Trustworthy AI, Spring 2022

Course Overview

Course banner for 24-784 Machine Learning for Engineers

Instructor: Prof. Ding Zhao

Teaching Assistant: Wenhao Ding, Zuxin Liu, Jiayin Xia, Zijian Guo, Sameer Bharadwaj, Yihan He

Objectives: Machine learning is a fast developing field. This is a big opportunity for young generation but also a big challenge as one may get lost in the pyramid of technical news and papers. This course aims to achieve two goals:

Help students have a high-level understanding of the fast-developing TAIAT area, quickly grasp key concepts and skills, and get familiar with the latest technological tools in the early phase of their graduate program, so that they could focus on a direction to establish their expertise;
Develop full-cycle research capabilities including paper reviews, writing research proposals, milestone reports, academic papers, conference-style paper reviews, making presentations, and serving as an academic critic.

Lecture schedule for Spring 2022

Lecture 1 - Tuesday, Jan 18, 2022: Overview, autonomy framework, trustworthy autonomy [Slides]

Lecture 2 - Thursday, Jan 20, 2022: Deep learning basics, vision models [Slides]

Lecture 3 - Tuesday, Jan 25, 2022: Latent space visualization, explainability [Slides]

Lecture 4 - Thursday, Jan 27, 2022: Security attacks: poisoning, evasion, FGSM, robust physical attack [Slides]

Lecture 5 - Tuesday, Feb 1, 2022: Robustness-Adversarial and defensive ML: randomization, robust AI, certification [Slides]

Lecture 6 - Thursday, Feb 3, 2022: Model-free decision making: imitation learning, reinforcement learning, Q learning [Slides]

Lecture 7 - Tuesday, Feb 8, 2022: Model-free Deep RL: REINFORCE, Actor-Critic [Slides]

Lecture 8 - Thursday, Feb 10, 2022: Model-based Deep RL: MPC [Slides]

Lecture 9 - Tuesday, Feb 15, 2022: Adversarial AI

Lecture 10 - Thursday, Feb 17, 2022: Gaussian processes: GP [Slides]

Lecture 11 - Tuesday, Feb 22, 2022: Safety: CMDP, Lagrangian-based Method (TRPO-lag, PPO-lag), Constrained Optimization [Slides]

Lecture 12 - Thursday, Feb 24, 2022: RL for real world autonomy (and model based RL)

Lecture 13 - Tuesday, Mar 1, 2022: Safety: CMDP, Lagrangian-based Method (TRPO-lag, PPO-lag), Constrained Optimization [Slides]

Lecture 14 - Thursday, Mar 3, 2022: Safety: reachability, Control Lyapunov, barrier function [Slides]

Tuesday, Mar 8, 2022: No class

Thursday, Mar 10, 2022: No class

Lecture 15 - Tuesday, Mar 15, 2022: Certification: overview, digital twin simulation, safety critical scenario generation [Slides]

Lecture 16 - Thursday, Mar 17, 2022: Safe RL

Lecture 17 - Tuesday, Mar 22, 2022: Digital twin - data-driven: VAE, GAN, and Flow [Slides]

Lecture 18 - Thursday, Mar 24, 2022: Digital twin - adversarial: worst-case, IS, splitting [Slides]

Lecture 19 - Tuesday, Mar 29, 2022: Scenario generation, evaluation and certification

Lecture 20 - Thursday, Mar 31, 2022: Generalization: Working with real world robots, domain randomization, DDPG, SAC [Slides]

Lecture 21 - Tuesday, Apr 5, 2022: Generalization: Nonstationary environment: delay, RARL, meta learning, NP [Slides]

Thursday, Apr 7, 2022: No class

Lecture 22 - Tuesday, Apr 12, 2022: Milestone review

Lecture 23 - Thursday, Apr 14, 2022: Generalization: hierarchical AI, lifelong learning, DPGP [Slides]

Lecture 24 - Tuesday, Apr 19, 2022: Generalization

Lecture 25 - Thursday, Apr 21, 2022: Human centricity: Privacy, fairness [Slides]

Lecture 26 - Tuesday, Apr 26, 2022: Invited speaker: Rahul Mangharam

Lecture 27 - Thursday, Apr 28, 2022: Alumni session: presentation