Prospector Heads: Generalized Feature Attribution for Large Models & Data

arXiv:2402.11729v1 Announce Type: new
Abstract: Feature attribution, the ability to localize regions of the input data that are relevant for classification, is an important capability for machine learning models in scientific and biomedical domains. Current methods for feature attribution, which rely on "explaining" the predictions of end-to-end classifiers, suffer from imprecise feature localization and are inadequate for use with small sample sizes and high-dimensional datasets due to computational challenges. We introduce prospector heads, an efficient and interpretable alternative to explanation-based …

abstract arxiv attribution biomedical capability classification classifiers cs.ai cs.lg current data domains feature generalized large models machine machine learning machine learning models predictions q-bio.qm type