BIRLA INSTITUTE OF TECHNOLOGY & SCIENCE, PILANI

WORK INTEGRATED LEARNING PROGRAMMES

Digital
Part A: Content Design

Course Title FAccT Machine Learning

Course No(s) AIML * ZG517

Credit Units 4

Credit Model 1 - 0.5 - 1.5

1 unit for class room hours, 0.5 unit for Tutorial, 1.5 units for
Student preparation.
1 unit = 32 hours
Content Authors Dr. Sugata Ghosal

Version 1.0

Date January 11th , 2023

Course Description

Machine Learning systems have demonstrated remarkable learning capabilities. A growing area in
machine learning focuses on improving the Fairness, Accountability, and Transparency (FAccT) of a
model in addition to its accuracy-based performance metrics. Although FAccT will be difficult to
achieve, emerging technical approaches in this topic show promise in making better FAccT ML
systems. In this course, we will study the rigorous computer science techniques necessary for FAccT
machine learning and dive into the technical underpinnings of topics including fairness, robustness,
interpretability, accountability, and privacy. These topics reflect state-of-the-art research in FAccT, are
socially important, and they have strong industrial interest due to government and other policy
regulation. This course will focus on the algorithmic and statistical aspects of FAccT ML. We will also
discuss several application areas where we can apply these techniques. This course requires students to
have mathematical and programming backgrounds in machine learning.

Topics
Biases and fairness, fair representation learning, Interpretability and Transparency, Example and Visualization
Based Methods for Interpretability, Interpreting deep neural networks, Fairness Through Input Manip ulation,
Fair NLP/Vision, Robustness and adversarial attacks/defence, ML auditing, privacy
Course Objectives

No

CO1 Introduce students to the concepts of bias and fairness and techniques for incorporating
these in ML

CO2 Introduce students to the concepts of interpretability and transparency and techniques for
incorporating these in ML

CO3 Introduce students to the concepts of robustness and techniques for robust ML

CO4 Introduce students to the concepts of privacy in ML

Text Book(s)

T1 Barocas, Solon, Moritz Hardt, and Arvind Narayanan. Fairness and Machine Learning,
2018.
T2 Molnar, Christoph. Interpretable machine learning, 2019.

Reference Book(s) & other resources

R1 Christopher M. Bishop, Pattern Recognition & Machine Learning, Springer, 2006

R2 https://www.borealisai.com/research-blogs/tutotial*

R3

Content Structure

1. Introduction
1.1 Review - traditional performance criteria for ML
1.2 Emerging requirements
1.2.1 Fairness
1.2.2 Accountability
1.2.3 Transparency
1.2.4 Privacy
1.2.5 Robustness
1.3 Motivating use cases
2. Fairness and Bias
2.1 Sources of Bias
2.2 Real world examples
2.2.1 school admissions, criminal justice, hiring, gender/occupation bias
2.3 Sensitive Features
2.4 Fairness through unawareness

3. Learning Fair Representations

3.1 Major Fairness criteria
3.1.1 Direct Solution Method
3.1.2 Demographic Parity
3.1.3 Equality of Odds/Opportunity
3.1.4 FICO Case Study
3.2 Prejudice Removing Regularizer
3.2.1 Prejudice index (PI)
3.2.2 Optimizing PI
3.3 Adult Income Case Study

4. Fairness thru input manipulation

4.1 Basic Data Manipulation Techniques
4.1.1 Reweighing
4.1.2 Universal Sampling
4.1.3 Preferential Sampling
4.2 Individual Fairness
4.3 Optimized Pre-processing
4.4 Learning to Defer

5. Fair Casual Reasoning

5.1 Causal Fairness and Inherent Bias
5.2 Counterfactual Fairness
5.2.1 Formal Methods
5.2.2 Case Study – Success in Law School
5.2.3 Case Study - Crime Rates and Arrest
5.3 Equalized Counterfactual Odds
5.4 Multiple Causal Worlds

