From ed40b45e3798a7823e0923c2843d85821de98d5a Mon Sep 17 00:00:00 2001 From: Piyush Mohite Date: Tue, 24 Aug 2021 17:53:35 +0530 Subject: [PATCH 1/5] create landing page with TOC --- .gitignore | 1 + index.md | 198 ++++++++++++++++++++++++++++++++++++++++++++++++++++- 2 files changed, 198 insertions(+), 1 deletion(-) diff --git a/.gitignore b/.gitignore index 918de83..8c621f0 100644 --- a/.gitignore +++ b/.gitignore @@ -4,3 +4,4 @@ _site .jekyll-metadata vendor Gemfile.lock +.vscode \ No newline at end of file diff --git a/index.md b/index.md index bc88dc0..dd1e1e8 100644 --- a/index.md +++ b/index.md @@ -3,4 +3,200 @@ layout: default title: Adversarial Deep Learning Website nav_exclude: true --- -This is the ADL homepage \ No newline at end of file +# Table of Contents + +## Unit-1: Introduction + +### 1. Introduction + +1. Motivation +2. Overview +3. Outline of following chapters + +### 2. Brief Introduction to Deep Learning Concepts + +1. Types of learning paradigms: + - Supervised learning + - Semi-supervised learning + - Unsupervised learning + +2. Artificial Neural Network +3. Convolutions and CNN +4. Recurrent Neural Networks +5. Encoder-Decoder +6. Auto-Encoders +7. Domain adaptation +8. Transformer + +### 3. Background and Notation on Adversarial Deep Learning + +1. History +2. Why do adversarial examples occur + - High dimensionality + - Insufficient data + - Linearity of distribution + - Robust and Non-Robust features + - Overparameterization +3. Basic Notation +4. Taxonomy +5. Threat Model + - Adversary’s goal + - Adversary’s knowledge + - White Box + - Black Box + - Grey Box + - Victim models +6. Security Evaluation + +### 4. Common Algorithms for Adversarial Testing of Deep Learning + +1. Generation of Adversarial Examples + - Box-Constrained L-BFGS + - Fast Gradient Sign Method (FGSM) + - Basic Iterative Method (BIM) +2. Evasion Black-Box Attacks + - Houdini + - Substitute Mode + +## Unit-2: Adversarial Learning in Computer Vision + +### 5. Attacking an Image Classifier + +1. Evasion White-Box Attacks + - Carlini and Wagner Attacks (C&W) + - Deep Fool + - Universal Attack + - Ground Truth Attack +2. Evasion Black-Box Attacks + - One-Pixel Attack + - Upset and Angri + - Zeroth Order Optimization (ZOO) based Attack + - Query Efficient Black Box Attack + - Adversarial Transformation Networks (ATNs) + - Generative Adversarial Networks (GANs) +3. Poisoning Attacks +4. Attacks in the Real World + - Cell-phone Camera Attack + - Road Sign Attack + - Cyberspace Attack + - 3D Adversarial Object + - Robotic Vision + +### 6. Defending Against Adversarial Attacks + +1. Recovering the True Labels of Adversarial Examples + - Robust Optimization + - Network Regularization + - Adversarial Training + - Provable Defenses + - Gradient Masking/Obfuscation + - Defensive Distillation + - Shattered Gradients + - Randomized Gradients + - Exploding & Vanishing Gradients +2. Detecting and Rejecting Adversarial Example + - Training-based Detection + - Criteria-based Detection + - Feature Squeezing + - Artifacts + - MagNet + +## Unit-3: Adversarial Learning in Natural Language Processing + +### Defense on NLP has just started last year so current existing papers are less than 10 + +1. Character-level + - Typo correction + - Robust encoding + - Learning to Discriminate Perturbations +2. Word-level + +### 8. Attacking + +### 9. Defense + +## Unit-4: Exploring Further + +### 10. Adversarial Attacks on Security + +### 11. Adversarial Attacks on Interpretability + +### 12. Adversarial Attacks on Privacy + +### 13. Fairness and Adversarial Learning + +1. The Concept of Fairness + - Social Perspective + - Probabilistic Perspective +2. Fairness by Adversarial Learning + +### 14. Adversarial Attacks against Bayesian Learning + +### 15. Adversarial Attacks against Reinforcement Learning + +### 16. Adversarial Attacks against Quantum + +### 17. Adversarial Attacks against Graph Learning + +### 18. Adversarial Training for Improving Model Generalization + +1. Domain Adaptation +2. Supervised and Semi-Supervised Classification +3. Regularization through Adversarial Learning +4. Why does adversarial training work? + +### 19. Generative Adversarial Networks + +### 20. A lawyer’s perspective on Adversaries and Adversarial Examples + +--- + +## Week 3 — Privacy Attacks + +- Stealing Machine Learning Models via Prediction APIs +- Model Reconstruction from Model Explanations +- Membership Inference Attacks Against Machine Learning Model + +## Week 4 - Poisoning Attacks + +- Poisoning Attacks against Support Vector Machines +- Poison Frogs! Targeted Clean-Label