Adog64/Adversarial-Machine-Learning-Clinic
1
0
Fork 0
Code Issues 3 Pull Requests Packages Projects 1 Releases Wiki Activity

Final report revisions

Browse Source
This commit is contained in:
Adog64 2024-05-11 21:39:29 -04:00
parent fd8fa12e75
commit 8b26371dab
23 changed files with 924 additions and 239 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
poster/pseudocode.py → examples/pseudocode.py
View File

@ -1,7 +1,7 @@
model = Net()
accuracies = {}
for filter in filters:
for filter_name in filters:
for epsilon in epsilons:
for strength in range(5):
correct = 0
@ -9,13 +9,13 @@ for filter in filters:
for data, target in dataset:
atk_data = fgsm_attack(data, epsilon)
filt_data = filter(atk_data, filter, strength)
filt_data = filtered(atk_data, filter_name, strength)
prediction = model(filt_data)
total += 1
if prediction == target:
correct += 1
accuracies[filter][epsilon][strength] = correct/total
accuracies[filter_name][epsilon][strength] = correct/total
save_json("results.json", accuracies)

BIN
impact_statements/Coverpage_for_Enhanching_Image_Classifiers_with_Denoising_Filters_Polikar.pdf
View File

Binary file not shown.

BIN
poster/CIFAR-10_excerpt.png
View File

Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 221 KiB

After

Width:  |  Height:  |  Size: 221 KiB

BIN
poster/MNIST_excerpt.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 14 KiB

88
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux
View File

@ -1,18 +1,26 @@
\relax
\abx@aux@refcontext{nty/global//global/global}
\abx@aux@cite{0}{ieee3129}
\abx@aux@segm{0}{0}{ieee3129}
\abx@aux@cite{0}{carlini2017evaluating}
\abx@aux@segm{0}{0}{carlini2017evaluating}
\abx@aux@cite{0}{szegedy2014intriguing}
\abx@aux@segm{0}{0}{szegedy2014intriguing}
\abx@aux@cite{0}{goodfellow2015explaining}
\abx@aux@segm{0}{0}{goodfellow2015explaining}
\abx@aux@cite{0}{goodfellow2015explaining}
\abx@aux@segm{0}{0}{goodfellow2015explaining}
\@writefile{toc}{\contentsline {section}{\numberline {1}Introduction}{2}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {2}Approach and Experimental Timeline}{2}{}\protected@file@percent }
\@writefile{lof}{\contentsline {figure}{\numberline {1}{\ignorespaces Simple block diagram of the proposed pipeline\relax }}{2}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {2}Approach and Background}{2}{}\protected@file@percent }
\newlabel{sec:approach_background}{{2}{2}}
\newlabel{eqn:fgsm}{{2}{2}}
\abx@aux@cite{0}{goodfellow2016deep}
\abx@aux@segm{0}{0}{goodfellow2016deep}
\abx@aux@cite{0}{carlini2017evaluating}
\abx@aux@segm{0}{0}{carlini2017evaluating}
\@writefile{lof}{\contentsline {figure}{\numberline {1}{\ignorespaces Excerpts each dataset used with elements arranged by class\relax }}{3}{}\protected@file@percent }
\providecommand*\caption@xref[2]{\@setref\relax\@undefined{#1}}
\newlabel{fig:concept-overview}{{1}{2}}
\newlabel{fig:dataset_excerpts}{{1}{3}}
\@writefile{toc}{\contentsline {section}{\numberline {3}Experimental Timeline}{3}{}\protected@file@percent }
\newlabel{sec:exp_timeline}{{3}{3}}
\abx@aux@cite{0}{ecma404}
\abx@aux@segm{0}{0}{ecma404}
\abx@aux@cite{0}{yu2019deep}
@ -21,33 +29,63 @@
\abx@aux@segm{0}{0}{simonyan2014very}
\abx@aux@cite{0}{carlini2017evaluating}
\abx@aux@segm{0}{0}{carlini2017evaluating}
\abx@aux@cite{0}{ieee3129}
\abx@aux@segm{0}{0}{ieee3129}
\@writefile{lof}{\contentsline {figure}{\numberline {2}{\ignorespaces Simple block diagram of the proposed pipeline\relax }}{4}{}\protected@file@percent }
\newlabel{fig:concept-overview}{{2}{4}}
\@writefile{toc}{\contentsline {section}{\numberline {4}Constraints}{4}{}\protected@file@percent }
\abx@aux@cite{0}{ieee3129}
\abx@aux@segm{0}{0}{ieee3129}
\abx@aux@cite{0}{ecma404}
\abx@aux@segm{0}{0}{ecma404}
\abx@aux@cite{0}{kuwahara1976processing}
\abx@aux@segm{0}{0}{kuwahara1976processing}
\@writefile{toc}{\contentsline {section}{\numberline {3}Constraints}{3}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {4}Alternative Filters}{3}{}\protected@file@percent }
\@writefile{lof}{\contentsline {figure}{\numberline {2}{\ignorespaces The seven selected filters operating on a randomly selected sample from MNIST attacked with FGSM with $\epsilon =0.2$\relax }}{3}{}\protected@file@percent }
\newlabel{fig:alternative_filters}{{2}{3}}
\newlabel{eqn:rad_odd_strength}{{1}{4}}
\@writefile{toc}{\contentsline {section}{\numberline {5}Results and Evaluation}{4}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {5}Applicable Engineering Standards}{5}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {5.1}IEEE 3129-2023}{5}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {5.2}ECMA-404}{5}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {6}Alternative Filters}{5}{}\protected@file@percent }
\@writefile{lof}{\contentsline {figure}{\numberline {3}{\ignorespaces The seven selected filters operating on a randomly selected sample from MNIST attacked with FGSM with $\epsilon =0.2$\relax }}{5}{}\protected@file@percent }
\newlabel{fig:alternative_filters}{{3}{5}}
\newlabel{eqn:rad_odd_strength}{{6}{5}}
\@writefile{toc}{\contentsline {section}{\numberline {7}Results and Evaluation}{6}{}\protected@file@percent }
\@writefile{lol}{\contentsline {lstlisting}{\numberline {1}{\ignorespaces Testing procedure overview}}{6}{}\protected@file@percent }
\abx@aux@cite{0}{schapire1989strength}
\abx@aux@segm{0}{0}{schapire1989strength}
\@writefile{lof}{\contentsline {figure}{\numberline {3}{\ignorespaces The highest rank filter and its accuracy for each attack strength and filter strength\relax }}{5}{}\protected@file@percent }
\newlabel{fig:filter_performances}{{3}{5}}
\@writefile{toc}{\contentsline {section}{\numberline {6}Conclusions}{5}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {7}Ethical Considerations}{5}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {7.1}Heath and Safety Considerations}{5}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {7.2}Social and Cultural Considerations}{6}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {7.3}Environmental Considerations}{6}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {7.4}Economic and Financial Considerations}{6}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {7.5}Generative AI Disclaimer}{6}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {8}Key Takeaways}{6}{}\protected@file@percent }
\@writefile{lof}{\contentsline {figure}{\numberline {4}{\ignorespaces The highest rank filter and its accuracy for each attack strength and filter strength\relax }}{7}{}\protected@file@percent }
\newlabel{fig:filter_performances}{{4}{7}}
\@writefile{toc}{\contentsline {section}{\numberline {8}Conclusions}{8}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {9}Ethical Considerations and Impact Factors}{8}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {9.1}Health and Safety Ethical Considerations}{8}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {9.2}Heath and Safety Impact Factors}{8}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {9.3}Social and Cultural Ethical Considerations}{8}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {9.4}Social and Cultural Impact Factors}{8}{}\protected@file@percent }
\abx@aux@cite{0}{carlini2017evaluating}
\abx@aux@segm{0}{0}{carlini2017evaluating}
\@writefile{toc}{\contentsline {section}{\numberline {9}Future Work}{7}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {10}Acknowledgements}{7}{}\protected@file@percent }
\abx@aux@read@bbl@mdfivesum{9111094E0DE770FBB5CC5AAD69E9088A}
\abx@aux@cite{0}{ieee3129}
\abx@aux@segm{0}{0}{ieee3129}
\@writefile{toc}{\contentsline {subsection}{\numberline {9.5}Environmental Ethical Considerations}{9}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {9.6}Environmental Impact Factors}{9}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {9.7}Economic and Financial Ethical Considerations}{9}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {9.8}Economic and Financial Impact Factors}{9}{}\protected@file@percent }
\@writefile{toc}{\contentsline {subsection}{\numberline {9.9}Generative AI Disclaimer}{9}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {10}Key Takeaways}{9}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {11}Future Work}{10}{}\protected@file@percent }
\newlabel{sec:future_work}{{11}{10}}
\@writefile{toc}{\contentsline {section}{\numberline {12}Acknowledgements}{10}{}\protected@file@percent }
\@writefile{toc}{\contentsline {section}{\numberline {A}Additional Views of MNIST Results}{12}{}\protected@file@percent }
\@writefile{lof}{\contentsline {figure}{\numberline {5}{\ignorespaces Performances for each filter with filter strength 1 on MNIST attacked with FGSM\relax }}{12}{}\protected@file@percent }
\newlabel{fig:mnist_fgsm_strength1}{{5}{12}}
\@writefile{lof}{\contentsline {figure}{\numberline {6}{\ignorespaces Performances for each filter with filter strength 4 on MNIST attacked with FGSM\relax }}{12}{}\protected@file@percent }
\newlabel{fig:mnist_fgsm_strength4}{{6}{12}}
\@writefile{toc}{\contentsline {section}{\numberline {B}Additional Views of CIFAR-10 Results}{13}{}\protected@file@percent }
\@writefile{lof}{\contentsline {figure}{\numberline {7}{\ignorespaces Performances for each filter with filter strength 1 on CIFAR-10 attacked with FGSM\relax }}{13}{}\protected@file@percent }
\newlabel{fig:mnist_fgsm_strength1}{{7}{13}}
\@writefile{lof}{\contentsline {figure}{\numberline {8}{\ignorespaces Performances for each filter with filter strength 4 on CIFAR-10 attacked with FGSM\relax }}{13}{}\protected@file@percent }
\newlabel{fig:mnist_fgsm_strength4}{{8}{13}}
\abx@aux@read@bbl@mdfivesum{9BF6AC5F7471C86B73B3CF4964B65FD9}
\abx@aux@defaultrefcontext{0}{carlini2017evaluating}{nty/global//global/global}
\abx@aux@defaultrefcontext{0}{ecma404}{nty/global//global/global}
\abx@aux@defaultrefcontext{0}{goodfellow2016deep}{nty/global//global/global}
\abx@aux@defaultrefcontext{0}{goodfellow2015explaining}{nty/global//global/global}
\abx@aux@defaultrefcontext{0}{ieee3129}{nty/global//global/global}
\abx@aux@defaultrefcontext{0}{kuwahara1976processing}{nty/global//global/global}
@ -55,4 +93,4 @@
\abx@aux@defaultrefcontext{0}{simonyan2014very}{nty/global//global/global}
\abx@aux@defaultrefcontext{0}{szegedy2014intriguing}{nty/global//global/global}
\abx@aux@defaultrefcontext{0}{yu2019deep}{nty/global//global/global}
\gdef \@abspage@last{9}
\gdef \@abspage@last{14}

35
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl
View File

@ -78,6 +78,41 @@
\verb 10.13140/RG.2.2.28181.14560
\endverb
\endentry
\entry{goodfellow2016deep}{book}{}
\name{author}{3}{}{%
{{hash=5d2585c11210cf1d4512e6e0a03ec315}{%
family={Goodfellow},
familyi={G\bibinitperiod},
given={Ian},
giveni={I\bibinitperiod}}}%
{{hash=40a8e4774982146adc2688546f54efb2}{%
family={Bengio},
familyi={B\bibinitperiod},
given={Yoshua},
giveni={Y\bibinitperiod}}}%
{{hash=ccec1ccd2e1aa86960eb2e872c6b7020}{%
family={Courville},
familyi={C\bibinitperiod},
given={Aaron},
giveni={A\bibinitperiod}}}%
}
\list{publisher}{1}{%
{MIT Press}%
}
\strng{namehash}{3ae53fe582e8a815b118d26947eaa326}
\strng{fullhash}{3ae53fe582e8a815b118d26947eaa326}
\strng{bibnamehash}{3ae53fe582e8a815b118d26947eaa326}
\strng{authorbibnamehash}{3ae53fe582e8a815b118d26947eaa326}
\strng{authornamehash}{3ae53fe582e8a815b118d26947eaa326}
\strng{authorfullhash}{3ae53fe582e8a815b118d26947eaa326}
\field{sortinit}{G}
\field{sortinithash}{32d67eca0634bf53703493fb1090a2e8}
\field{labelnamesource}{author}
\field{labeltitlesource}{title}
\field{note}{\url{http://www.deeplearningbook.org}}
\field{title}{Deep Learning}
\field{year}{2016}
\endentry
\entry{goodfellow2015explaining}{misc}{}
\name{author}{3}{}{%
{{hash=d4f74ef4c79f3bb1e51e378184d8850e}{%

