Discriminative k-shot learning using probabilistic models

TL;DR

This paper presents a probabilistic framework for k-shot image classification that leverages feature and class information, achieving state-of-the-art results and well-calibrated uncertainty estimates.

Contribution

It introduces a probabilistic model for the final layer weights that enhances k-shot learning by combining representational and concept transfer, with improved accuracy and flexibility.

Findings

Achieves state-of-the-art performance on standard k-shot datasets.

Models uncertainty accurately, resulting in well-calibrated classifiers.

Flexible and easily extensible approach compared to recent methods.

Abstract

This paper introduces a probabilistic framework for k-shot image classification. The goal is to generalise from an initial large-scale classification task to a separate task comprising new classes and small numbers of examples. The new approach not only leverages the feature-based representation learned by a neural network from the initial task (representational transfer), but also information about the classes (concept transfer). The concept information is encapsulated in a probabilistic model for the final layer weights of the neural network which acts as a prior for probabilistic k-shot learning. We show that even a simple probabilistic model achieves state-of-the-art on a standard k-shot learning dataset by a large margin. Moreover, it is able to accurately model uncertainty, leading to well calibrated classifiers, and is easily extensible and flexible, unlike many recent approaches to k-shot learning.