Интерпретируемое машинное обучение (2022): различия между версиями

Материал из SecSem Wiki
Перейти к навигации Перейти к поиску
(Лекции)
(Задания)
 
(не показано 9 промежуточных версий этого же участника)
Строка 16: Строка 16:
 
* '''31.03''' - PDP, ICE, ALE [https://www.dropbox.com/s/ld05h4fh9zmo5l7/5_PDP_ICE_ALE.pdf?dl=0 слайды] [https://www.dropbox.com/s/fzxasn5m2zq9wcc/5_PDP_ICE_ALE.mp4?dl=0 видео]
 
* '''31.03''' - PDP, ICE, ALE [https://www.dropbox.com/s/ld05h4fh9zmo5l7/5_PDP_ICE_ALE.pdf?dl=0 слайды] [https://www.dropbox.com/s/fzxasn5m2zq9wcc/5_PDP_ICE_ALE.mp4?dl=0 видео]
 
* ''' 7.04''' - Saliency maps, DeconvNet, Occlusion Sensitivity, Integrated gradients [https://www.dropbox.com/s/db2c6y4xaccuhpw/6_SDOI.ipynb?dl=0 тетрадка] [https://www.dropbox.com/s/fq6p6vjeswrs08p/6_SDOI.mp4?dl=0 видео]
 
* ''' 7.04''' - Saliency maps, DeconvNet, Occlusion Sensitivity, Integrated gradients [https://www.dropbox.com/s/db2c6y4xaccuhpw/6_SDOI.ipynb?dl=0 тетрадка] [https://www.dropbox.com/s/fq6p6vjeswrs08p/6_SDOI.mp4?dl=0 видео]
 +
* '''14.04''' - Backpropagation methods: Vanila backprop, Guided backprop, CAM (class activation methods), Grad-CAM+, Grad-CAM++ [https://www.dropbox.com/s/hyk5dollr4d7iik/7_Backprop_methods.ipynb?dl=0 тетрадка]
 +
* '''21.04''' - Visual Transformers [https://www.dropbox.com/s/gfyb3g4y6epikxt/8_ViT.ipynb?dl=0 тетрадка] [https://www.dropbox.com/s/x1kx7kzlzxq0otb/8_ViTs.mp4?dl=0 видео]
 +
* '''28.04''' - Deep Taylor decomposition, Layer-wise Relevance Propagation [https://www.dropbox.com/s/n2ho4s8nfk2fpn6/9_LRP.ipynb?dl=0 тетрадка] [https://www.dropbox.com/s/28fod9sjccvgpdm/9_LRP.mp4?dl=0 видео]
  
 
== Задания ==
 
== Задания ==
 
# Реализация LIME и SHAP [https://t.me/c/1554651430/20 Описание задания] Дедлайн: 26 марта
 
# Реализация LIME и SHAP [https://t.me/c/1554651430/20 Описание задания] Дедлайн: 26 марта
# Интерпретация нейронных сетей для картинок [https://t.me/c/1554651430/78 Распределение методов интерпретации по людям] Дедлайн: конец апреля
+
# Интерпретация нейронных сетей для картинок [https://t.me/c/1554651430/78 Распределение методов интерпретации по людям] Дедлайн: 20 мая
  
 
== Полезные ссылки и литература ==
 
== Полезные ссылки и литература ==
 +
=== Книги ===
 
# [https://cset.georgetown.edu/publication/machine-learning-and-cybersecurity/ Machine Learning and Cybersecurity (2021)]
 
# [https://cset.georgetown.edu/publication/machine-learning-and-cybersecurity/ Machine Learning and Cybersecurity (2021)]
 
# [https://www.dropbox.com/s/wd6g3jhi0qy3eyv/interpretable-machine-learning.pdf?dl=0 Molnar "Interpretable Machine Learning" (2020)]
 
# [https://www.dropbox.com/s/wd6g3jhi0qy3eyv/interpretable-machine-learning.pdf?dl=0 Molnar "Interpretable Machine Learning" (2020)]
 
# [https://github.com/PacktPublishing/Interpretable-Machine-Learning-with-Python Masis "Interpretable Machine Learning with Python" (2021)]
 
# [https://github.com/PacktPublishing/Interpretable-Machine-Learning-with-Python Masis "Interpretable Machine Learning with Python" (2021)]
 +
# [https://www.dropbox.com/s/ive59hcn7783vpm/%D0%9B%D0%B5%D0%BA%D1%86%D0%B8%D0%B8_%D0%BF%D0%BE_%D1%84%D1%83%D0%BD%D0%BA%D1%86%D0%B8%D0%BE%D0%BD%D0%B0%D0%BB%D1%8C%D0%BD%D0%BE%D0%BC%D1%83_%D0%B0%D0%BD%D0%B0%D0%BB%D0%B8%D0%B7%D1%83_%D0%A5%D0%B5%D0%BB%D0%B5%D0%BC%D1%81%D0%BA%D0%B8%D0%B9_A_%D0%AF_z_lib_org.pdf?dl=0 Хелемский "Лекции по функциональному анализу"]
 +
=== Статьи ===
 