26
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bcf
View File

@ -2361,17 +2361,23 @@
<bcf:datasource type="file" datatype="bibtex" glob="false">references.bib</bcf:datasource>
</bcf:bibdata>
<bcf:section number="0">
<bcf:citekey order="1" intorder="1">ieee3129</bcf:citekey>
<bcf:citekey order="2" intorder="1">carlini2017evaluating</bcf:citekey>
<bcf:citekey order="3" intorder="1">szegedy2014intriguing</bcf:citekey>
<bcf:citekey order="1" intorder="1">carlini2017evaluating</bcf:citekey>
<bcf:citekey order="2" intorder="1">szegedy2014intriguing</bcf:citekey>
<bcf:citekey order="3" intorder="1">goodfellow2015explaining</bcf:citekey>
<bcf:citekey order="4" intorder="1">goodfellow2015explaining</bcf:citekey>
<bcf:citekey order="5" intorder="1">ecma404</bcf:citekey>
<bcf:citekey order="6" intorder="1">yu2019deep</bcf:citekey>
<bcf:citekey order="7" intorder="1">simonyan2014very</bcf:citekey>
<bcf:citekey order="8" intorder="1">carlini2017evaluating</bcf:citekey>
<bcf:citekey order="9" intorder="1">kuwahara1976processing</bcf:citekey>
<bcf:citekey order="10" intorder="1">schapire1989strength</bcf:citekey>
<bcf:citekey order="11" intorder="1">carlini2017evaluating</bcf:citekey>
<bcf:citekey order="5" intorder="1">goodfellow2016deep</bcf:citekey>
<bcf:citekey order="6" intorder="1">carlini2017evaluating</bcf:citekey>
<bcf:citekey order="7" intorder="1">ecma404</bcf:citekey>
<bcf:citekey order="8" intorder="1">yu2019deep</bcf:citekey>
<bcf:citekey order="9" intorder="1">simonyan2014very</bcf:citekey>
<bcf:citekey order="10" intorder="1">carlini2017evaluating</bcf:citekey>
<bcf:citekey order="11" intorder="1">ieee3129</bcf:citekey>
<bcf:citekey order="12" intorder="1">ieee3129</bcf:citekey>
<bcf:citekey order="13" intorder="1">ecma404</bcf:citekey>
<bcf:citekey order="14" intorder="1">kuwahara1976processing</bcf:citekey>
<bcf:citekey order="15" intorder="1">schapire1989strength</bcf:citekey>
<bcf:citekey order="16" intorder="1">carlini2017evaluating</bcf:citekey>
<bcf:citekey order="17" intorder="1">ieee3129</bcf:citekey>
</bcf:section>
<!-- SORTING TEMPLATES -->
<bcf:sortingtemplate name="nty">

26
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.blg
View File

@ -1,15 +1,15 @@
[0] Config.pm:307> INFO - This is Biber 2.19
[0] Config.pm:310> INFO - Logfile is 'Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.blg'
[82] biber:340> INFO - === Wed May 8, 2024, 20:57:47
[96] Biber.pm:419> INFO - Reading 'Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bcf'
[158] Biber.pm:979> INFO - Found 9 citekeys in bib section 0
[173] Biber.pm:4419> INFO - Processing section 0
[183] Biber.pm:4610> INFO - Looking for bibtex file 'references.bib' for section 0
[184] bibtex.pm:1713> INFO - LaTeX decoding ...
[192] bibtex.pm:1519> INFO - Found BibTeX data source 'references.bib'
[291] UCollate.pm:68> INFO - Overriding locale 'en-US' defaults 'normalization = NFD' with 'normalization = prenormalized'
[291] UCollate.pm:68> INFO - Overriding locale 'en-US' defaults 'variable = shifted' with 'variable = non-ignorable'
[291] Biber.pm:4239> INFO - Sorting list 'nty/global//global/global' of type 'entry' with template 'nty' and locale 'en-US'
[291] Biber.pm:4245> INFO - No sort tailoring available for locale 'en-US'
[299] bbl.pm:660> INFO - Writing 'Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl' with encoding 'UTF-8'
[302] bbl.pm:763> INFO - Output to Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl
[77] biber:340> INFO - === Sat May 11, 2024, 14:28:39
[92] Biber.pm:419> INFO - Reading 'Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bcf'
[154] Biber.pm:979> INFO - Found 10 citekeys in bib section 0
[170] Biber.pm:4419> INFO - Processing section 0
[180] Biber.pm:4610> INFO - Looking for bibtex file 'references.bib' for section 0
[190] bibtex.pm:1713> INFO - LaTeX decoding ...
[199] bibtex.pm:1519> INFO - Found BibTeX data source 'references.bib'
[297] UCollate.pm:68> INFO - Overriding locale 'en-US' defaults 'variable = shifted' with 'variable = non-ignorable'
[297] UCollate.pm:68> INFO - Overriding locale 'en-US' defaults 'normalization = NFD' with 'normalization = prenormalized'
[297] Biber.pm:4239> INFO - Sorting list 'nty/global//global/global' of type 'entry' with template 'nty' and locale 'en-US'
[297] Biber.pm:4245> INFO - No sort tailoring available for locale 'en-US'
[306] bbl.pm:660> INFO - Writing 'Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl' with encoding 'UTF-8'
[309] bbl.pm:763> INFO - Output to Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl

59
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.fdb_latexmk
View File

@ -1,31 +1,55 @@
# Fdb version 4
["biber Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar"] 1715216266.79693 "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bcf" "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl" "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar" 1715216269.40277 0
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bcf" 1715216269.31768 108499 95a47fd2fc684088b90f67f61dca5899 "pdflatex"
"references.bib" 1715049098.74874 2739 5a48774d00665200923a145db75385de ""
["biber Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar"] 1715452119.28702 "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bcf" "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl" "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar" 1715469651.47844 0
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bcf" 1715469651.37563 108921 6820876556521df27b59f0d479a70cf7 "pdflatex"
"references.bib" 1715404651.38054 2956 732703545485ef32f8dba6d55b81c8ac ""
(generated)
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl"
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.blg"
(rewritten before read)
["pdflatex"] 1715216267.698 "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.tex" "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.pdf" "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar" 1715216269.4031 0
["pdflatex"] 1715469649.0722 "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.tex" "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.pdf" "Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar" 1715469651.47884 0
"../examples/bilateral_filter_eps_0.2.png" 1714589885.31582 19297 a377e84c1b2287303d66a65095c4f66f ""
"../examples/bit_depth_eps_0.2.png" 1714589885.31582 17392 8fbda7f34c4a1b766481d7448d6c6a6d ""
"../examples/gaussian_blur_eps_0.2.png" 1714589885.31682 20939 b8f00f2cf22ad296bc257cef04fbd329 ""
"../examples/gaussian_kuwahara_eps_0.2.png" 1714589885.31682 24613 40ffc963acce3b07a9ae332bbda1b96a ""
"../examples/mean_kuwahara_eps_0.2.png" 1714589885.31682 22802 40baccf677bc3568b40aa6cc2e8129fa ""
"../examples/pseudocode.py" 1715298588.97525 593 1df67f3e89331a32b1b0d665c8def003 ""
"../examples/random_noise_eps_0.2.png" 1714589885.31682 24314 c5738d11c496c013f543d0ceca6f4c76 ""
"../examples/threshold_eps_0.2.png" 1714589885.31682 17748 16fe06b182d4b12cd0ff0db1f22818cc ""
"../examples/unfiltered_eps_0.2.png" 1714589885.31782 17055 9bd4320ffca5bddf882ef20a12722551 ""
"../poster/CIFAR-10_excerpt.png" 1715390814.37055 225943 68ee413abe2c512ff9a7dbbf32192933 ""
"../poster/Concept.png" 1714589885.34182 49562 1f25e2373f82dec47edf226266717baf ""
"../poster/MNIST_excerpt.png" 1715390802.80857 13965 76c7545efc248683798ae505d9525d07 ""
"../results/CIFAR-10_FGSM_3d.png" 1714589885.68882 171512 62693c4d9adfd475017080143dadb6c4 ""
"../results/CIFAR-10_FGSM_line_strength1.png" 1715405918.24979 108521 810050b2be240ec1650e04c5ef2d3bc6 ""
"../results/CIFAR-10_FGSM_line_strength4.png" 1715405938.41475 99784 2e1f739f81b6ae35fb26cf0b2935be57 ""
"../results/MNIST_FGSM_3d_new.png" 1714589885.70382 174972 e032c9953482469a5639e79b7085256e ""
"../results/MNIST_FGSM_line_strength1.png" 1715376687.87043 133036 9303bd76a3caab8c478882bf5619e2d4 ""
"../results/MNIST_FGSM_line_strength4.png" 1715374256.82344 136901 da7f401a87183eddbc433cc6081bad6e ""
"/usr/share/texlive/texmf-dist/fonts/map/fontname/texfonts.map" 1577235249 3524 cb3e574dea2d1052e39280babc910dc8 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmbsy5.tfm" 1246382020 1120 1e8878807317373affa7f7bba4cf2f6a ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmbsy6.tfm" 1246382020 1124 14ccf5552bc7f77ca02a8a402bea8bfb ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmbsy7.tfm" 1246382020 1120 7f9f170e8aa57527ad6c49feafd45d54 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmbsy8.tfm" 1246382020 1120 200be8b775682cdf80acad4be5ef57e4 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmex7.tfm" 1246382020 1004 54797486969f23fa377b128694d548df ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmex8.tfm" 1246382020 988 bdf658c3bfc2d96d3c8b02cfc1c94c20 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmmib5.tfm" 1246382020 1496 c79f6914c6d39ffb3759967363d1be79 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmmib6.tfm" 1246382020 1516 a3bf6a5e7ec4401b1f52092dfaaed242 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmmib7.tfm" 1246382020 1508 6e807ff901c35a5f1fde0ca275533df8 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmmib8.tfm" 1246382020 1528 dab402b9d3774ca98baa037071cee7ae ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbsy10.tfm" 1136768653 1116 4e6ba9d7914baa6482fd69f67d126380 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx10.tfm" 1136768653 1328 c834bbb027764024c09d3d2bf908b5f0 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx12.tfm" 1136768653 1324 c910af8c371558dc20f2d7822f66fe64 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx5.tfm" 1136768653 1332 f817c21a1ba54560425663374f1b651a ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx6.tfm" 1136768653 1344 8a0be4fe4d376203000810ad4dc81558 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx7.tfm" 1136768653 1336 3125ccb448c1a09074e3aa4a9832f130 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx8.tfm" 1136768653 1332 1fde11373e221473104d6cc5993f046e ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx9.tfm" 1136768653 1328 5442e22a7072966dbaf88ca900acf3f0 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmcsc10.tfm" 1136768653 1300 63a6111ee6274895728663cf4b4e7e81 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmex10.tfm" 1136768653 992 662f679a0b3d2d53c1b94050fdaa3f50 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmmi12.tfm" 1136768653 1524 4414a8315f39513458b80dfc63bff03a ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmmi6.tfm" 1136768653 1512 f21f83efb36853c0b70002322c1ab3ad ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmmi8.tfm" 1136768653 1520 eccf95517727cb11801f4f1aee3a21b4 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmmib10.tfm" 1136768653 1524 554068197b70979a55370e6c6495f441 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmr12.tfm" 1136768653 1288 655e228510b4c2a1abe905c368440826 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmr17.tfm" 1136768653 1292 296a67155bdbfc32aa9c636f21e91433 ""
"/usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmr6.tfm" 1136768653 1300 b62933e007d01cfd073f79b963c01526 ""
@ -42,6 +66,7 @@
"/usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmcsc10.pfb" 1248133631 32001 6aeea3afe875097b1eb0da29acd61e28 ""
"/usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmi10.pfb" 1248133631 36299 5f9df58c2139e7edcf37c8fca4bd384d ""
"/usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmi7.pfb" 1248133631 36281 c355509802a035cadc5f15869451dcee ""
"/usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmib10.pfb" 1248133631 36912 b448ef9ad9d7228ec3c6e71005136d55 ""
"/usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr10.pfb" 1248133631 35752 024fb6c41858982481f6968b5fc26508 ""
"/usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr12.pfb" 1248133631 32722 d7379af29a190c3f453aba36302ff5a9 ""
"/usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr17.pfb" 1248133631 32362 179c33bbf43f19adbb3825bb4e36e57a ""
@ -55,10 +80,18 @@
"/usr/share/texlive/texmf-dist/tex/generic/iftex/iftex.sty" 1644112042 7237 bdd120a32c8fdb4b433cf9ca2e7cd98a ""
"/usr/share/texlive/texmf-dist/tex/generic/iftex/ifvtex.sty" 1572645307 1057 525c2192b5febbd8c1f662c9468335bb ""
"/usr/share/texlive/texmf-dist/tex/generic/infwarerr/infwarerr.sty" 1575499628 8356 7bbb2c2373aa810be568c29e333da8ed ""
"/usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty" 1576878844 5412 d5a2436094cd7be85769db90f29250a6 ""
"/usr/share/texlive/texmf-dist/tex/generic/ltxcmds/ltxcmds.sty" 1600895880 17859 4409f8f50cd365c68e684407e5350b1b ""
"/usr/share/texlive/texmf-dist/tex/generic/pdftexcmds/pdftexcmds.sty" 1593379760 20089 80423eac55aa175305d35b49e04fe23b ""
"/usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty" 1654720880 2222 78b930a5a6e3dc2ac69b78c2057b94d7 ""
"/usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty" 1654720880 4173 c989ee3ced31418e3593916ab26c793a ""
"/usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty" 1654720880 88393 1adf6fa3f245270d06e3d4f8910f7fc5 ""
"/usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty" 1654720880 4474 f04cd1cc7bd76eb033e6fb12eb6a0d77 ""
"/usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty" 1654720880 2444 70065bddd85997dc1fd0bb7ae634e5fa ""
"/usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty" 1581200180 8878 d9f65b39ca82f1d70030390eca653b1c ""
"/usr/share/texlive/texmf-dist/tex/latex/base/article.cls" 1689984000 20144 d5ecf0a5140c8d8d8b72cbe86e320eff ""
"/usr/share/texlive/texmf-dist/tex/latex/base/ifthen.sty" 1689984000 5319 48d7f3cfa322abd2788e3c09d624b922 ""
"/usr/share/texlive/texmf-dist/tex/latex/base/omscmr.fd" 1689984000 2469 24e8437abfd2393aed5f0ac9f00f801e ""
"/usr/share/texlive/texmf-dist/tex/latex/base/size10.clo" 1689984000 8448 c33a4e1cb35cee9b33c2b21033b73e39 ""
"/usr/share/texlive/texmf-dist/tex/latex/biblatex/bbx/numeric.bbx" 1609451401 1818 9ed166ac0a9204a8ebe450ca09db5dde ""
"/usr/share/texlive/texmf-dist/tex/latex/biblatex/bbx/standard.bbx" 1609451401 25680 409c3f3d570418bc545e8065bebd0688 ""
@ -83,23 +116,29 @@
"/usr/share/texlive/texmf-dist/tex/latex/graphics/graphicx.sty" 1654720880 8010 a8d949cbdbc5c983593827c9eec252e1 ""
"/usr/share/texlive/texmf-dist/tex/latex/graphics/keyval.sty" 1654720880 2671 7e67d78d9b88c845599a85b2d41f2e39 ""
"/usr/share/texlive/texmf-dist/tex/latex/graphics/trig.sty" 1654720880 4023 293ea1c16429fc0c4cf605f4da1791a9 ""
"/usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty" 1575499774 7133 b94bbacbee6e4fdccdc7f810b2aec370 ""
"/usr/share/texlive/texmf-dist/tex/latex/kvoptions/kvoptions.sty" 1655478651 22555 6d8e155cfef6d82c3d5c742fea7c992e ""
"/usr/share/texlive/texmf-dist/tex/latex/kvsetkeys/kvsetkeys.sty" 1665067230 13815 760b0c02f691ea230f5359c4e1de23a7 ""
"/usr/share/texlive/texmf-dist/tex/latex/l3backend/l3backend-pdftex.def" 1673989714 30429 213676d4c7327a21d91ddaed900e7b81 ""
"/usr/share/texlive/texmf-dist/tex/latex/l3kernel/expl3.sty" 1677186603 6107 5cfea8a675c58918b8c04be10261e48c ""
"/usr/share/texlive/texmf-dist/tex/latex/l3packages/xparse/xparse.sty" 1675461949 6812 3c152a1c8d562d7b7291c4839b61a5c3 ""
"/usr/share/texlive/texmf-dist/tex/latex/latexconfig/epstopdf-sys.cfg" 1279039959 678 4792914a8f45be57bb98413425e4c7af ""
"/usr/share/texlive/texmf-dist/tex/latex/listings/listings.cfg" 1677530001 1829 d8258b7d94f5f955e70c623e525f9f45 ""
"/usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty" 1677530001 80947 75a96bb4c9f40ae31d54a01d924df2ff ""
"/usr/share/texlive/texmf-dist/tex/latex/listings/lstlang1.sty" 1677530001 205154 31132370016e8c97e49bc3862419679b ""
"/usr/share/texlive/texmf-dist/tex/latex/listings/lstmisc.sty" 1677530001 77021 d05e9115c67855816136d82929db8892 ""
"/usr/share/texlive/texmf-dist/tex/latex/logreq/logreq.def" 1284153563 1620 fb1c32b818f2058eca187e5c41dfae77 ""
"/usr/share/texlive/texmf-dist/tex/latex/logreq/logreq.sty" 1284153563 6187 b27afc771af565d3a9ff1ca7d16d0d46 ""
"/usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty" 1654720880 13231 b52297489a0e9d929aae403417d92a02 ""
"/usr/share/texlive/texmf-dist/tex/latex/url/url.sty" 1388531844 12796 8edb7d69a20b857904dd0ea757c14ec9 ""
"/usr/share/texlive/texmf-dist/web2c/texmf.cnf" 1689984000 40900 887e0dc8cac988a9e9c574af364cf837 ""
"/var/lib/texmf/fonts/map/pdftex/updmap/pdftex.map" 1705885142.27725 4602615 ac1019cb4c4ba39710c2cf7f752b378d ""
"/var/lib/texmf/web2c/pdftex/pdflatex.fmt" 1710776733 7871985 a923b025655ef6c2f0aa4b5ff9642a36 ""
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux" 1715216269.31568 4132 6743685263894671c0aa2ccde3f16fa2 "pdflatex"
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl" 1715216267.67469 12728 9111094e0de770fbb5cc5aad69e9088a "biber Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar"
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.run.xml" 1715216269.31768 2809 f4f54a2692c91e06b589c996b2bb321b "pdflatex"
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.tex" 1715215108.46121 21144 12f80f3fdb285a7029a4f6a86e4e569d ""
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.toc" 1715216269.31768 1056 ffb6043435a176e7044172b0eaef5e7d "pdflatex"
"/var/lib/texmf/web2c/pdftex/pdflatex.fmt" 1715441823 7871986 49831ca64e8868b85935ca2a969d18f1 ""
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux" 1715469651.37363 7290 57452a6170a517a7a574c82201992b31 "pdflatex"
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl" 1715452120.22387 14072 9bf6ac5f7471c86b73b3cf4964b65fd9 "biber Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar"
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.run.xml" 1715469651.37663 2809 f4f54a2692c91e06b589c996b2bb321b "pdflatex"
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.tex" 1715469624.76728 30234 682006c047f8c209ed8e8cbd309d21a1 ""
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.toc" 1715469651.37663 1877 eed594d04747640f866b7bbf9a508b00 "pdflatex"
(generated)
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux"
"Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bcf"

297
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.fls
View File

@ -337,6 +337,107 @@ INPUT /usr/share/texlive/texmf-dist/tex/latex/caption/subcaption.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/caption/subcaption.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/caption/subcaption.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/caption/caption.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/graphics/keyval.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/lstmisc.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/lstmisc.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/lstmisc.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/lstmisc.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.cfg
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.cfg
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.cfg
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/listings.cfg
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
INPUT ./Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux
INPUT Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux
INPUT Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux
@ -356,6 +457,32 @@ INPUT /usr/share/texlive/texmf-dist/tex/latex/epstopdf-pkg/epstopdf-base.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/epstopdf-pkg/epstopdf-base.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/epstopdf-pkg/epstopdf-base.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/epstopdf-pkg/epstopdf-base.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/infwarerr/infwarerr.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/infwarerr/infwarerr.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/kvoptions/kvoptions.sty
INPUT /usr/share/texlive/texmf-dist/tex/generic/pdftexcmds/pdftexcmds.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/latexconfig/epstopdf-sys.cfg
INPUT /usr/share/texlive/texmf-dist/tex/latex/latexconfig/epstopdf-sys.cfg
INPUT /usr/share/texlive/texmf-dist/tex/latex/latexconfig/epstopdf-sys.cfg
@ -381,6 +508,18 @@ INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmmi6.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmsy10.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmsy8.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmsy6.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmex10.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmex8.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmex7.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx12.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx8.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx6.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmmib10.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmmib8.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmmib6.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbsy10.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmbsy8.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmbsy6.tfm
OUTPUT Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.pdf
INPUT /var/lib/texmf/fonts/map/pdftex/updmap/pdftex.map
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmr12.tfm
@ -389,68 +528,121 @@ INPUT ./Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Pol
INPUT Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.toc
INPUT Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.toc
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx10.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmex7.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmex7.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx7.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx5.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmmib10.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmmib7.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmmib5.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbsy10.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmbsy7.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/amsfonts/cmextra/cmbsy5.tfm
OUTPUT Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.toc
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmr9.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx9.tfm
INPUT ../poster/MNIST_excerpt.png
INPUT ../poster/MNIST_excerpt.png
INPUT ../poster/MNIST_excerpt.png
INPUT ../poster/MNIST_excerpt.png
INPUT ../poster/MNIST_excerpt.png
INPUT ../poster/CIFAR-10_excerpt.png
INPUT ../poster/CIFAR-10_excerpt.png
INPUT ../poster/CIFAR-10_excerpt.png
INPUT ../poster/CIFAR-10_excerpt.png
INPUT ../poster/CIFAR-10_excerpt.png
INPUT ../poster/Concept.png
INPUT ../poster/Concept.png
INPUT ../poster/Concept.png
INPUT ../poster/Concept.png
INPUT ../poster/Concept.png
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmti10.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmbx12.tfm
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmcsc10.tfm
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/unfiltered_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/bit_depth_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_blur_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/gaussian_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/mean_kuwahara_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/random_noise_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/bilateral_filter_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT ../examples/threshold_eps_0.2.png
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/lstlang1.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/lstlang1.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/lstlang1.sty
INPUT /usr/share/texlive/texmf-dist/tex/latex/listings/lstlang1.sty
INPUT ../examples/pseudocode.py
INPUT ../examples/pseudocode.py
INPUT ../examples/pseudocode.py
INPUT ../examples/pseudocode.py
INPUT ../examples/pseudocode.py
INPUT ../examples/pseudocode.py
INPUT /usr/share/texlive/texmf-dist/tex/latex/base/omscmr.fd
INPUT /usr/share/texlive/texmf-dist/tex/latex/base/omscmr.fd
INPUT /usr/share/texlive/texmf-dist/tex/latex/base/omscmr.fd
INPUT /usr/share/texlive/texmf-dist/tex/latex/base/omscmr.fd
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmtt10.tfm
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/MNIST_FGSM_3d_new.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT ../results/CIFAR-10_FGSM_3d.png
INPUT /usr/share/texlive/texmf-dist/fonts/tfm/public/cm/cmcsc10.tfm
INPUT ../results/MNIST_FGSM_line_strength1.png
INPUT ../results/MNIST_FGSM_line_strength1.png
INPUT ../results/MNIST_FGSM_line_strength1.png
INPUT ../results/MNIST_FGSM_line_strength1.png
INPUT ../results/MNIST_FGSM_line_strength1.png
INPUT ../results/MNIST_FGSM_line_strength4.png
INPUT ../results/MNIST_FGSM_line_strength4.png
INPUT ../results/MNIST_FGSM_line_strength4.png
INPUT ../results/MNIST_FGSM_line_strength4.png
INPUT ../results/MNIST_FGSM_line_strength4.png
INPUT ../results/CIFAR-10_FGSM_line_strength1.png
INPUT ../results/CIFAR-10_FGSM_line_strength1.png
INPUT ../results/CIFAR-10_FGSM_line_strength1.png
INPUT ../results/CIFAR-10_FGSM_line_strength1.png
INPUT ../results/CIFAR-10_FGSM_line_strength1.png
INPUT ../results/CIFAR-10_FGSM_line_strength4.png
INPUT ../results/CIFAR-10_FGSM_line_strength4.png
INPUT ../results/CIFAR-10_FGSM_line_strength4.png
INPUT ../results/CIFAR-10_FGSM_line_strength4.png
INPUT ../results/CIFAR-10_FGSM_line_strength4.png
INPUT Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux
INPUT Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.run.xml
OUTPUT Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.run.xml
@ -460,6 +652,7 @@ INPUT /usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmbx9.pfb
INPUT /usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmcsc10.pfb
INPUT /usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmi10.pfb
INPUT /usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmi7.pfb
INPUT /usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmib10.pfb
INPUT /usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr10.pfb
INPUT /usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr12.pfb
INPUT /usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr17.pfb

