Description
Training examples can be analyzed for influence on PyTorch model predictions. ML researchers use pytorch-influence-functions for interpretability, debugging datasets, data valuation, and research experiments. Results are approximations and depend on model and Hessian assumptions.