Poisoning Attacks on Neural Networks +- Stronger Data Poisoning Attacks Break Data Sanitization Defenses +- Transferable Clean-Label Poisoning Attacks on Deep Neural Nets + +## Week 5 - Evasion Attacks (Adversarial Examples) + +- Explaining and Harnessing Adversarial Examples +- Towards Evaluating the Robustness of Neural Networks +- Why Do Adversarial Attacks Transfer? Explaining Transferability of Evasion and Poisoning Attacks + +## Week 6 - Defense against Poisoning Attacks + +- Certified Defenses for Data Poisoning Attacks +- Co-teaching: Robust Training of Deep Neural Networks with Extremely Noisy Labels +- Robust Logistic Regression and Classification + +## Week 7 - Advanced Adversarial Attacks + +- Understanding Black-box Predictions via Influence Functions +- Machine Learning with Adversaries: Byzantine Tolerant Gradient Descent +- Comprehensive Privacy Analysis of Deep Learning: Passive and Active White-box Inference +Attacks against Centralized and Federated Learning + +## Week 8 - Privacy Defenses + +- Machine Learning with Membership Privacy using Adversarial Regularization +- Privacy-preserving Prediction +- Deep Learning with Differential Privacy + +## Week 9 - Defenses against Adversarial Examples + +- Towards Deep Learning Models Resistant to Adversarial Attacks +- Certified Defenses against Adversarial Examples +- An abstract domain for certifying neural networks + +## Week 10 - Advanced topics on Adversarial Examples + +- Adversarially Robust Generalization Requires More Data +- Adversarial Examples Are Not Bugs, They Are Features +- Theoretically Principled Trade-off between Robustness and Accuracy \ No newline at end of file From 6a8101742c905d16925ad0f20888f1cca4348b2d Mon Sep 17 00:00:00 2001 From: Piyush Mohite Date: Tue, 24 Aug 2021 21:11:00 +0530 Subject: [PATCH 2/5] Add Color Scheme toggle - Since the color scheme does not retain it's state on refresh, the color-toggler has been added to the header --- _config.yml | 1 + _includes/head.html | 39 ++++++++ _layouts/default.html | 216 ++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 256 insertions(+) create mode 100644 _includes/head.html create mode 100644 _layouts/default.html diff --git a/_config.yml b/_config.yml index b7488b1..085c8a2 100644 --- a/_config.yml +++ b/_config.yml @@ -33,6 +33,7 @@ theme: just-the-docs plugins: - jekyll-feed +color_scheme: dark # Exclude from processing. # The following items will not be processed, by default. # Any item listed under the `exclude:` key here will be automatically added to diff --git a/_includes/head.html b/_includes/head.html new file mode 100644 index 0000000..eaa5320 --- /dev/null +++ b/_includes/head.html @@ -0,0 +1,39 @@ + + + + + {% unless site.plugins contains "jekyll-seo-tag" %} + {{ page.title }} - {{ site.title }} + + {% if page.description %} + + {% endif %} + {% endunless %} + + + + + + {% if site.ga_tracking != nil %} + + + + {% endif %} + + {% if site.search_enabled != false %} + + {% endif %} + + + + + {% seo %} + + + \ No newline at end of file diff --git a/_layouts/default.html b/_layouts/default.html new file mode 100644 index 0000000..6adbf7e --- /dev/null +++ b/_layouts/default.html @@ -0,0 +1,216 @@ +--- +layout: table_wrappers +--- + + + + +{% include head.html %} + + + + Link + + + + + + Search + + + + + + Menu + + + + + + Expand + + + + + + Document + + + + + + +
+
+ {% include title.html %} + + + +
+ +
+ This site uses Just the Docs, a documentation theme for Jekyll. +
+
+
+
+ {% if site.search_enabled != false %} +
+
+ + +
+
+
+ {% endif %} + + {% include header_custom.html %} + {% if site.aux_links %} + + {% endif %} +
+
+ {% unless page.url == "/" %} + {% if page.parent %} + {%- for node in pages_list -%} + {%- if node.parent == nil -%} + {%- if page.parent == node.title or page.grand_parent == node.title -%} + {%- assign first_level_url = node.url | absolute_url -%} + {%- endif -%} + {%- if node.has_children -%} + {%- assign children_list = pages_list | where: "parent", node.title -%} + {%- for child in children_list -%} + {%- if page.url == child.url or page.parent == child.title -%} + {%- assign second_level_url = child.url | absolute_url -%} + {%- endif -%} + {%- endfor -%} + {%- endif -%} + {%- endif -%} + {%- endfor -%} + + {% endif %} + {% endunless %} +
+ {% if site.heading_anchors != false %} + {% include vendor/anchor_headings.html html=content beforeHeading="true" anchorBody="" anchorClass="anchor-heading" anchorAttrs="aria-labelledby=\"%html_id%\"" %} + {% else %} + {{ content }} + {% endif %} + + {% if page.has_children == true and page.has_toc != false %} +
+

