Daniel E. Hernández

Dustin Morrill, Thomas J. Walsh, Daniel Hernandez et al.

Modern reinforcement learning systems produce many high-quality policies throughout the learning process. However, to choose which policy to actually deploy in the real world, they must be tested under an intractable number of environmental conditions. We introduce RPOSST, an algorithm to select a small set of test cases from a larger pool based on a relatively small number of sample evaluations. RPOSST treats the test case selection problem as a two-player game and optimizes a solution with provable $k$-of-$N$ robustness, bounding the error relative to a test that used all the test cases in the pool. Empirical results demonstrate that RPOSST finds a small set of test cases that identify high quality policies in a toy one-shot game, poker datasets, and a high-fidelity racing simulator.

Daniel Hernandez, Hendrik Baier, Michael Kaisers

Finding a best response policy is a central objective in game theory and multi-agent learning, with modern population-based training approaches employing reinforcement learning algorithms as best-response oracles to improve play against candidate opponents (typically previously learnt policies). We propose Best Response Expert Iteration (BRExIt), which accelerates learning in games by incorporating opponent models into the state-of-the-art learning algorithm Expert Iteration (ExIt). BRExIt aims to (1) improve feature shaping in the apprentice, with a policy head predicting opponent policies as an auxiliary task, and (2) bias opponent moves in planning towards the given or learnt opponent model, to generate apprentice targets that better approximate a best response. In an empirical ablation on BRExIt's algorithmic variants against a set of fixed test agents, we provide statistical evidence that BRExIt learns better performing policies than ExIt.

Leonardo Trujillo, Jose Manuel Muñoz Contreras, Daniel E Hernandez et al.

Geometric Semantic Genetic Programming (GSGP) is a state-of-the-art machine learning method based on evolutionary computation. GSGP performs search operations directly at the level of program semantics, which can be done more efficiently then operating at the syntax level like most GP systems. Efficient implementations of GSGP in C++ exploit this fact, but not to its full potential. This paper presents GSGP-CUDA, the first CUDA implementation of GSGP and the most efficient, which exploits the intrinsic parallelism of GSGP using GPUs. Results show speedups greater than 1,000X relative to the state-of-the-art sequential implementation.