Program Overview

Overview of CSCI 6962/4180 Trustworthy Machine Learning

The CSCI 6962/4180 Trustworthy Machine Learning course, offered in Fall 2025, delves into the critical aspects of trustworthiness in machine learning systems. Beyond traditional metrics of accuracy, the course explores alignment, fairness, robustness, privacy, and attack surfaces of machine learning systems. This seminar is designed for students with a basic understanding of machine learning, aiming to introduce them to fundamental questions and equip them with tools and methods to measure and ensure these aspects of trustworthiness.

Course Topics

The course covers five broad areas:

• Alignment of LLMs: Examining how large language models (LLMs) respect human norms, methods to measure and improve alignment, and the limitations thereof.
• Attack Models: Categorizing types of attacks on machine learning frameworks, understanding targeted aspects, goals, and adversary capabilities, and strategies for defense.
• Privacy and Confidentiality: Assessing the trustworthiness of machine learning frameworks with access to personal data, quantifying disclosure risk, and ensuring models do not reveal sensitive information.
• Robustness: Investigating the ability of algorithms to learn robust decision rules from noisy or adversarially crafted data and make correct predictions on challenging testing data.
• Algorithmic Fairness: Addressing bias in machine learning, building trustworthy algorithms that provide fair predictions for all groups, and quantifying fairness.

Course Logistics

Instructor and Lectures

The course is instructed by Alex Gittens, with lectures held on Mondays and Thursdays from 2 pm to 3:50 am ET in Ricketts 212.

Grading Criteria

The grading criteria for the course include:

• Seminar: 60%
• Project: 20%
• Attendance and Participation: 20%
  • Note: Attendance is mandatory.

Rubrics for each graded component are available. Letter grades are computed from the semester average, with specific lower-bound cutoffs for undergraduates (A: 90%, B: 80%, C: 70%, D: 60%) and graduates (A: 90%, B: 80%, C: 70%). These cutoffs may be adjusted at the instructor's discretion.

Course Materials and Project

Course materials, including papers, presentations, lecture notes, and discussions, are available. Each student must complete a final project, with graduate students undertaking a research project related to the course subject matter and undergraduate students having the option to complete either a research or a pedagogical project.

Supplementary Materials

For students needing a refresher on machine learning architectures and concepts, supplementary materials from a related course are available.