Alignment with human representations supports robust few-shot learning
This work addresses the problem of improving few-shot learning and robustness in AI systems for researchers and practitioners, though it is incremental as it builds on existing alignment concepts.
The study investigated whether AI systems with human-like representations perform better on few-shot learning tasks, finding a U-shaped relationship between representational alignment and performance across 491 computer vision models, with highly-aligned models showing greater robustness to adversarial attacks and domain shifts.
Should we care whether AI systems have representations of the world that are similar to those of humans? We provide an information-theoretic analysis that suggests that there should be a U-shaped relationship between the degree of representational alignment with humans and performance on few-shot learning tasks. We confirm this prediction empirically, finding such a relationship in an analysis of the performance of 491 computer vision models. We also show that highly-aligned models are more robust to both natural adversarial attacks and domain shifts. Our results suggest that human-alignment is often a sufficient, but not necessary, condition for models to make effective use of limited data, be robust, and generalize well.