Unbiased Max-Min Embedding Classification for Transductive Few-Shot Learning: Clustering and Classification Are All You Need

TL;DR

The paper addresses transductive few-shot learning under data scarcity by introducing UMMEC, a framework that unifies embedding and classification objectives. It mitigates hubness with a Decentralized Covariance Matrix, balances inter-class separation with intra-class clustering through adaptive local alignment and a global uniformity loss, and enhances classification via a Variational Sinkhorn OT-based classifier. The combination yields state-of-the-art results on Mini-ImageNet, Tiered-ImageNet, and CUB across 1- and 5-shot tasks on multiple backbones, supported by ablations and visualizations. This work advances TFSL by providing a robust, scalable approach that excels with minimal labeled data and unlabeled query support, with practical implications for low-label regimes in vision tasks.

Abstract

Convolutional neural networks and supervised learning have achieved remarkable success in various fields but are limited by the need for large annotated datasets. Few-shot learning (FSL) addresses this limitation by enabling models to generalize from only a few labeled examples. Transductive few-shot learning (TFSL) enhances FSL by leveraging both labeled and unlabeled data, though it faces challenges like the hubness problem. To overcome these limitations, we propose the Unbiased Max-Min Embedding Classification (UMMEC) Method, which addresses the key challenges in few-shot learning through three innovative contributions. First, we introduce a decentralized covariance matrix to mitigate the hubness problem, ensuring a more uniform distribution of embeddings. Second, our method combines local alignment and global uniformity through adaptive weighting and nonlinear transformation, balancing intra-class clustering with inter-class separation. Third, we employ a Variational Sinkhorn Few-Shot Classifier to optimize the distances between samples and class prototypes, enhancing classification accuracy and robustness. These combined innovations allow the UMMEC method to achieve superior performance with minimal labeled data. Our UMMEC method significantly improves classification performance with minimal labeled data, advancing the state-of-the-art in TFSL.

Figures (3)

• Figure 1: The UMMEC creates a feature space where each class forms a compact cluster, while the clusters of different classes are widely separated.
• Figure 2: The structure of UMMEC. The UMMEC framework utilizes a pre-trained backbone for feature extraction, applies Decentralized Covariance to achieve inter-class uniformity and intra-class alignment in embeddings, and employs a variational Sinkhorn approach based on optimal transport theory for classifier optimization.
• Figure 3: The results of the t-SNE experiment are shown in the figure. Figure (a) depicts the class distribution before embedding, while Figure (b) shows the class distribution after applying the L2Nwang2019simpleshot, CL2Nwang2019simpleshot, TCPRxu2022alleviating and our UMME methods. The L2N, CL2N, TCPR methods exhibit some degree of class confusion, with indistinct class boundaries. In contrast, our UMME method achieves clear inter-class separation and strong intra-class clustering, resulting in well-defined class boundaries.