320
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.log
View File

@ -1,4 +1,4 @@
This is pdfTeX, Version 3.141592653-2.6-1.40.25 (TeX Live 2023) (preloaded format=pdflatex 2024.3.18) 8 MAY 2024 20:57
This is pdfTeX, Version 3.141592653-2.6-1.40.25 (TeX Live 2023) (preloaded format=pdflatex 2024.5.11) 11 MAY 2024 19:20
entering extended mode
restricted \write18 enabled.
file:line:error style messages enabled.
@ -338,42 +338,156 @@ Package caption Info: New subtype `subfigure' on input line 239.
\c@subfigure=\count393
Package caption Info: New subtype `subtable' on input line 239.
\c@subtable=\count394
) (/usr/share/texlive/texmf-dist/tex/latex/tools/bm.sty
Package: bm 2022/01/05 v1.2f Bold Symbol Support (DPC/FMi)
\symboldoperators=\mathgroup4
\symboldletters=\mathgroup5
\symboldsymbols=\mathgroup6
Package bm Info: No bold for \OMX/cmex/m/n, using \pmb.
LaTeX Font Info: Redeclaring math alphabet \mathbf on input line 149.
) (/usr/share/texlive/texmf-dist/tex/latex/amsmath/amsmath.sty
Package: amsmath 2022/04/08 v2.17n AMS math features
\@mathmargin=\skip72
For additional information on amsmath, use the `?' option.
(/usr/share/texlive/texmf-dist/tex/latex/amsmath/amstext.sty
Package: amstext 2021/08/26 v2.01 AMS text
(/usr/share/texlive/texmf-dist/tex/latex/amsmath/amsgen.sty
File: amsgen.sty 1999/11/30 v2.0 generic functions
\@emptytoks=\toks20
\ex@=\dimen159
)) (/usr/share/texlive/texmf-dist/tex/latex/amsmath/amsbsy.sty
Package: amsbsy 1999/11/29 v1.2d Bold Symbols
LaTeX Info: Redefining \boldsymbol on input line 28.
\pmbraise@=\dimen160
) (/usr/share/texlive/texmf-dist/tex/latex/amsmath/amsopn.sty
Package: amsopn 2022/04/08 v2.04 operator names
)
\inf@bad=\count395
LaTeX Info: Redefining \frac on input line 234.
\uproot@=\count396
\leftroot@=\count397
LaTeX Info: Redefining \overline on input line 399.
LaTeX Info: Redefining \colon on input line 410.
\classnum@=\count398
\DOTSCASE@=\count399
LaTeX Info: Redefining \ldots on input line 496.
LaTeX Info: Redefining \dots on input line 499.
LaTeX Info: Redefining \cdots on input line 620.
\Mathstrutbox@=\box54
\strutbox@=\box55
LaTeX Info: Redefining \big on input line 722.
LaTeX Info: Redefining \Big on input line 723.
LaTeX Info: Redefining \bigg on input line 724.
LaTeX Info: Redefining \Bigg on input line 725.
\big@size=\dimen161
LaTeX Font Info: Redeclaring font encoding OML on input line 743.
LaTeX Font Info: Redeclaring font encoding OMS on input line 744.
\macc@depth=\count400
LaTeX Info: Redefining \bmod on input line 905.
LaTeX Info: Redefining \pmod on input line 910.
LaTeX Info: Redefining \smash on input line 940.
LaTeX Info: Redefining \relbar on input line 970.
LaTeX Info: Redefining \Relbar on input line 971.
\c@MaxMatrixCols=\count401
\dotsspace@=\muskip21
\c@parentequation=\count402
\dspbrk@lvl=\count403
\tag@help=\toks21
\row@=\count404
\column@=\count405
\maxfields@=\count406
\andhelp@=\toks22
\eqnshift@=\dimen162
\alignsep@=\dimen163
\tagshift@=\dimen164
\tagwidth@=\dimen165
\totwidth@=\dimen166
\lineht@=\dimen167
\@envbody=\toks23
\multlinegap=\skip73
\multlinetaggap=\skip74
\mathdisplay@stack=\toks24
LaTeX Info: Redefining \[ on input line 2953.
LaTeX Info: Redefining \] on input line 2954.
) (/usr/share/texlive/texmf-dist/tex/latex/listings/listings.sty
\lst@mode=\count407
\lst@gtempboxa=\box56
\lst@token=\toks25
\lst@length=\count408
\lst@currlwidth=\dimen168
\lst@column=\count409
\lst@pos=\count410
\lst@lostspace=\dimen169
\lst@width=\dimen170
\lst@newlines=\count411
\lst@lineno=\count412
\lst@maxwidth=\dimen171
(/usr/share/texlive/texmf-dist/tex/latex/listings/lstmisc.sty
File: lstmisc.sty 2023/02/27 1.9 (Carsten Heinz)
\c@lstnumber=\count413
\lst@skipnumbers=\count414
\lst@framebox=\box57
) (/usr/share/texlive/texmf-dist/tex/latex/listings/listings.cfg
File: listings.cfg 2023/02/27 1.9 listings configuration
))
Package: listings 2023/02/27 1.9 (Carsten Heinz)
(/usr/share/texlive/texmf-dist/tex/latex/appendix/appendix.sty
Package: appendix 2020/02/08 v1.2c extra appendix facilities
\c@@pps=\count415
\c@@ppsavesec=\count416
\c@@ppsaveapp=\count417
)
\@quotelevel=\count418
\@quotereset=\count419
(./Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux
LaTeX Warning: Label `fig:mnist_fgsm_strength1' multiply defined.
LaTeX Warning: Label `fig:mnist_fgsm_strength4' multiply defined.
)
\@quotelevel=\count395
\@quotereset=\count396
(./Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux)
\openout1 = `Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux'.
LaTeX Font Info: Checking defaults for OML/cmm/m/it on input line 15.
LaTeX Font Info: ... okay on input line 15.
LaTeX Font Info: Checking defaults for OMS/cmsy/m/n on input line 15.
LaTeX Font Info: ... okay on input line 15.
LaTeX Font Info: Checking defaults for OT1/cmr/m/n on input line 15.
LaTeX Font Info: ... okay on input line 15.
LaTeX Font Info: Checking defaults for T1/cmr/m/n on input line 15.
LaTeX Font Info: ... okay on input line 15.
LaTeX Font Info: Checking defaults for TS1/cmr/m/n on input line 15.
LaTeX Font Info: ... okay on input line 15.
LaTeX Font Info: Checking defaults for OMX/cmex/m/n on input line 15.
LaTeX Font Info: ... okay on input line 15.
LaTeX Font Info: Checking defaults for U/cmr/m/n on input line 15.
LaTeX Font Info: ... okay on input line 15.
LaTeX Font Info: Checking defaults for OML/cmm/m/it on input line 22.
LaTeX Font Info: ... okay on input line 22.
LaTeX Font Info: Checking defaults for OMS/cmsy/m/n on input line 22.
LaTeX Font Info: ... okay on input line 22.
LaTeX Font Info: Checking defaults for OT1/cmr/m/n on input line 22.
LaTeX Font Info: ... okay on input line 22.
LaTeX Font Info: Checking defaults for T1/cmr/m/n on input line 22.
LaTeX Font Info: ... okay on input line 22.
LaTeX Font Info: Checking defaults for TS1/cmr/m/n on input line 22.
LaTeX Font Info: ... okay on input line 22.
LaTeX Font Info: Checking defaults for OMX/cmex/m/n on input line 22.
LaTeX Font Info: ... okay on input line 22.
LaTeX Font Info: Checking defaults for U/cmr/m/n on input line 22.
LaTeX Font Info: ... okay on input line 22.
(/usr/share/texlive/texmf-dist/tex/context/base/mkii/supp-pdf.mkii
[Loading MPS to PDF converter (version 2006.09.02).]
\scratchcounter=\count397
\scratchdimen=\dimen159
\scratchbox=\box54
\nofMPsegments=\count398
\nofMParguments=\count399
\everyMPshowfont=\toks20
\MPscratchCnt=\count400
\MPscratchDim=\dimen160
\MPnumerator=\count401
\makeMPintoPDFobject=\count402
\everyMPtoPDFconversion=\toks21
\scratchcounter=\count420
\scratchdimen=\dimen172
\scratchbox=\box58
\nofMPsegments=\count421
\nofMParguments=\count422
\everyMPshowfont=\toks26
\MPscratchCnt=\count423
\MPscratchDim=\dimen173
\MPnumerator=\count424
\makeMPintoPDFobject=\count425
\everyMPtoPDFconversion=\toks27
) (/usr/share/texlive/texmf-dist/tex/latex/epstopdf-pkg/epstopdf-base.sty
Package: epstopdf-base 2020-01-24 v2.11 Base part for package epstopdf
(/usr/share/texlive/texmf-dist/tex/latex/grfext/grfext.sty
Package: grfext 2019/12/03 v1.3 Manage graphics extensions (HO)
(/usr/share/texlive/texmf-dist/tex/generic/kvdefinekeys/kvdefinekeys.sty
Package: kvdefinekeys 2019-12-19 v1.6 Define keys (HO)
))
Package epstopdf-base Info: Redefining graphics rule for `.eps' on input line 485.
Package grfext Info: Graphics extension search list:
(grfext) [.pdf,.png,.jpg,.mps,.jpeg,.jbig2,.jb2,.PDF,.PNG,.JPG,.JPEG,.JBIG2,.JB2,.eps]
(grfext) \AppendGraphicsExtensions on input line 504.
(/usr/share/texlive/texmf-dist/tex/latex/latexconfig/epstopdf-sys.cfg
File: epstopdf-sys.cfg 2010/07/13 v1.3 Configuration of (r)epstopdf for TeX Live
))
@ -424,103 +538,149 @@ Package biblatex Info: Automatic encoding selection.
Package biblatex Info: Trying to load bibliographic data...
Package biblatex Info: ... file 'Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl' found.
(./Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.bbl)
Package biblatex Info: Reference section=0 on input line 15.
Package biblatex Info: Reference segment=0 on input line 15.
Package biblatex Info: Reference section=0 on input line 22.
Package biblatex Info: Reference segment=0 on input line 22.
Package caption Info: Begin \AtBeginDocument code.
Package caption Info: listings package is loaded.
Package caption Info: End \AtBeginDocument code.
LaTeX Font Info: External font `cmex10' loaded for size
(Font) <12> on input line 17.
LaTeX Font Info: External font `cmex10' loaded for size
(Font) <8> on input line 17.
LaTeX Font Info: External font `cmex10' loaded for size
(Font) <6> on input line 17.
\c@lstlisting=\count426
[1
{/var/lib/texmf/fonts/map/pdftex/updmap/pdftex.map}] (./Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.toc
LaTeX Font Info: External font `cmex10' loaded for size
(Font) <7> on input line 8.
LaTeX Font Info: External font `cmex10' loaded for size
(Font) <5> on input line 8.
)
{/var/lib/texmf/fonts/map/pdftex/updmap/pdftex.map}] (./Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.toc)
\tf@toc=\write4
\openout4 = `Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.toc'.
[1]
<../poster/Concept.png, id=14, 338.2236pt x 211.992pt>
[1] [2]
<../poster/MNIST_excerpt.png, id=23, 99.37125pt x 100.87688pt>
File: ../poster/MNIST_excerpt.png Graphic file (type png)
<use ../poster/MNIST_excerpt.png>
Package pdftex.def Info: ../poster/MNIST_excerpt.png used on input line 81.
(pdftex.def) Requested size: 155.02pt x 157.38574pt.
<../poster/CIFAR-10_excerpt.png, id=24, 272.51813pt x 273.27094pt>
File: ../poster/CIFAR-10_excerpt.png Graphic file (type png)
<use ../poster/CIFAR-10_excerpt.png>
Package pdftex.def Info: ../poster/CIFAR-10_excerpt.png used on input line 85.
(pdftex.def) Requested size: 155.02pt x 155.44913pt.
<../poster/Concept.png, id=25, 338.2236pt x 211.992pt>
File: ../poster/Concept.png Graphic file (type png)
<use ../poster/Concept.png>
Package pdftex.def Info: ../poster/Concept.png used on input line 40.
Package pdftex.def Info: ../poster/Concept.png used on input line 98.
(pdftex.def) Requested size: 234.8775pt x 147.21593pt.
[2 <../poster/Concept.png>]
<../examples/unfiltered_eps_0.2.png, id=22, 682.2288pt x 708.9687pt>
[3 <../poster/MNIST_excerpt.png (PNG copy)> <../poster/CIFAR-10_excerpt.png (PNG copy)>] [4 <../poster/Concept.png>]
<../examples/unfiltered_eps_0.2.png, id=34, 682.2288pt x 708.9687pt>
File: ../examples/unfiltered_eps_0.2.png Graphic file (type png)
<use ../examples/unfiltered_eps_0.2.png>
Package pdftex.def Info: ../examples/unfiltered_eps_0.2.png used on input line 66.
Package pdftex.def Info: ../examples/unfiltered_eps_0.2.png used on input line 137.
(pdftex.def) Requested size: 58.71938pt x 61.01338pt.
<../examples/bit_depth_eps_0.2.png, id=23, 682.2288pt x 708.9687pt>
<../examples/bit_depth_eps_0.2.png, id=35, 682.2288pt x 708.9687pt>
