Ashutosh Trivedi
Associate Professor of Computer Science
University of Colorado Boulder
I develop mathematical foundations and tools for trustworthy AI systems that learn and make decisions under uncertainty while satisfying formal guarantees.
My research brings together reinforcement learning, formal verification, and symbolic reasoning, with applications to safety-critical control and accountable software.
Explore research · Prospective students
CV · Google Scholar · GitHub · CU Boulder · Academic genealogy · Email
Recent News
- Gave How do we know an algorithm is treating us fairly? in CU Boulder’s Beyond the Bytes faculty speaker series. Slides (PDF).
- On the Robustness of Fairness Practices appeared in ICSE 2026.
- An LLM Agentic Approach for Legal-Critical Software appeared in ICSE 2026.
- Received the 2026 Outstanding Faculty Research Advisor/Mentor Award from CU Boulder’s College of Engineering and Applied Science.
Selected Contributions
Learning with symbolic structure
Regular Reinforcement Learning
How can reinforcement learning work with sets of states, rather than individual states alone? We use regular languages to represent these sets and symbolic transformations to describe transitions, connecting reinforcement learning with ideas from model checking.
CAV, 2024CAV Distinguished Paper Award
Beyond safety guarantees
Closure Certificates
Showing that a system avoids unsafe states is only part of verification. Closure certificates reason about transitions across time, extending certificate-based verification to richer temporal requirements for dynamical systems.
HSCC 2024
Finding discrimination in the extremes
Fairness Testing through Extreme Value Theory
Average fairness measures can hide severe disadvantage for some individuals. We use extreme value theory to estimate worst-case discrimination and guide mitigation, making the tails of a model’s outcomes part of its fairness assessment.
ICSE, 2025
Students and Group
I work with students and postdoctoral researchers in the Programming Languages and Verification (CUPLV) group.
Teaching
I teach theoretical computer science, reinforcement learning, and cyber-physical systems.