Table of contents

+
    + {%- assign children_list = pages_list | where: "parent", page.title | where: "grand_parent", page.parent -%} + {% for child in children_list %} +
  • + {{ child.title }}{% if child.summary %} - {{ child.summary }}{% endif %} +
    • + {% endfor %} +
+ {% endif %} + + {% capture footer_custom %} + {%- include footer_custom.html -%} + {% endcapture %} + {% if footer_custom != "" or site.last_edit_timestamp or site.gh_edit_link %} +
+
+ {% if site.back_to_top %} +

{{ site.back_to_top_text }}

+ {% endif %} + + {{ footer_custom }} + + {% if site.last_edit_timestamp or site.gh_edit_link %} +
+ {% if site.last_edit_timestamp and site.last_edit_time_format and page.last_modified_date %} +

+ Page last modified: {{ page.last_modified_date | date: site.last_edit_time_format }}. +

+ {% endif %} + {% if + site.gh_edit_link and + site.gh_edit_link_text and + site.gh_edit_repository and + site.gh_edit_branch and + site.gh_edit_view_mode + %} +

+ {{ site.gh_edit_link_text }} +

+ {% endif %} +
+ {% endif %} +
+ {% endif %} + +
+
+ + {% if site.search_enabled != false %} + {% if site.search.button %} + + + + {% endif %} +
+ {% endif %} +
+ + + \ No newline at end of file From acdbc64b0965451166d8aebd50bf66c925e46136 Mon Sep 17 00:00:00 2001 From: Piyush Mohite Date: Sat, 28 Aug 2021 18:44:47 +0530 Subject: [PATCH 3/5] Modify the dark-mode toggle function --- _layouts/default.html | 12 ++++++++---- 1 file changed, 8 insertions(+), 4 deletions(-) diff --git a/_layouts/default.html b/_layouts/default.html index 6adbf7e..7565631 100644 --- a/_layouts/default.html +++ b/_layouts/default.html @@ -202,12 +202,16 @@

Table of contents