A Model of Inductive Bias Learning

TL;DR

This paper proposes a model for automatically learning inductive biases by leveraging related tasks in an environment, improving generalization across multiple tasks and reducing reliance on hand-crafted biases.

Contribution

It introduces a formal model for automatic bias learning from related tasks and provides theoretical bounds showing improved generalization in multi-task environments.

Findings

Learning multiple tasks enhances generalization performance.

A hypothesis space that performs well on many tasks also generalizes well to new tasks.

Explicit bounds demonstrate the benefits of multi-task learning for inductive bias.

Abstract

A major problem in machine learning is that of inductive bias: how to choose a learner's hypothesis space so that it is large enough to contain a solution to the problem being learnt, yet small enough to ensure reliable generalization from reasonably-sized training sets. Typically such bias is supplied by hand through the skill and insights of experts. In this paper a model for automatically learning bias is investigated. The central assumption of the model is that the learner is embedded within an environment of related learning tasks. Within such an environment the learner can sample from multiple tasks, and hence it can search for a hypothesis space that contains good solutions to many of the problems in the environment. Under certain restrictions on the set of all hypothesis spaces available to the learner, we show that a hypothesis space that performs well on a sufficiently large number of training tasks will also perform well when learning novel tasks in the same environment. Explicit bounds are also derived demonstrating that learning multiple tasks within an environment of related tasks can potentially give much better generalization than learning a single task.