6. Fairness in NLP, Computer Vision

6.1 Biases in NLP Models
6.2 Data Augmentation
6.3 Debiasing Word Embedding
6.4 Counterfactual Fairness
6.5 Counterfactual Face Attribution
6.6 Gender Equalized Image Captioning
6.7 Adversarial Removal of Gender Features
7. Interpretability and Transparency
7.1 ML Interpretability
7.2 Intrinsically Interpretable Models
7.2.1 Simple interpretable models
7.2.2 Intrinsically interpretable techniques for deep learning
7.3 Interpretability Concepts
7.3.1 Intrinsic and post hoc methods
7.3.2 model-specific and model-agnostic methods
7.3.3 Local and global interpretable methods
7.4 Interpretability and performance trade-offs Instance-based Learning

8. Feature interaction for interpretability

8.1 Feature Interaction
8.2 Layerwise Relevance Propagation
8.3 DeepLift
8.4 Shapley Additive Explanations (SHAP)
8.4.1 Coalitional Game and Shapley Values
8.4.2 Kernel SHAP
8.4.3 Deep SHAP
8.4.4 Tree SHAP
8.5 Equitable Value of Data

9. Example and Visualization Based Methods for Interpretability

9.1 Example Based Methods
9.2 Counterfactual Explanations
9.2.1 Targeted counterfactual examples
9.2.2 Untargeted counterfactual examples
9.3 Contrastive Examples
9.4 Concept Based Methods

10. Interpreting Deep Networks

10.1 Visualization Based Methods
10.2 Activation Visualization
10.2.1 Saliency Maps
10.2.2 GoogLeNet Activation Atlas
10.2.3 Interpretability via Activation Visualization
10.3 Gradient Based Feature Attribution ○
10.3.1 Integrated Gradient
10.3.2 Baselines for Integrated Gradient

11. Robustness and Adversarial Attacks / Defense

11.1 Adversarial Attack
11.2 White-box Evasion Attack
11.2.1 FGSM
11.2.2 C&W
11.2.3 Physical Attack
11.3 Transferability of Attack
 11.4 Black-box Evasion Attack
11.4.1 Jacobian-based Data Augmentation
11.5 Adversarial Defense
11.6 Defense Strategies
11.6.1 Adversarial Training
11.6.2 Input Transformations
11.6.3 Stochastic Gradients
11.6.4 Obfuscated Gradients and BPDA
11.7 Robust Optimization
11.8 Certified Defense

12. ML Auditing and Privacy

12.1 ML Auditing
12.1.1 Distill-and-Compare
12.2 Privacy in ML
12.2.1 Differential Privacy with Deep Learning
12.2.2 Model Inversion Attack and Differential Privacy
12.2.3 Local Differential Privacy
12.2.4 Federated Learning

Learning Outcomes:

No Learning Outcomes

LO1 A strong understanding of the foundations of Machine Learning algorithms

LO2 Able to solve Machine Learning problems using appropriate learning techniques

LO3 Evaluate machine learning solutions to problems

LO4 Identify appropriate tools to implement the solutions to machine learning problems

Part B: Learning Plan

Academic Term

Course Title Machine Learning

Course No ZG 565

Lead Instructor

Session No. Topic Title Study/HW Resource

 Reference

1 Introduction
Review - traditional performance criteria for ML; T1 – Ch1
Emerging requirements such as Fairness,
Accountability, Transparency, Privacy, Robustness,
Motivating use cases

2 Fairness and Bias

Sources of Bias, Real world examples such as school
admissions, criminal justice, hiring, gender/occupatio n
bias, Sensitive Features, Fairness through unawareness

3 Learning Fair Representations

Major Fairness criteria such as Demographic Parity,
Equality of Odds/Opportunity, FICO Case Study, ,
Prejudice Removing Regularizer, Adult Income Case
Study

4 Fairness thru input manipulation

Basic Data Manipulation Techniques, such as
Reweighing, Universal Sampling, Preferentia l
Sampling, Individual Fairness, Optimized Pre-
processing, Learning to Defer

5 Fair Casual Reasoning

Causal Fairness and Inherent Bias, Counterfactua l
Fairness, Case Study – Success in Law School, Crime
Rates and Arrest, Equalized Counterfactual Odds,
Multiple Causal Worlds

6 Fairness in NLP, Computer Vision

Biases in NLP Models. Data Augmentation, Debiasing
Word Embedding, Counterfactual Fairness,
Counterfactual Face Attribution, Gender Equalized
Image Captioning, Adversarial Removal of Gender
Features

7 Disentangled Fair Representation

Disentangled Representations, Flexibly Fair
Representation, Orthogonal Disentangled Fair
Representations, Measurements for Disentangled Fair
Representations

