uber-research/safemutations

This project offers tools to experiment with different ways of modifying deep and recurrent neural networks, particularly for tasks like classifying sequences or navigating mazes. You provide a neural network model and choose a 'mutation' method, then observe how these changes impact the model's performance. It's designed for machine learning researchers or practitioners who want to understand and test various network mutation strategies.

144 stars. No commits in the last 6 months.