Steve Hsu

Steve in his office
graduating Masters student at the University of Pennsylvania

Contact Information

stevehsu at seas dot upenn dot edu
triangular dot pyramid at gmail dot com

Professional Interests



Research Assistant, University of Pennsylvania, September 2019–August 2022
Counterexample-guided Reinforcement Learning
  • Designed a compositional reinforcement learning algorithm for a complex task decomposed (by the user) into simpler subtasks and achieves a higher success rate with less training time
  • Sped up training and devised a natural framework for testing the policy
  • Created a benchmark suite of environments decomposed into subtasks for running the algorithm
Compositional Learning and Verification of Neural Network Controllers
  • Designed and trained a neural network policy that can safely drive in a hallway with an arbitrary sequence of turns
Internship, Amazon Connect (Amazon Web Services)
Implemented a variation of a feature called least-cost routing, which sends an outgoing call through the cheapest available carrier
Teaching Assistant, University of Pennsylvania, September 2020–May 2021
  • CIS 519 Applied Machine Learning, fall 2020 semester
  • CIS 520 Machine Learning, spring 2021 semester


Machine learning libraries


Programming Languages



Compositional Learning and Verification of Neural Network Controllers, Emsoft 2021 (PDF)
Radoslav Ivanov, Kishor Jothimurugan, Steve Hsu, Shaan Vaidya, Rajeev Alur, and Osbert Bastani
  • trained a car to drive in a hallway using only LiDAR scans without access to its true position
  • proved that the car doesn't crash by modeling the system as a hybrid automaton and using a hybrid automaton reachability checker