Ashutosh Trivedi

Associate Professor of Computer Science
University of Colorado Boulder

I work on formal methods for reinforcement learning, trustworthy AI, and safety-critical software and cyber-physical systems.

My research combines verification, learning, and symbolic reasoning to make intelligent systems safer, fairer, and easier to explain.

Research

Foundations of Learning and Decision-Making
How can formal methods advance reinforcement learning and decision-making with rich objectives, structured environments, and differing time preferences?
Verification for Learning and Control
How can we provide formal guarantees for learned controllers operating under uncertainty?
Accountable AI and Software
How can we detect failures, explain decisions, and assess fairness in consequential software?

Regular Reinforcement Learning
A symbolic approach to reinforcement learning that represents sets of states with regular languages and transitions with rational transductions.
CAV, 2024 · CAV Distinguished Paper Award
Paper (PDF) Publisher BibTeX
Recursive Reinforcement Learning
Foundations for learning in recursive decision processes with an unbounded call structure.
NeurIPS, 2022
Paper (PDF) arXiv BibTeX
Closure Certificates
Transition-based certificates that extend safety reasoning to richer temporal properties of dynamical systems.
HSCC 2024
Paper (PDF) arXiv Publisher BibTeX
Stochastic Neural Simulation Relations for Control Transfer
Neural simulation relations for transferring controllers between stochastic systems with probabilistic guarantees.
NeuS, 2025 · DARPA Disruptive Idea Award
Paper (PDF) BibTeX
Fairness Testing through Extreme Value Theory
Extreme value theory for measuring and mitigating worst-case discrimination in machine-learning software.
ICSE, 2025
Paper (PDF) arXiv BibTeX

I work with students and postdoctoral researchers in the Programming Languages and Verification (CUPLV) group.

I teach theoretical computer science, reinforcement learning, and cyber-physical systems.