A separability-based approach to quantifying generalization: which layer is best?

arXiv:2405.01524v1 Announce Type: new
Abstract: Generalization to unseen data remains poorly understood for deep learning classification and foundation models. How can one assess the ability of networks to adapt to new or extended versions of their input space in the spirit of few-shot learning, out-of-distribution generalization, and domain adaptation? Which layers of a network are likely to generalize best? We provide a new method for evaluating the capacity of networks to represent a sampled domain, regardless of whether the network …

abstract adapt arxiv classification cs.ai cs.cv cs.lg data deep learning distribution domain domain adaptation few-shot few-shot learning foundation layer networks space type versions