I am Thiago. Currently, I am a PostDoc researcher with the Department of Software Science (SWS) at Radboud University Nijmegen advised by Dr. Nils Jansen. Previously, I was a Ph.D. candidate in the Algorithmics Group at Delft University of Technology, advised by Dr. Matthijs Spaan. For more details checkout my biography or my cv .
Research Interests: The motivation for my research revolves around making AI techniques more reliable, to enable their deployment in real-world applications. I focus on developing AI algorithms for scenarios with constrained interactions with an unknown environment. I am currently interested in safe reinforcement learning, a research topic concerned with problems where a minimum performance must be guaranteed and catastrophic events must be avoided.
- Organization committee of the BeNeRL Workshop 2018.
- Local organizing committee of the 28th ICAPS.
- PC for NeurIPS22, ICML22, ICAPS22, AAAI21.
- Reviewer for JAAMAS, ICRA, AAAI and BRACIS.
Besides my professional activities, I like to run, play boardgames, listen to music and read.
- Our paper “Robust Anytime Learning of Markov Decision Processes” has been accepted at NeurIPS-22.
- I am serving as a PC member for NeurIPS 2022.
- Our paper “Safety-constrained reinforcement learning with a distributional safety critic” has been published at Machine Learning.
- Talk at the ADML meetup about Ensuring Safety for Reinforcement Learning.
- I am serving as a PC member for ICML 2022.
- Talk at the iVerif workshop on Safety Abstractions.
- I am serving as a PC member for the Planning and Learning track at ICAPS 2022.
- Invited talk at the Center for Artificial Intelligence.
- Our paper “AlwaysSafe: Reinforcement Learning Without Safety Constraint Violations During Training” has been accepted at AAMAS-21.
- Our paper “WCSAC: Worst-Case Soft Actor Critic for Safety-Constrained Reinforcement Learning” has been accepted at AAAI-21.
- I am serving as a PC member for AAAI-21.
- @ AAMAS-20 presenting the paper Safe Policy Improvement with an Estimated Baseline Policy.
- Released gym-factored, a collection of factored environments that are OpenAI Gym compliant.
- In Hilversum, presenting our work on reinforcement learning at the ICT.Open-19.
- I am co-organizing the Belgium Netherlands Workshop on Reinforcement Learning (BeNeRL-18).
- I am attending the 14th European Workshop on Reinforcement Learning (EWRL-18).
- I gave a contributed talk at the ICML-18 Workshop on Planning and Learning.
- I presented a poster at the Energy Event promoted by the PowerWeb Institute.
- I attended the 19th European Agent Systems Summer School.
MLSafety-constrained reinforcement learning with a distributional safety criticMachine Learning 2022
AlwaysSafe: Reinforcement Learning Without Safety Constraint Violations During TrainingIn Proceedings of the 20th International Conference on Autonomous Agents and MultiAgent Systems (AAMAS) 2021
Safe Policy Improvement with an Estimated Baseline PolicyIn Proceedings of the 19th International Conference on Autonomous Agents and MultiAgent Systems (AAMAS) 2020
Safe Policy Improvement with Baseline Bootstrapping in Factored EnvironmentIn Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence 2019