File: ../examples/bit_depth_eps_0.2.png Graphic file (type png)
<use ../examples/bit_depth_eps_0.2.png>
Package pdftex.def Info: ../examples/bit_depth_eps_0.2.png used on input line 67.
Package pdftex.def Info: ../examples/bit_depth_eps_0.2.png used on input line 138.
(pdftex.def) Requested size: 58.71938pt x 61.01338pt.
<../examples/gaussian_blur_eps_0.2.png, id=24, 682.2288pt x 708.9687pt>
<../examples/gaussian_blur_eps_0.2.png, id=36, 682.2288pt x 708.9687pt>
File: ../examples/gaussian_blur_eps_0.2.png Graphic file (type png)
<use ../examples/gaussian_blur_eps_0.2.png>
Package pdftex.def Info: ../examples/gaussian_blur_eps_0.2.png used on input line 68.
Package pdftex.def Info: ../examples/gaussian_blur_eps_0.2.png used on input line 139.
(pdftex.def) Requested size: 58.71938pt x 61.01338pt.
<../examples/gaussian_kuwahara_eps_0.2.png, id=25, 682.2288pt x 708.9687pt>
<../examples/gaussian_kuwahara_eps_0.2.png, id=37, 682.2288pt x 708.9687pt>
File: ../examples/gaussian_kuwahara_eps_0.2.png Graphic file (type png)
<use ../examples/gaussian_kuwahara_eps_0.2.png>
Package pdftex.def Info: ../examples/gaussian_kuwahara_eps_0.2.png used on input line 69.
Package pdftex.def Info: ../examples/gaussian_kuwahara_eps_0.2.png used on input line 140.
(pdftex.def) Requested size: 58.71938pt x 61.01338pt.
<../examples/mean_kuwahara_eps_0.2.png, id=26, 682.2288pt x 708.9687pt>
<../examples/mean_kuwahara_eps_0.2.png, id=38, 682.2288pt x 708.9687pt>
File: ../examples/mean_kuwahara_eps_0.2.png Graphic file (type png)
<use ../examples/mean_kuwahara_eps_0.2.png>
Package pdftex.def Info: ../examples/mean_kuwahara_eps_0.2.png used on input line 71.
Package pdftex.def Info: ../examples/mean_kuwahara_eps_0.2.png used on input line 142.
(pdftex.def) Requested size: 58.71938pt x 61.01338pt.
<../examples/random_noise_eps_0.2.png, id=27, 682.2288pt x 708.9687pt>
<../examples/random_noise_eps_0.2.png, id=39, 682.2288pt x 708.9687pt>
File: ../examples/random_noise_eps_0.2.png Graphic file (type png)
<use ../examples/random_noise_eps_0.2.png>
Package pdftex.def Info: ../examples/random_noise_eps_0.2.png used on input line 72.
Package pdftex.def Info: ../examples/random_noise_eps_0.2.png used on input line 143.
(pdftex.def) Requested size: 58.71938pt x 61.01338pt.
<../examples/bilateral_filter_eps_0.2.png, id=28, 682.2288pt x 708.9687pt>
<../examples/bilateral_filter_eps_0.2.png, id=40, 682.2288pt x 708.9687pt>
File: ../examples/bilateral_filter_eps_0.2.png Graphic file (type png)
<use ../examples/bilateral_filter_eps_0.2.png>
Package pdftex.def Info: ../examples/bilateral_filter_eps_0.2.png used on input line 73.
Package pdftex.def Info: ../examples/bilateral_filter_eps_0.2.png used on input line 144.
(pdftex.def) Requested size: 58.71938pt x 61.01338pt.
<../examples/threshold_eps_0.2.png, id=29, 682.2288pt x 708.9687pt>
<../examples/threshold_eps_0.2.png, id=41, 682.2288pt x 708.9687pt>
File: ../examples/threshold_eps_0.2.png Graphic file (type png)
<use ../examples/threshold_eps_0.2.png>
Package pdftex.def Info: ../examples/threshold_eps_0.2.png used on input line 74.
Package pdftex.def Info: ../examples/threshold_eps_0.2.png used on input line 145.
(pdftex.def) Requested size: 58.71938pt x 61.01338pt.
[3 <../examples/unfiltered_eps_0.2.png> <../examples/bit_depth_eps_0.2.png> <../examples/gaussian_blur_eps_0.2.png> <../examples/gaussian_kuwahara_eps_0.2.png> <../examples/mean_kuwahara_eps_0.2.png> <../examples/random_noise_eps_0.2.png> <../examples/bilateral_filter_eps_0.2.png> <../examples/threshold_eps_0.2.png>]
<../results/MNIST_FGSM_3d_new.png, id=42, 650.43pt x 722.7pt>
[5 <../examples/unfiltered_eps_0.2.png> <../examples/bit_depth_eps_0.2.png> <../examples/gaussian_blur_eps_0.2.png> <../examples/gaussian_kuwahara_eps_0.2.png> <../examples/mean_kuwahara_eps_0.2.png> <../examples/random_noise_eps_0.2.png> <../examples/bilateral_filter_eps_0.2.png> <../examples/threshold_eps_0.2.png>]
Package amsmath Warning: Foreign command \over;
(amsmath) \frac or \genfrac should be used instead
(amsmath) on input line 179.
(/usr/share/texlive/texmf-dist/tex/latex/listings/lstlang1.sty
File: lstlang1.sty 2023/02/27 1.9 listings language file
) (../examples/pseudocode.py
LaTeX Font Info: Trying to load font information for OMS+cmr on input line 2.
(/usr/share/texlive/texmf-dist/tex/latex/base/omscmr.fd
File: omscmr.fd 2022/07/10 v2.5l Standard LaTeX font definitions
)
LaTeX Font Info: Font shape `OMS/cmr/m/n' in size <10> not available
(Font) Font shape `OMS/cmsy/m/n' tried instead on input line 2.
[6])
<../results/MNIST_FGSM_3d_new.png, id=60, 650.43pt x 722.7pt>
File: ../results/MNIST_FGSM_3d_new.png Graphic file (type png)
<use ../results/MNIST_FGSM_3d_new.png>
Package pdftex.def Info: ../results/MNIST_FGSM_3d_new.png used on input line 116.
Package pdftex.def Info: ../results/MNIST_FGSM_3d_new.png used on input line 193.
(pdftex.def) Requested size: 155.02pt x 172.23851pt.
<../results/CIFAR-10_FGSM_3d.png, id=43, 650.43pt x 722.7pt>
<../results/CIFAR-10_FGSM_3d.png, id=61, 650.43pt x 722.7pt>
File: ../results/CIFAR-10_FGSM_3d.png Graphic file (type png)
<use ../results/CIFAR-10_FGSM_3d.png>
Package pdftex.def Info: ../results/CIFAR-10_FGSM_3d.png used on input line 117.
Package pdftex.def Info: ../results/CIFAR-10_FGSM_3d.png used on input line 194.
(pdftex.def) Requested size: 155.02pt x 172.23851pt.
[4] [5 <../results/MNIST_FGSM_3d_new.png> <../results/CIFAR-10_FGSM_3d.png>] [6] [7] [8] (./Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux)
[7 <../results/MNIST_FGSM_3d_new.png> <../results/CIFAR-10_FGSM_3d.png>] [8] [9] [10]
Overfull \hbox (0.96643pt too wide) in paragraph at lines 273--273
[]\OT1/cmr/m/n/10 Ian Good-fel-low, Yoshua Ben-gio, and Aaron Courville. \OT1/cmr/m/it/10 Deep Learn-ing\OT1/cmr/m/n/10 . $\OT1/cmtt/m/n/10 http : / / www . deeplearningbook .
[]
[11]
<../results/MNIST_FGSM_line_strength1.png, id=81, 1156.32pt x 650.43pt>
File: ../results/MNIST_FGSM_line_strength1.png Graphic file (type png)
<use ../results/MNIST_FGSM_line_strength1.png>
Package pdftex.def Info: ../results/MNIST_FGSM_line_strength1.png used on input line 279.
(pdftex.def) Requested size: 469.75502pt x 264.23653pt.
<../results/MNIST_FGSM_line_strength4.png, id=82, 1156.32pt x 650.43pt>
File: ../results/MNIST_FGSM_line_strength4.png Graphic file (type png)
<use ../results/MNIST_FGSM_line_strength4.png>
Package pdftex.def Info: ../results/MNIST_FGSM_line_strength4.png used on input line 286.
(pdftex.def) Requested size: 469.75502pt x 264.23653pt.
<../results/CIFAR-10_FGSM_line_strength1.png, id=83, 1156.32pt x 650.43pt>
File: ../results/CIFAR-10_FGSM_line_strength1.png Graphic file (type png)
<use ../results/CIFAR-10_FGSM_line_strength1.png>
Package pdftex.def Info: ../results/CIFAR-10_FGSM_line_strength1.png used on input line 294.
(pdftex.def) Requested size: 469.75502pt x 264.23653pt.
[12 <../results/MNIST_FGSM_line_strength1.png> <../results/MNIST_FGSM_line_strength4.png>]
<../results/CIFAR-10_FGSM_line_strength4.png, id=90, 1156.32pt x 650.43pt>
File: ../results/CIFAR-10_FGSM_line_strength4.png Graphic file (type png)
<use ../results/CIFAR-10_FGSM_line_strength4.png>
Package pdftex.def Info: ../results/CIFAR-10_FGSM_line_strength4.png used on input line 301.
(pdftex.def) Requested size: 469.75502pt x 264.23653pt.
[13 <../results/CIFAR-10_FGSM_line_strength1.png> <../results/CIFAR-10_FGSM_line_strength4.png>] (./Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.aux)
LaTeX Warning: There were multiply-defined labels.
Package logreq Info: Writing requests to 'Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.run.xml'.
\openout1 = `Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.run.xml'.
)
Here is how much of TeX's memory you used:
11539 strings out of 476041
224526 string characters out of 5793174
1861388 words of memory out of 6000000
31900 multiletter control sequences out of 15000+600000
517964 words of font info for 51 fonts, out of 8000000 for 9000
14441 strings out of 476041
265701 string characters out of 5793174
1869388 words of memory out of 6000000
34726 multiletter control sequences out of 15000+600000
524003 words of font info for 72 fonts, out of 8000000 for 9000
1137 hyphenation exceptions out of 8191
63i,7n,68p,1221b,1248s stack positions out of 10000i,1000n,20000p,200000b,200000s
</usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmbx10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmbx12.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmbx9.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmcsc10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmi10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmi7.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr12.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr17.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr7.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr9.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmsy10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmsy7.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmti10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmtt10.pfb>
Output written on Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.pdf (9 pages, 646007 bytes).
63i,9n,68p,1489b,1806s stack positions out of 10000i,1000n,20000p,200000b,200000s
</usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmbx10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmbx12.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmbx9.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmcsc10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmi10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmi7.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmmib10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr12.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr17.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr7.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmr9.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmsy10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmsy7.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmti10.pfb></usr/share/texlive/texmf-dist/fonts/type1/public/amsfonts/cm/cmtt10.pfb>
Output written on Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.pdf (14 pages, 1331754 bytes).
PDF statistics:
132 PDF objects out of 1000 (max. 8388607)
68 compressed objects within 1 object stream
163 PDF objects out of 1000 (max. 8388607)
82 compressed objects within 1 object stream
0 named destinations out of 1000 (max. 500000)
56 words of extra memory for PDF output out of 10000 (max. 10000000)
86 words of extra memory for PDF output out of 10000 (max. 10000000)

BIN
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.pdf
View File

Binary file not shown.

