Diagnosing and Remedying Shot Sensitivity with Cosine Few-Shot Learners
This addresses a practical limitation in few-shot learning for computer vision, making models more reliable in realistic settings where shot numbers are unknown, though it is incremental as it builds on existing cosine distance approaches.
The paper tackles the problem of shot sensitivity in few-shot image recognition, where performance degrades when the number of training and test examples mismatch, and finds that using cosine distance-based methods consistently improves robustness to shot variation, achieving competitive accuracy and outperforming prior state-of-the-art, with notable gains in very-low-shot regimes.
Few-shot recognition involves training an image classifier to distinguish novel concepts at test time using few examples (shot). Existing approaches generally assume that the shot number at test time is known in advance. This is not realistic, and the performance of a popular and foundational method has been shown to suffer when train and test shots do not match. We conduct a systematic empirical study of this phenomenon. In line with prior work, we find that shot sensitivity is broadly present across metric-based few-shot learners, but in contrast to prior work, larger neural architectures provide a degree of built-in robustness to varying test shot. More importantly, a simple, previously known but greatly overlooked class of approaches based on cosine distance consistently and greatly improves robustness to shot variation, by removing sensitivity to sample noise. We derive cosine alternatives to popular and recent few-shot classifiers, broadening their applicability to realistic settings. These cosine models consistently improve shot-robustness, outperform prior shot-robust state of the art, and provide competitive accuracy on a range of benchmarks and architectures, including notable gains in the very-low-shot regime.