Congratulations to Wen Chi Yang, Gavin Rens, Giuseppe Marra and Luc De Raedt on winning the IJCAI 2023 Distinguished Paper Award!
The IJCAI Distinguished Paper Awards recognise some of the best papers presented at the conference each year. The winners were selected from among more than 4500 papers by the associate programme committee chairs, the programme and general chairs, and the president of EurAI. This year, three articles were named as distinguished papers, and one of them is a paper presented by the four TAILOR scientists from KU Leuven.
The paper is a typical TAILOR style of work, positioning itself at the intersection of WP4 (Unifying Paradigms) and WP5 (Action)
Title: “Safe Reinforcement Learning via Probabilistic Logic Shields”
Abstract: Safe Reinforcement learning (Safe RL) aims at learning optimal policies while staying safe. A popular solution to Safe RL is shielding, which uses a logical safety specification to prevent an RL agent from taking unsafe actions. However, traditional shielding techniques are difficult to integrate with continuous, end-to-end deep RL methods. To this end, we introduce Probabilistic Logic Policy Gradient (PLPG). PLPG is a model-based Safe RL technique that uses probabilistic logic programming to model logical safety constraints as differentiable functions. Therefore, PLPG can be seamlessly applied to any policy gradient algorithm while still providing the same convergence guarantees. In our experiments, we show that PLPG learns safer and more rewarding policies compared to other state-of-the-art shielding techniques.