BIN
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.synctex.gz
View File

Binary file not shown.

187
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.tex
View File

@ -1,10 +1,17 @@
\documentclass{article}
\usepackage{graphicx}
\usepackage{float}
\usepackage[margin=1in]{geometry}
\usepackage[backend=biber]{biblatex}
\usepackage{caption}
\usepackage{subcaption}
\usepackage{bm}
\usepackage{amsmath}
\usepackage{listings}
\usepackage[toc,page]{appendix}
\DeclareMathOperator{\sign}{sign}
\addbibresource{references.bib}
@ -22,18 +29,69 @@
\pagebreak
\begin{abstract}
With the sudden increase in machine learning applications brought forth in the last eighteen months or so, a new cybersecurity threat has been posed. Not only are traditional cloud-based threats a risk---denial of service attacks for example---AI services are also at risk of being sabotaged by adversarial examples. For this reason, standards have been passed by international governing bodies such as the IEEE to streamline the process of making such services more robust \cite{ieee3129}. To further support increased robustness for online AI services, we propose a pipeline that applies denoising filters to inputs before they reach the model. In doing so, the goal is to remove, or at least reduce the impact of, adversarial attacks. We find that in cases where data is particularly high contrast, some filters can maintain the classifying accuracy of the model at high attack strengths. However, in cases where the data is low contrast, the best filters are only able to keep accuracy slightly better than if the model was simply guessing classes at random.
With the sudden increase in machine learning applications brought forth in the last eighteen months or so, a new cybersecurity threat has been posed. Not only are traditional cloud-based threats a risk---denial of service attacks for example---AI services are also at risk of being sabotaged by adversarial examples. Several defense methods have been proposed in the last decade that have seen varying degrees of success and transferability across models and attack methods. Even the best defenses, however, have some drawbacks. One recent technique used a denoising autoencoder to essentially remove the attack from an image. However, this technique requires training the autoencoder and pipelining multiple machine learning models. To further support increased robustness for online AI services without a training stage and without multiple models, we designed a pipeline that instead applies denoising filters to inputs before they reach the model. In doing so, the goal is to maintain the accuracy of the classifier. Our tests on a classifier trained on the MNIST dataset show that some filters could maintain the classifying accuracy of the model at high attack strengths. However, the when instead a classifier was trained on CIFAR-10, a lower-contrast dataset, even the best filters were only able to keep accuracy slightly better than if the model was guessing classes at random.
\end{abstract}
\section{Introduction}
All neural networks are prone to adversarial attacks \cite{carlini2017evaluating} \cite{szegedy2014intriguing}. When it comes to defending against these attacks, one method is to train using both adversarial (attacked) and clean (not attacked) data. While this method has been shown to somewhat regularize the neural network \cite{goodfellow2015explaining}, the technique fails to generalize to other attacks. A defense technique, from which our approach draws inspiration, involves using a denoising autoencoder to produce only clean data from both adversarial and clean data. This technique shows promise and transferability across a range of attacks.
All neural networks are prone to adversarial attacks \cite{carlini2017evaluating} \cite{szegedy2014intriguing}. When it comes to defending against these attacks, one method is to train using both adversarial (attacked) and clean (not attacked) data. While this method has been shown to somewhat regularize the neural network \cite{goodfellow2015explaining}, the technique fails to generalize to other attacks. Another defense technique, from which our approach draws inspiration, involves using a denoising autoencoder to produce only clean data from both adversarial and clean data. This technique shows promise and transferability across a range of attacks.
Unfortunately, this approach involves training a secondary model to protect the original, which can drastically increase the amount of computing power required for set up. Therefore, in scenarios where reducing computing power is a factor or a simple drop-in defense is required, we propose a pipeline in which commonly available denoising filters are used to pre-process inputs before being fed to the model.
Unfortunately, this approach involves training a secondary model to protect the original, which can drastically increase the amount of computing power required. Therefore, in scenarios where reducing computing power is important or a simple drop-in defense is required, we propose a pipeline in which commonly available denoising filters are used to pre-process inputs before being fed to the model.
By implementing this approach, we examined whether preprocessing filters are effective at defending adversarial attacks. To do so, we compared the efficacy of different filters at different filtering strengths. Finally, we plan to investigate the transferability of this approach across models and attacks.
By implementing this approach, we examined whether preprocessing filters are effective at defending adversarial attacks. Therefore, we set out to:
\begin{enumerate}
\item Compare the efficacy of different filters at different attack and filtering strengths
\item Investigate the efficacy of different filters on different datasets and classifier architectures
\item Investigate the effectiveness of a filter-based defense across multiple adversarial attacks
\end{enumerate}
\section{Approach and Experimental Timeline}
We began by establishing a baseline performance with an FGSM attack on a pre-trained MNIST CNN classifier. After verifying that the attack was working, the first filter could be implemented: a Gaussian Kuwahara filter. The Kuwahara filter was initially chosen because it excels at removing noise from smooth areas of an image while also preserving edges. Once this preliminary pipeline was up and running, the code was made more modular with a standard interface to facilitate swapping out models, attacks, and filters.
\section{Approach and Background}
\label{sec:approach_background}
Consider an input image $\bm{x}$. We obtain its attacked counterpart $\tilde{\bm{x}}$ by adding some calculated perturbation.
\begin{equation}
\tilde{\bm{x}} = \bm{x} + \bm{\eta}
\end{equation}
where $\bm{\eta}$ is the calculated perturbation for the given attack method. For the case of the FGSM attack,
\begin{equation}
\bm{\eta} = \epsilon \sign(\nabla_x J(\bm{\theta}, \bm{x}, y))
\label{eqn:fgsm}
\end{equation}
where $J(\bm{\theta}, \bm{x}, y)$ is the cost function used to train the neural network and $\epsilon$ is a free parameter indicating the strength of the attack \cite{goodfellow2015explaining}. For our method we used $\epsilon \in [0,0.3]$ with a step size of 0.025.
The goal of the denoising autoencoder approach, which we drew inspiration from, is to remove the perturbation and recover the original image. This approach requires a training process on to minimize the error, so that as error decreases the output given a noisy input as similar as possible to the unperturbed case. Specifically, while a regular autoencoder attempts to essentially compress and recover an image, minimizing its loss function:
\begin{equation}
L(\bm{x}, g(f(\bm{x})))
\end{equation}
where $L$ is the loss function, $g(\bm{h})$ is the decoded output and $\bm{h} = f(\bm{x})$ is the encoder output, a denoising autoencoder tries to minimize a slightly different loss:
\begin{equation}
L(\bm{x}, g(f(\tilde{\bm{x}})))
\end{equation}
In doing so, the original image can be almost perfectly recovered from the perturbed one \cite{goodfellow2016deep}.
Solving this minimization problem requires additional computational power that a filtering approach would not. Unfortunately, hoping to use a simple filter to fully recover an image is too lofty a goal. However, our goal should not be recovering the image at all, since the point of a defense is only to get the same output from the model. Therefore, given an ideal filter $F$ and a classifier $C$,
\begin{equation}
C(F(\tilde{\bm{x}})) = C(F(\bm{x})) = C(\bm{x})
\end{equation}
Extending this concept to accuracy, for an ideal filter, the accuracy of the classifier with clean inputs should be the same as the accuracy of the classifier with filtered inputs whether attacked or not.
Demonstrating filters on only one attack or only one dataset would not provide any information on the transferability of our approach. To test dataset transferability we selected MNIST and CIFAR-10, excerpts of each are seen in figure \ref{fig:dataset_excerpts}. To test attack transferability, we began with the FGSM attack given in equation \ref{eqn:fgsm} and plan to implement the Carlini and Wagner attack soon \cite{carlini2017evaluating}.
\begin{figure}[H]
\centering
\begin{subfigure}[b]{0.33\textwidth}
\includegraphics[width=\textwidth]{../poster/MNIST_excerpt.png}
\caption{MNIST dataset excerpt}
\end{subfigure}
\begin{subfigure}[b]{0.33\textwidth}
\includegraphics[width=\textwidth]{../poster/CIFAR-10_excerpt.png}
\caption{CIFAR-10 dataset excerpt}
\end{subfigure}
\caption{Excerpts each dataset used with elements arranged by class}
\label{fig:dataset_excerpts}
\end{figure}
\section{Experimental Timeline}
\label{sec:exp_timeline}
We began by establishing a baseline performance with an FGSM attack on a pre-trained MNIST CNN classifier. After verifying that the attack was working, the first filter could be implemented: a Gaussian Kuwahara filter. The Kuwahara filter was initially chosen because it excels at blurring smooth areas of an image while also preserving edges. Once this preliminary pipeline was up and running, the code was made more modular with a standard interface to facilitate swapping out models, attacks, and filters.
\begin{figure}[H]
\centering
@ -44,22 +102,35 @@ We began by establishing a baseline performance with an FGSM attack on a pre-tra
With this interface constructed, all alternative filters were tested over a range of attack strengths and with a range of variation on each filter's free parameter. This free parameter is referred to as filtering "strength" for a lack of better terms, although some filters, such as bit-depth reduction, make larger alterations at lower "strength".
To prevent rerunning the program every time we wanted to change how results were visualized, data for each run was saved in the JSON format specified by the ECMA 404 standard \cite{ecma404}. This allowed for a companion program to read and display results in a number of ways.
To prevent the need to rerun the filter testing program every time we wanted to change how results were visualized, data for each run was saved in the JSON format specified by the ECMA-404 standard \cite{ecma404}. This allowed for a companion program to read and display results in a number of ways.
Once we had results for MNIST, we shifted focus to implementing a CIFAR-10 classifier. Originally, we attempted to adapt the LeNet classifier we were already using. This was only able to achieve about 65\% to 70\% accuracy on the validation dataset before plateauing. Therefore, we decided to research what others had found success with. We started to test DLA \cite{yu2019deep}, but after taking too long to train, we pivoted again to VGG16, which we stuck with for testing on filters \cite{simonyan2014very}.
Once we had results for MNIST, we shifted focus to implementing a CIFAR-10 classifier. Originally, we attempted to adapt the LeNet classifier we were already using. This could achieve about 65\% to 70\% accuracy on the validation dataset before plateauing. Therefore, we decided to research what others had found success with. We started to test DLA \cite{yu2019deep}, but after taking too long to train, we pivoted again to VGG16, which we stuck with for testing on filters \cite{simonyan2014very}.
Once we had two working classifiers, LeNet trained on MNIST and VGG16 trained on CIFAR-10, we were able to compare the effectiveness of the filters across models.
The next logical progression, which will be implemented in the near future, is to implement another attack to evaluate the transferability of our approach across attack methods. This attack is likely to be Carlini and Wagner \cite{carlini2017evaluating}.
The next logical progression, which will be implemented in the near future, is another attack to evaluate the transferability of our approach across attack methods. Currently we plan to use the Carlini and Wagner attack \cite{carlini2017evaluating}.
We specifically chose these two attacks because they appear in the list of example transferable adversarial attacks (Annex A.2.1 of IEEE 3129-2023) \cite{ieee3129}. Additionally both FGSM and Carlini and Wagner attacks were tested on the desnoising autoencoder approach, so by implementing them, comparing results will be much easier.
\section{Constraints}
In switching from LeNet to deeper networks, two constraints were posed. Firstly, the GPU in the machine used did not support any machine learning hardware acceleration tools like CUDA or ROCm. Being forced to do everything on the CPU imposed a limit on how much training and testing could be done within a reasonable amount of time. Limited computing resources also guided our decision to pick CIFAR-10 over a higher-resolution dataset like ImageNet.
In switching from LeNet to deeper networks, two constraints were posed. Firstly, the GPU in the machine used did not support any machine learning hardware acceleration tools like CUDA or ROCm. Being forced to do everything on the CPU imposed a limit on how much training and testing could be done within a reasonable amount of time. Essentially, hardware options put a limit on model complexity. Limited computing resources also guided our decision to pick CIFAR-10 over a higher-resolution dataset like ImageNet.
The other constraint was posed by the version management system used. While Git itself does not really have a limit, the current server-side setup requires that files be under about 100 MB each. Since we wanted to store the trained models alongside the other files in the repository, this limited the number of parameters our models had. While the largest model used in testing was 62MiB, the limit significantly narrowed the list of possible choices.
\section{Applicable Engineering Standards}
Both standards used are mentioned throughout. This section serves mainly as an overview of each standard and why it was selected. Both standards were highly valuable resources, and an overview alone would not do them justice. Specifically, IEEE 3129 is discussed further in sections \ref{sec:approach_background}, \ref{sec:exp_timeline}, and \ref{sec:future_work}, and ECMA-404 is discussed further in section \ref{sec:exp_timeline}.
\subsection{IEEE 3129-2023}
IEEE standard 3129-2023, \emph{IEEE Standard for Robustness Testing and Evaluation of Artificial Intelligence (AI)-based Image Recognition Service} proved itself an invaluable resource throughout the project. Definitions and experimental processes alike were informed by this standard. For example, IEEE 3129 provides, in Annex A.2.1, examples of attacks that must be defended against for a defense to be considered \emph{transferable}. Therefore, our experimental process selected attacks from their list \cite{ieee3129}.
\subsection{ECMA-404}
Standard ECMA-404, \emph{The JSON Data Interchange Syntax} specifies the JSON file format: a text-based serialization format, syntactically similar to Python dictionaries \cite{ecma404}. This file format was selected as the primary method of storing results due to its popularity across engineering and non-engineering disciplines, its easy-to-use Python library, and most importantly, human-readability. By using the JSON file format for our results, any issues were debugged rapidly, and any data reformatting could be done with a very short helper program.
\section{Alternative Filters}
Seven image processing filters were selected, and their performances were compared over a range of attack strengths, while parametrically varying each filter's free parameter. In this way, the optimal parameters for each filter could be found, and their effectiveness against adversarial examples could be examined. An overview of how each filter operates on a randomly selected MNIST example attacked with FGSM with $\epsilon=0.2$ is seen in figure \ref{fig:alternative_filters}.
Seven image processing filters were selected, and their performances were compared over a range of attack strengths, while parametrically varying filter strength. In this way, the optimal parameters for each filter could be found, and their effectiveness against adversarial examples could be examined. An overview of how each filter operates on a randomly selected MNIST example attacked with FGSM with $\epsilon=0.2$ is seen in figure \ref{fig:alternative_filters}.
\begin{figure}[H]
\centering
@ -76,7 +147,7 @@ Seven image processing filters were selected, and their performances were compar
\label{fig:alternative_filters}
\end{figure}
The first set of filters examined were the Kuwahara filters, both Gaussian and Mean variants. They operate based on the variance of four overlapping regions in a kernel. These filters were selected due to their ability to preserve edges and blur smooth areas: a desirable property for removing noise. The Kuwahara filter is typically used to implement an oil painting effect in some software packages, but was originally used to make interpreting medical imaging easier \cite{kuwahara1976processing}. The free parameter for both Kuwahara filters is the radius of the kernel, and its relation to filter strength is given by:
The first set of filters examined were the Kuwahara filters, both Gaussian and Mean variants. They operate based on the variance of four overlapping regions in a kernel. These filters were selected due to their ability to preserve edges and blur smooth areas: a desirable property for removing noise. The Kuwahara filter is commonly used to implement an oil painting effect in some software packages, but was originally used to make interpreting medical imaging easier \cite{kuwahara1976processing}. The free parameter for both Kuwahara filters is the radius of the kernel, and its relation to filter strength is given by:
\begin{equation}
r = 2s + 1
\label{eqn:rad_odd_strength}
@ -101,7 +172,7 @@ While not typically used as a filter, a bit-depth reduction was chosen because i
\begin{equation}
b = 2^s
\end{equation}
