Building an Explainable Reinforcement Learning Framework

Explainable Results Through Symbolic Policy Discovery

Symbolic genetic algorithms, action potentials, and equation trees

We’ve learned to train models that can beat world champions at a games like Chess and Go, with one major limitation: explainability. Many methods exist to create a black-box model that knows how to play a game or system better than any human, but creating a model with a human-readable closed-form strategy is another problem altogether.

The potential upsides of being better at this problem are …

data science genetic-algorithm machine learning programming reinforcement learning