8 Review of Session 1 to 7 Books, Web references

and Slides
9 Interpretability and Transparency
ML Interpretability, Intrinsically Interpretable Models,
Interpretability Concepts such as Intrinsic and post hoc
methods, model-specific and model-agnostic methods,
Local and global interpretable methods, Interpretability
and performance trade-offs Instance-based Learning

10 Feature interaction for interpretability

Feature Interaction such as Layerwise Relevance
Propagation, DeepLift, Shapley Additive Explanatio ns
(SHAP) and variants

11 Example and Visualization Based Methods for

Interpretability
Example Based Methods, Counterfactual Explanations,
Contrastive Examples, Concept Based Methods

12 Interpreting Deep Networks

Visualization Based Methods, Activation Visualizatio n,
saliency maps, Interpretability via Activatio n
Visualization, Gradient Based Feature Attribution ○

13 Robustness and Adversarial Attacks / Defense

Adversarial Attack, White-box Evasion Attack,
Transferability of Attack, Black-box Evasion Attack

14 Robustness and Adversarial Attacks / Defense

(contd.)
Adversarial Defense, Defense Strategies such as
Adversarial Training, Input Transformations,
Stochastic Gradients, Obfuscated Gradients
ML Auditing and Privacy
ML Auditing, Distill-and-Compare, Privacy in ML,

15 ML Auditing and Privacy (contd)

Differential Privacy with Deep Learning, Model
Inversion Attack and Differential Privacy, Local
Differential Privacy, Federated Learning

16 Review of session 9 to 15 Books, Web references

and Slides
Detailed Plan for Lab work

Lab Lab Objective Lab Sheet Session

No. Access URL Reference
1 Detecting bias in ML model using different fairness metrics
2 Bias mitigation in ML model using Disparate impact
remover
3 Improving the Fairness of a Classifier using Prejudice
Removal Regularizer
4 Implementing LIME and SHAP algorithms to interpret
different machine learning classifiers
5 Visualizing and interpreting a CNN based Image classifier
Implementing Adversarial attacks on CNN based Image
classifier

Evaluation Scheme :
Legend: EC = Evaluation Component; AN = After Noon Session; FN = Fore Noon Session
No Name Type Duration Weight Day, Date, Session, Time
EC-1 Quiz – Best 2 out of 3 Online ~1 hour 10% TBA
Assignment-I Take ~2-3 10% TBA
Home weeks
Assignment-II Take ~2-3 10% TBA
Home weeks
EC-2 Mid-Semester Test Closed TBA 30% TBA
Book
EC-3 Comprehensive Exam Open Book TBA 40% TBA

8310758441

Note:
Syllabus for Mid-Semester Test (Open Book): Topics in Session Nos. 1 to 8
Syllabus for Comprehensive Exam (Open Book): All topics (Session Nos. 1 to 16)
Important links and information:
Elearn portal: https://elearn.bits-pilani.ac.in or Canvas
Students are expected to visit the Elearn portal on a regular basis and stay up to date with the latest
announcements and deadlines.
Contact sessions: Students should attend the online lectures as per the schedule provided on the Elearn
portal.
Evaluation Guidelines:
1. EC-1 consists of either two Assignments or three Quizzes. Students will attempt them through
the course pages on the Elearn portal. Announcements will be made on the portal, in a timely
manner.
 2. For Closed Book tests: No books or reference material of any kind will be permitted.
3. For Open Book exams: Use of books and any printed / written reference material (filed or
bound) is permitted. However, loose sheets of paper will not be allowed. Use of calculators
is permitted in all exams. Laptops/Mobiles of any kind are not allowed. Exchange of any
material is not allowed.
4. If a student is unable to appear for the Regular Test/Exam due to genuine exigencies, the
student should follow the procedure to apply for the Make-Up Test/Exam which will be made
available on the Elearn portal. The Make-Up Test/Exam will be conducted only at selected
exam centres on the dates to be announced later.

It shall be the responsibility of the individual student to be regular in maintaining the self study
schedule as given in the course handout, attend the online lectures, and take all the prescribed
evaluation components such as Assignment/Quiz, Mid-Semester Test and Comprehensive Exam
according to the evaluation scheme provided in the handout.