Towards Class-Balanced Transductive Few-Shot Learning

Ran Tao, Hao Chen, Marios Savvides

Keywords: Few-shot classification, Transductive learning, Class-imbalanced Prediction

Abstract: In this work, we present an observation of severe class-imbalanced predictions in few-shot learning and propose solving it by acquiring a more balanced marginal probability through Transductive Fine-tuning with Margin-based uncertainty weighting and Class-balanced normalization (TF-MC). Margin-based uncertainty weighting compresses the utilization of wrong predictions with lower loss weights to stabilize predicted marginal distribution. Class-balanced normalization adjusts the predicted probability for testing data to pursue class-balanced fine-tuning without directly regularizing the marginal testing distribution. TF-MC effectively improves the class balance in predictions with state-of-the-art performance on in- / out-of-distribution evaluations of Meta-Dataset and surpasses previous transductive methods by a large margin.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Deep Learning and representational learning

TL;DR: We develop transductive fine-tuning with margin-based uncertainty weighting and class-balanced normalization to tackle the issue of class imbalanced predictions in few-shot learning.