The new AI planning method, T-UCT, smartly estimates cost-reward trade-offs (Pareto curves) to find strategies that are much better at both getting rewards and staying within safety limits compared to existing approaches

Paper finds provably minimal counterfactual explanations

Secret identities in Dwarf Fortress (2017)

Adversarial Robust Deep Reinforcement Learning Requires Redefining Robustness

Deep Reinforcement Learning Policies Learn Shared Adversarial Features across MDPs

Knowledge Representation in Sanskrit and Artificial Intelligence

Nearly half of all political comments on Reddit are posted in non-political subreddits