Discriminative k-shot learning using probabilistic models

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 form of the classes (concept transfer). The concept information is encapsulated in a probabilistic model for the final layer weights of the neural network which then acts as a prior when probabilistic k-shot learning is performed. Surprisingly, simple probabilistic models and inference schemes outperform many existing k-shot learning approaches and compare favourably with the state-of-the-art method in terms of error-rate. The new probabilistic methods are also able to accurately model uncertainty, leading to well calibrated classifiers, and they are easily extensible and flexible, unlike many recent approaches to k-shot learning.