where $b$ is the number of bits. The bit-depth reduction is one such case where increasing strength is produces counter-intuitive effects.
where $b$ is the number of bits. The bit-depth reduction is one such case where increasing the filter strength makes the output more similar to the input.
The final filter tested was a simple threshold. For each channel, the value would set to the maximum value when above the threshold and to the minimum value when below the threshold. For a floating-point greyscale image, the maximum value is one and the minimum value is zero. The threshold filter is another such filter where mapping the free parameter to strength seems odd. The free parameter is the threshold, which is given by
\begin{equation}
@ -110,6 +181,12 @@ The final filter tested was a simple threshold. For each channel, the value woul
where $T$ is the threshold. Perhaps this definition could be better as it restricts strength to be values in the range $[-0.5,4.5]$. Regardless, this is the definition used in testing to get an even spread of thresholds between zero and one.
\section{Results and Evaluation}
In Python-like pseudocode, a rough outline of our testing procedure is:
\lstinputlisting[language=Python, caption=Testing procedure overview]{../examples/pseudocode.py}
With this procedure, all filters are tested automatically at all filtering and attack strengths. To change the model, a different \verb|Net| class is imported, and to change the attack, the \verb|fgsm_attack()| function may be swapped out.
Since both CIFAR-10 and MNIST include a validation datasets, only one run was conducted for each model. Both datasets contain 10,000 test images, which should provide a good estimate on the accuracy of each model over each parameter's range.
With the two free parameters, filter strength and attack strength, the accuracy of each filter can be thought of as a point in 3D space. For each filter strength and attack strength, the highest accuracy filter was chosen to represent that point. For our purposes, the accuracy was defined as the ratio of correct predictions to total predictions. The best filters and their accuracies for each attack strength and filter strength for MNIST and CIFAR-10 are seen in figure \ref{fig:filter_performances}.
\begin{figure}[H]
\centering
@ -125,49 +202,62 @@ The second measure of effectiveness we defined is a \emph{minimally effective} f
This definition was chosen, because if a classifier is performing better than randomly guessing, then it may be boosted to increase accuracy \cite{schapire1989strength}.
The results as seen in figure \ref{fig:filter_performances} only displays the highest rank filter for each attack and filter strength. To get a better comparison of the filters, several additional views are listed in the appendices.
\section{Conclusions}
As seen in figure \ref{fig:filter_performances}, the threshold filter operates as a nearly ideal filter for MNIST even at the highest attack strengths tested. Additionally for the MNIST case, the Gaussian Kuwahara filter prevails at high filter strength and high attack strength.
For CIFAR-10, however, the results are not nearly as promising. While for every attack strength tested there was a filter at some strength that kept the classifier above the random guessing threshold, there were no filters that came even close to the ideal case.
We hypothesize that, at least for low-resolution datasets, having low contrast images makes defending attacks more difficult. Taking a look at what the threshold filter does to an attacked image in figure \ref{fig:alternative_filters}, it becomes apparent why it works so well for MNIST. Essentially, when attacked, the SNR is still quite low.
We conjecture that, at least for low-resolution datasets, having low contrast images makes defending attacks more difficult. Taking a look at what the threshold filter does to an attacked image in figure \ref{fig:alternative_filters}, it becomes apparent why it works so well for MNIST. Essentially, when attacked, the SNR is still quite low.
In the case of CIFAR-10, discerning the clean images can be difficult for a human observer at times. Therefore, it is understandable why FGSM impacted CIFAR-10 much more than MNIST. To better understand how filters apply to color photos in general, testing with the ImageNet database is a logical next step.
In the case of CIFAR-10, discerning the images can be difficult at times even for a human observer. Therefore, it is understandable why FGSM impacted CIFAR-10 much more than MNIST. To better understand how filters apply to color photos in general, testing with the ImageNet database is a logical next step.
\section{Ethical Considerations}
\subsection{Heath and Safety Considerations}
When it comes to image classifier services, many tools will not have any direct impact on the health and safety of others. In some cases, however, if the classifier is being used to operate machinery, such as in self-driving vehicles, having a robust pipeline is important to keeping people out of harm's way. Additionally, for services such as plant classifiers, if a poisonous plant is identified as safe to eat, people may fall ill or perish.
\section{Ethical Considerations and Impact Factors}
Ethical considerations were any ethical issues that we found that may arise as a direct effect of or adjacent to our work, while impact factors were our solutions and guidance informed by those considerations.
Therefore, as such machine learning services become more popular, being robust to adversarial attacks is of the utmost importance to human health. For this reason, unless further research shows that filtering is an effective defense for higher resolution color images, we cannot recommend that filters be used as an adversarial defense in any service that may involve machinery, food, or any other potentially hazardous element.
\subsection{Health and Safety Ethical Considerations}
When it comes to image classifier services, many tools will not have any direct impact on the health and safety of others. In some cases, however, if the classifier is being used to operate machinery, such as in self-driving vehicles, having a robust pipeline is important to keeping people out of harm's way. Additionally, for services such as plant classifiers, if attacked, and a poisonous plant is identified as safe to eat, people may fall ill or perish. Therefore, as such machine learning services become more popular, being robust to adversarial attacks is of the utmost importance to human health.
We initially chose the filtering approach because it is potentially faster than a denoising autoencoder. This way, if our approach is found to be effective against adversarial attacks, we can increase the safety of equipment operators in time-sensitive operations.
\subsection{Heath and Safety Impact Factors}
For this reason, unless further research shows that filtering is an effective defense for higher resolution color images, we cannot recommend that filters be used as an adversarial defense in any service that may involve machinery, food, or any other potentially hazardous element.
\subsection{Social and Cultural Considerations}
Be it source code, experimental procedures, raw results, or otherwise, having closed-source anything in an academic project should raise a red flag. To ensure equal access and full transparency with the readers and those wishing to reproduce the results found in this paper, all code, results, document files, and their editing histories are free and open source (FOSS) under the MIT license. All software packages and external libraries used were also ensured to be FOSS before use.
Additionally, one of the reasons we chose to investigate a filtering approach was because it is potentially faster than a denoising autoencoder. This way, if our approach is found to be effective against adversarial attacks, we can increase the safety of equipment operators in time-sensitive operations.
\subsection{Environmental Considerations}
One possible source for environmental impact, when it comes to adversarial defenses, is reducing energy consumption. Outside of academia, the most likely victim for an adversarial attack is AI web services hosted in data centers. These data centers are always online, so by reducing the energy use, the amount of fossil fuels consumed to power the data center can be reduced. For services dealing with only high-contrast, greyscale datasets, we recommend looking into a filter defense to avoid wasting energy training a denoising autoencoder.
\subsection{Social and Cultural Ethical Considerations}
Be it source code, experimental procedures, raw results, or otherwise, having closed-source anything in an academic project should raise a red flag. In too many cases papers have been retracted due to academic dishonesty or improper manipulation of data.
While admittedly, slightly reducing energy consumption will not make the difference in the long run, doing anything to reduce power consumption helps.
\subsection{Social and Cultural Impact Factors}
To ensure equal access and full transparency with the readers and those wishing to reproduce the results found in this paper, all code, results, document files, and their editing histories are free and open source (FOSS) under the MIT license. All software packages and external libraries used were also ensured to be FOSS before use.
\subsection{Economic and Financial Considerations}
As noted several times previously, online services have new cybersecurity risks when it comes to machine learning. Dealing with adversarial examples may result in financial damages whether in the form of lawsuits or lost sales due to sabotage.
\subsection{Environmental Ethical Considerations}
When it comes to adversarial defenses, energy consumption will likely impact the environment the most. Outside of academia, the most likely victim for an adversarial attack is AI web services hosted in data centers. These data centers are always online, so by reducing the energy use, the amount of fossil fuels consumed may be reduced greatly in the long-term.
One benefit of filtering over a denoising autoencoder is a deterministic output. Since filters behave like any other algorithmic method, the output of any input is easily determined. This way, auxiliary algorithms such as checksums or cryptographic signing can be used to verify inputs to the model. The determinism of filtering opens the door to enhanced cybersecurity integrations.
\subsection{Environmental Impact Factors}
For services dealing with only high-contrast, greyscale datasets, we recommend looking into a filter defense to avoid wasting energy training a denoising autoencoder. Additionally, we plan to perform a cradle to grave life-cycle assessments on both denoising autoencoders and denoising filters to determine which is better from an environmental perspective.
Unfortunately, this determinism may also put the service at risk from reverse engineered attacks, since it is much easier to reverse engineer an algorithm than an autoencoder's parameters. Still, the potential cybersecurity advantages that come with an algorithmic approach drove much of the curiosity behind this project.
\subsection{Economic and Financial Ethical Considerations}
As noted previously, online services have new cybersecurity risks when it comes to machine learning. Dealing with adversarial examples may result in financial damages whether in the form of lawsuits or lost sales due to sabotage.
We must also be aware that filters may pose a risk to reverse-engineering the defense mechanism. As our pipeline is made from commonly available components, an adversary may devise an attack that produces an adversarial example after the input is filtered.
\subsection{Economic and Financial Impact Factors}
One benefit of filtering over a denoising autoencoder is pre-deterministic output. Since filters behave like any other algorithmic method, the output of any input is easily determined. Compare this with an autoencoder: the output relies on the parameters of the model, which cannot possibly be known prior to the training process.
This pre-deterministic behavior may allow auxiliary algorithms such as checksums or cryptographic signing to be used to verify inputs to the model. This subtle trait of filtering opens the door to enhanced cybersecurity integrations, and drove much of the curiosity behind this project.
\subsection{Generative AI Disclaimer}
All code, results data, writing, figures, equations, typos, and errors, unless a source is cited, are the words and work of the authors listed on the title page. ChatGPT 3.5 was used to assist with the wording of the title of this paper, but no other generative AI was used in this project.
\section{Key Takeaways}
This project started with very little knowledge about machine learning, let alone adversarial machine learning. We had some prior experience working with image processing and filtering in Python, which partly inspired our approach. For the most part, however, all knowledge presented at the end of the semester was learned throughout the semester. Everything from builing custom models to implementing, training, and validating machine learning models, to learning how to attack a model, to even learning how to use Matplotlib's 3D functionionality.
This project started with very little knowledge about machine learning, let alone adversarial machine learning. We had some prior experience working with image processing and filtering in Python, which partly inspired our approach. For the most part, however, all knowledge presented at the end of the semester was learned throughout the semester. Everything from building custom models to implementing, training, and validating machine learning models, to learning how to attack a model, to even learning how to use Matplotlib's 3D functionality was all learned over the course of the project.
The primary source of information was library documentation, after that, it was teammates with prior experience in adversarial machine learning. Other questions not answerable by those sources were answered by reading prior papers published on filters and adversarial machine learning.
\section{Future Work}
While significant progress was made in developing the proposed pipeline, we did not meet all of our requirements, and therefore are not yet in a position to publish this work. To meet these requirements, we must still test at least one more attack, likely Carlini and Wagner \cite{carlini2017evaluating}.
\label{sec:future_work}
While significant progress was made in developing the proposed pipeline, we did not meet all of our requirements, and therefore are not yet in a position to publish this work. To meet these requirements, we must still test at least one more attack, likely Carlini and Wagner \cite{carlini2017evaluating}, and to fully show transferability, we would like to test all attacks listed in Annex A.2.1 of IEEE 3129-2023. This testing will assist any image classifier cloud services in developing a robust platform, the stated goal of IEEE 3129-2023 \cite{ieee3129}.
Additionally, we find it important to standardize the strength parameter with a more robust, likely SNR-based definition. In doing so, calling the parameter "strength" would make more intuitive sense, and it would provide a fairer comparison of the filters.
@ -176,8 +266,41 @@ Third, we would like to test the hypothesis that filtering is more energy effici
Finally, we plan to implement a few more alternative filters. Some of which will be quite easy to implement (simply a call to a library function), while others will require a custom implementation.
\section{Acknowledgements}
In any academic effort, we stand on the shoulders of giants. This project is no different and would not have been possible without help of several key resources. Special thanks to the PyTorch examples team for providing the baseline implementation of the LeNet classifier used on MNIST and the implementation of the FGSM attack with varying strengths. This project also heavily used the NumPy, Matplotlib, PyTorch libraries. Filters not implemented by the authors used the cv2 distribution of OpenCV for Python, and the Kuwahara filters were implemented using the pykuwahara library. Additionally, we would not have found VGG16 or DLA without the help from kuangliu on GitHub, who supplied the source code and performances of a set of models on CIFAR-10. To fix the rendering of Matplotlib's 3D bar chart, code from codeMonkey, marisano, and CYang on Stack Overflow was used. Finally, we would like to thank Dr. Robi Polikar and Chaz Allegra for their mentorship and unending support throughout the course of this project. Without their help and dedication, we would not have made nearly as much progress this semester.
In any academic effort, we stand on the shoulders of giants. This project is no different and would not have been possible without help of several key resources. Special thanks to the PyTorch examples team for providing the baseline implementation of the LeNet classifier used on MNIST and the implementation of the FGSM attack with varying strengths. This project also heavily used the NumPy, Matplotlib, and PyTorch libraries. Filters not implemented by the authors used the cv2 distribution of OpenCV for Python, and the Kuwahara filters were implemented using the pykuwahara library. Additionally, we would not have found VGG16 or DLA without the help from kuangliu on GitHub, who supplied the source code and performances of a set of models on CIFAR-10. To fix the rendering of Matplotlib's 3D bar chart, code from codeMonkey, marisano, and CYang on Stack Overflow was used. Finally, we would like to thank Dr. Robi Polikar and Chaz Allegra for their mentorship and unending support throughout the course of this project. Without their help and dedication, we would not have made nearly as much progress this semester.
\pagebreak
\printbibliography
\pagebreak
\appendix
\section{Additional Views of MNIST Results}
\begin{figure}[H]
\centering
\includegraphics[width=\textwidth]{../results/MNIST_FGSM_line_strength1.png}
\caption{Performances for each filter with filter strength 1 on MNIST attacked with FGSM}
\label{fig:mnist_fgsm_strength1}
\end{figure}
\begin{figure}[H]
\centering
\includegraphics[width=\textwidth]{../results/MNIST_FGSM_line_strength4.png}
\caption{Performances for each filter with filter strength 4 on MNIST attacked with FGSM}
\label{fig:mnist_fgsm_strength4}
\end{figure}
\section{Additional Views of CIFAR-10 Results}
\begin{figure}[H]
\centering
\includegraphics[width=\textwidth]{../results/CIFAR-10_FGSM_line_strength1.png}
\caption{Performances for each filter with filter strength 1 on CIFAR-10 attacked with FGSM}
\label{fig:mnist_fgsm_strength1}
\end{figure}
\begin{figure}[H]
\centering
\includegraphics[width=\textwidth]{../results/CIFAR-10_FGSM_line_strength4.png}
\caption{Performances for each filter with filter strength 4 on CIFAR-10 attacked with FGSM}
\label{fig:mnist_fgsm_strength4}
\end{figure}
\end{document}

