Alignment with human representations supports robust few-shot learning

Date:

NeurIPS 2023 Spotlight presentation on the U-shaped relationship between human-model representational alignment and downstream task performance, with Tom Griffiths. Paper.