# [https://www.dropbox.com/s/h7sa4zi3n44qmkb/lime_paper.pdf?dl=0 Ribeiro "“Why Should I Trust You?” Explaining the Predictions of Any Classifier" (2016)]
 
# [https://www.dropbox.com/s/h7sa4zi3n44qmkb/lime_paper.pdf?dl=0 Ribeiro "“Why Should I Trust You?” Explaining the Predictions of Any Classifier" (2016)]
 +
# Friedman, Jerome H. “Greedy function approximation: A gradient boosting machine.” Annals of statistics (2001): 1189-1232
 +
# Apley, Daniel W. “Visualizing the effects of predictor variables in black box supervised learning models.” arXiv preprint arXiv:1612.08468 (2016)
 +
# [https://arxiv.org/abs/1706.07979 "General Activation Maximization, Activation Maximization in Codespace, Simple Taylor Decomposition, Deep Taylor Decomposition, LRP-ab"]
 +
# [https://distill.pub/2017/feature-visualization/ "DeepDream"]
 +
# [https://arxiv.org/pdf/1312.6034.pdf "Saliency Map, Vanilla Backpropagation"]
 +
# [https://arxiv.org/pdf/1311.2901.pdf "DeConvNet Full Input Reconstruction, DeConvNet Partial Input Reconstruction, Occlusion Sensitivity"]
 +
# [https://arxiv.org/pdf/1412.6806.pdf "Guided Backpropagation"]
 +
# [https://arxiv.org/pdf/1703.01365.pdf "Integrated Gradients"]
 +
# [https://arxiv.org/pdf/1706.03825.pdf "SmoothGrad"]
 +
# [http://iphome.hhi.de/samek/pdf/MonXAI19.pdf "Deep Taylor Decomposition, LRP-0, -epsilon, -gamma, LRP-ab"]
 +
# [https://arxiv.org/pdf/1704.02685.pdf "DeepLIFT"]
 +
# [http://cnnlocalization.csail.mit.edu/Zhou_Learning_Deep_Features_CVPR_2016_paper.pdf "Class Activation Map (CAM)"]
 +
# [https://arxiv.org/pdf/1610.02391.pdf "Gradient-Weighted Class Activation Map (Grad-CAM)"]
 +
=== Репозитории и либы ===
 
# https://github.com/marcotcr/lime
 
# https://github.com/marcotcr/lime
 
# https://shap.readthedocs.io
 
# https://shap.readthedocs.io
 
# https://scikit-learn.org/stable/modules/partial_dependence.html
 
# https://scikit-learn.org/stable/modules/partial_dependence.html
# Friedman, Jerome H. “Greedy function approximation: A gradient boosting machine.” Annals of statistics (2001): 1189-1232
 
# Apley, Daniel W. “Visualizing the effects of predictor variables in black box supervised learning models.” arXiv preprint arXiv:1612.08468 (2016)
 

Текущая версия на 12:03, 30 апреля 2022

Курс Интерпретируемое машинное обучение читается в весеннем семестре первого года обучения магистерской программы "Искусственный интеллект в кибербезопасности".

Курс рекомендуется студентам, интересующимся методами интерпретации для моделей машинного обучения.

Оценка за курс будет определяться по результатам выполнения домашних заданий заданий. Выполнение всех домашних заданий позволяют получить "зачет" автоматически.

Лекции

Задания

  1. Реализация LIME и SHAP Описание задания Дедлайн: 26 марта
  2. Интерпретация нейронных сетей для картинок Распределение методов интерпретации по людям Дедлайн: 20 мая

Полезные ссылки и литература

Книги

  1. Machine Learning and Cybersecurity (2021)
  2. Molnar "Interpretable Machine Learning" (2020)
  3. Masis "Interpretable Machine Learning with Python" (2021)
  4. Хелемский "Лекции по функциональному анализу"

Статьи

  1. Ribeiro "“Why Should I Trust You?” Explaining the Predictions of Any Classifier" (2016)
  2. Friedman, Jerome H. “Greedy function approximation: A gradient boosting machine.” Annals of statistics (2001): 1189-1232
  3. Apley, Daniel W. “Visualizing the effects of predictor variables in black box supervised learning models.” arXiv preprint arXiv:1612.08468 (2016)
  4. "General Activation Maximization, Activation Maximization in Codespace, Simple Taylor Decomposition, Deep Taylor Decomposition, LRP-ab"
  5. "DeepDream"
  6. "Saliency Map, Vanilla Backpropagation"
  7. "DeConvNet Full Input Reconstruction, DeConvNet Partial Input Reconstruction, Occlusion Sensitivity"
  8. "Guided Backpropagation"
  9. "Integrated Gradients"
  10. "SmoothGrad"
  11. "Deep Taylor Decomposition, LRP-0, -epsilon, -gamma, LRP-ab"
  12. "DeepLIFT"
  13. "Class Activation Map (CAM)"
  14. "Gradient-Weighted Class Activation Map (Grad-CAM)"

Репозитории и либы

  1. https://github.com/marcotcr/lime
  2. https://shap.readthedocs.io
  3. https://scikit-learn.org/stable/modules/partial_dependence.html