Publication Type

Conference Proceeding Article

Version

acceptedVersion

Publication Date

12-2020

Abstract

We uncover an ever-overlooked deficiency in the prevailing Few-Shot Learning (FSL) methods: the pre-trained knowledge is indeed a confounder that limits the performance. This finding is rooted from our causal assumption: a Structural Causal Model (SCM) for the causalities among the pre-trained knowledge, sample features, and labels. Thanks to it, we propose a novel FSL paradigm: Interventional Few-Shot Learning (IFSL). Specifically, we develop three effective IFSL algorithmic implementations based on the backdoor adjustment, which is essentially a causal intervention towards the SCM of many-shot learning: the upper-bound of FSL in a causal view. It is worth noting that the contribution of IFSL is orthogonal to existing fine-tuning and meta-learning based FSL methods, hence IFSL can improve all of them, achieving a new 1-/5-shot state-of-the-art on miniImageNet, tieredImageNet, and cross-domain CUB. Code is released at https://github.com/yue-zhongqi/ifsl.

Keywords

Causal intervention, Causal model, Cross-domain, Fine tuning, Interventional, Metalearning, Sample features, State of the art

Discipline

Artificial Intelligence and Robotics | Databases and Information Systems

Research Areas

Intelligent Systems and Optimization

Publication

Advances in Neural Information Processing Systems: Proceedings of the 34th Conference NeurIPS 2020, Vancouver, Canada, December 6-12, Virtual

First Page

1

Last Page

23

Publisher

MIT

City or Country

Cambridge, MA

Copyright Owner and License

Authors

Additional URL

https://proceedings.neurips.cc/paper/2020/file/1cc8a8ea51cd0adddf5dab504a285915-Paper.pdf

Download

Share

COinS