Interpreting Black Box Predictions using Fisher Kernels
This work addresses the need for interpretability in machine learning models, particularly for practitioners dealing with mislabeled or noisy data, though it is incremental by building on existing methods like influence functions.
The paper tackles the problem of interpreting black box predictions by identifying which training examples are most responsible for given predictions, using Fisher kernels and Sequential Bayesian Quadrature to handle any subset of test predictions efficiently. It provides theoretical convergence bounds, recovers influence functions as a special case, and demonstrates applications in data cleaning, fixing mislabels, and summarization.
Research in both machine learning and psychology suggests that salient examples can help humans to interpret learning models. To this end, we take a novel look at black box interpretation of test predictions in terms of training examples. Our goal is to ask `which training examples are most responsible for a given set of predictions'? To answer this question, we make use of Fisher kernels as the defining feature embedding of each data point, combined with Sequential Bayesian Quadrature (SBQ) for efficient selection of examples. In contrast to prior work, our method is able to seamlessly handle any sized subset of test predictions in a principled way. We theoretically analyze our approach, providing novel convergence bounds for SBQ over discrete candidate atoms. Our approach recovers the application of influence functions for interpretability as a special case yielding novel insights from this connection. We also present applications of the proposed approach to three use cases: cleaning training data, fixing mislabeled examples and data summarization.