Unsupervised Concept Discovery Mitigates Spurious Correlations

arXiv:2402.13368v1 Announce Type: new
Abstract: Models prone to spurious correlations in training data often produce brittle predictions and introduce unintended biases. Addressing this challenge typically involves methods relying on prior knowledge and group annotation to remove spurious correlations, which may not be readily available in many applications. In this paper, we establish a novel connection between unsupervised object-centric learning and mitigation of spurious correlations. Instead of directly inferring sub-groups with varying correlations with labels, our approach focuses on discovering concepts: …

abstract annotation applications arxiv biases challenge concept correlations cs.cv cs.lg data discovery knowledge novel paper predictions prior training training data type unsupervised