38
report/Report_for_Enhancing_Image_Classifiers_with_Denoising_Filters_Sharpe_Polikar.toc
View File

@ -1,15 +1,25 @@
\contentsline {section}{\numberline {1}Introduction}{2}{}%
\contentsline {section}{\numberline {2}Approach and Experimental Timeline}{2}{}%
\contentsline {section}{\numberline {3}Constraints}{3}{}%
\contentsline {section}{\numberline {4}Alternative Filters}{3}{}%
\contentsline {section}{\numberline {5}Results and Evaluation}{4}{}%
\contentsline {section}{\numberline {6}Conclusions}{5}{}%
\contentsline {section}{\numberline {7}Ethical Considerations}{5}{}%
\contentsline {subsection}{\numberline {7.1}Heath and Safety Considerations}{5}{}%
\contentsline {subsection}{\numberline {7.2}Social and Cultural Considerations}{6}{}%
\contentsline {subsection}{\numberline {7.3}Environmental Considerations}{6}{}%
\contentsline {subsection}{\numberline {7.4}Economic and Financial Considerations}{6}{}%
\contentsline {subsection}{\numberline {7.5}Generative AI Disclaimer}{6}{}%
\contentsline {section}{\numberline {8}Key Takeaways}{6}{}%
\contentsline {section}{\numberline {9}Future Work}{7}{}%
\contentsline {section}{\numberline {10}Acknowledgements}{7}{}%
\contentsline {section}{\numberline {2}Approach and Background}{2}{}%
\contentsline {section}{\numberline {3}Experimental Timeline}{3}{}%
\contentsline {section}{\numberline {4}Constraints}{4}{}%
\contentsline {section}{\numberline {5}Applicable Engineering Standards}{5}{}%
\contentsline {subsection}{\numberline {5.1}IEEE 3129-2023}{5}{}%
\contentsline {subsection}{\numberline {5.2}ECMA-404}{5}{}%
\contentsline {section}{\numberline {6}Alternative Filters}{5}{}%
\contentsline {section}{\numberline {7}Results and Evaluation}{6}{}%
\contentsline {section}{\numberline {8}Conclusions}{8}{}%
\contentsline {section}{\numberline {9}Ethical Considerations and Impact Factors}{8}{}%
\contentsline {subsection}{\numberline {9.1}Health and Safety Ethical Considerations}{8}{}%
\contentsline {subsection}{\numberline {9.2}Heath and Safety Impact Factors}{8}{}%
\contentsline {subsection}{\numberline {9.3}Social and Cultural Ethical Considerations}{8}{}%
\contentsline {subsection}{\numberline {9.4}Social and Cultural Impact Factors}{8}{}%
\contentsline {subsection}{\numberline {9.5}Environmental Ethical Considerations}{9}{}%
\contentsline {subsection}{\numberline {9.6}Environmental Impact Factors}{9}{}%
\contentsline {subsection}{\numberline {9.7}Economic and Financial Ethical Considerations}{9}{}%
\contentsline {subsection}{\numberline {9.8}Economic and Financial Impact Factors}{9}{}%
\contentsline {subsection}{\numberline {9.9}Generative AI Disclaimer}{9}{}%
\contentsline {section}{\numberline {10}Key Takeaways}{9}{}%
\contentsline {section}{\numberline {11}Future Work}{10}{}%
\contentsline {section}{\numberline {12}Acknowledgements}{10}{}%
\contentsline {section}{\numberline {A}Additional Views of MNIST Results}{12}{}%
\contentsline {section}{\numberline {B}Additional Views of CIFAR-10 Results}{13}{}%

9
report/references.bib
View File

@ -84,3 +84,12 @@
year={1976},
publisher={Springer}
}
@book{goodfellow2016deep,
title={Deep Learning},
author={Ian Goodfellow and Yoshua Bengio and Aaron Courville},
publisher={MIT Press},
note={\url{http://www.deeplearningbook.org}},
year={2016}
}

BIN
results/CIFAR-10_FGSM_line_strength1.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 106 KiB

BIN
results/CIFAR-10_FGSM_line_strength4.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 97 KiB

BIN
results/MNIST_FGSM_line_strength1.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 130 KiB

BIN
results/MNIST_FGSM_line_strength4.png Normal file
View File

Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 134 KiB

1
src/display_results.py → src/display_results_3d.py
View File

@ -74,6 +74,7 @@ def main():
# ax.bar3d(x, y, 0, dstrength, depsilon, tops, color=greyscale, zsort="average", shade=True, label=filter)
# has_random_guessing_threshold = True
fig, ax = plt.subplots(subplot_kw={"projection": "3d"}, figsize=(9,10))
ax.view_init(45, 135, 0)
ax.set_proj_type('ortho')

44
src/display_results_line.py Normal file
View File

@ -0,0 +1,44 @@
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.lines import Line2D
import json
data = {}
with open("../results/cifar10_fgsm.json", "r") as infile:
data = json.load(infile)
attack = data["attack"]
epsilons = data["epsilons"]
filters = data["filters"]
dataset = data["dataset"]
strength_count = len(filters[list(filters.keys())[0]][0])
random_guess_threshold = np.ndarray((len(epsilons), strength_count))
random_guess_threshold.fill(0.1)
filters["Random Guess Threshold"] = random_guess_threshold
strengths = np.arange(strength_count)
epsilons = np.array(epsilons)
colors = ['blue', 'orange', 'red', 'purple', 'green', 'yellow', 'brown', 'black']
z = np.zeros_like(strengths)
best_performance = np.zeros_like(strengths)
strengths, epsilons = strengths.ravel(), epsilons.ravel()
strength = 4
plt.subplots(figsize=(16,9))
for i, filter in enumerate(filters):
performances = np.array([filt_eps[strength] for filt_eps in filters[filter]])
plt.plot(epsilons, performances, color=colors[i], label=filter.replace('_', ' ').title())
plt.legend(loc="upper right")
plt.title(f"Filter Performances on {dataset} at Filter Strength {strength}", fontsize=32)
plt.ylabel(f"{dataset} Classification Accuracy", fontsize=18)
plt.xlabel(f"{attack} Attack Strength ($\\epsilon$)", fontsize=18)
plt.savefig(f"../results/{dataset}_{attack}_line_strength{strength}.png")
plt.show()

27
src/tabulate_results.py Normal file
View File

@ -0,0 +1,27 @@
import json
import numpy as np
data = {}
with open("../results/mnist_fgsm_reformatted.json", "r") as infile:
data = json.load(infile)
attack = data["attack"]
epsilons = data["epsilons"]
filters = data["filters"]
dataset = data["dataset"]
strength_count = len(filters[list(filters.keys())[0]][0])
column_headers = ["$\\epsilon$"]
column_headers.extend([" & " + filter.replace('_', ' ').title() for filter in filters])
header_row = ""
for header in column_headers:
header_row += header
header_row += "\\"
print(header_row)
for i, epsilon in enumerate(epsilons):
row = str(epsilon)
for filter in filters:
row += filters[filter][i]filters[filter][