UNEM: UNrolled Generalized EM for Transductive Few-Shot Learning

TL;DR

This paper introduces UNEM, an unrolled EM algorithm that learns optimal hyper-parameters for transductive few-shot learning, significantly improving performance across diverse vision tasks.

Contribution

It proposes a novel unrolling paradigm for EM, enabling automatic hyper-parameter learning in few-shot learning, adaptable to various models and datasets.

Findings

Up to 10% accuracy improvement on vision benchmarks.

Up to 7.5% accuracy improvement on vision-language benchmarks.

Effective across multiple fine-grained classification tasks.

Abstract

Transductive few-shot learning has recently triggered wide attention in computer vision. Yet, current methods introduce key hyper-parameters, which control the prediction statistics of the test batches, such as the level of class balance, affecting performances significantly. Such hyper-parameters are empirically grid-searched over validation data, and their configurations may vary substantially with the target dataset and pre-training model, making such empirical searches both sub-optimal and computationally intractable. In this work, we advocate and introduce the unrolling paradigm, also referred to as "learning to optimize", in the context of few-shot learning, thereby learning efficiently and effectively a set of optimized hyper-parameters. Specifically, we unroll a generalization of the ubiquitous Expectation-Maximization (EM) optimizer into a neural network architecture, mapping each of its iterates to a layer and learning a set of key hyper-parameters over validation data. Our unrolling approach covers various statistical feature distributions and pre-training paradigms, including recent foundational vision-language models and standard vision-only classifiers. We report comprehensive experiments, which cover a breadth of fine-grained downstream image classification tasks, showing significant gains brought by the proposed unrolled EM algorithm over iterative variants. The achieved improvements reach up to 10% and 7.5% on vision-only and vision-language benchmarks, respectively.