ML / AI · KSE 2024, Published
Adversarial Robustness via Entropy Based Feature Selection in RL
An entropy-based feature selection framework for reinforcement learning agents that achieved 94 and 95 percent accuracy on Lunar Lander and Bipedal Walker under adversarial perturbations, outperforming KL Divergence and Joint Entropy baselines across Gym environments.
94% acc
Lunar Lander
95% acc
Bipedal Walker
KL, Joint-H
Baselines beaten
KSE 2024
Venue
Problem
Reinforcement learning policies degrade unpredictably under observation noise, and standard robustness methods such as KL divergence and joint entropy provide limited guidance on which features to retain.
Approach
I designed an entropy-based feature selection framework, applied adversarial perturbations across multiple Gym environments at varying intensities, and benchmarked the method against KL Divergence and Joint Entropy baselines under matched configurations.
Results
The framework achieved 94 and 95 percent accuracy on Lunar Lander and Bipedal Walker under perturbation, outperforming both baselines. The work was published at KSE 2024 as 'Imposter Injection: Learning to Select Features in Reinforcement Learning.'
Stack