One Network Fits All? Modular versus Monolithic Task Formulations in Neural Networks

TL;DR

This paper explores how neural networks can learn multiple diverse tasks simultaneously, analyzing the impact of task representation complexity on learning efficiency through theoretical insights and empirical experiments.

Contribution

It introduces a novel theoretical analysis of task encoding schemes and their learnability by neural networks, highlighting the effects of task complexity on sample requirements.

Findings

Neural networks can learn multiple tasks from combined data sets with well-structured task encodings.

Task complexity influences the sample size needed for successful learning.

Empirical results support the theoretical bounds across various task encoding methods.

Abstract

Can deep learning solve multiple tasks simultaneously, even when they are unrelated and very different? We investigate how the representations of the underlying tasks affect the ability of a single neural network to learn them jointly. We present theoretical and empirical findings that a single neural network is capable of simultaneously learning multiple tasks from a combined data set, for a variety of methods for representing tasks -- for example, when the distinct tasks are encoded by well-separated clusters or decision trees over certain task-code attributes. More concretely, we present a novel analysis that shows that families of simple programming-like constructs for the codes encoding the tasks are learnable by two-layer neural networks with standard training. We study more generally how the complexity of learning such combined tasks grows with the complexity of the task codes; we find that combining many tasks may incur a sample complexity penalty, even though the individual tasks are easy to learn. We provide empirical support for the usefulness of the learning bounds by training networks on clusters, decision trees, and SQL-style aggregation.