Task Discovery: Finding the Tasks that Neural Networks Generalize on

TL;DR

This paper introduces a framework to discover tasks that neural networks naturally generalize on by optimizing an agreement score, revealing insights into model biases and enabling new ways to evaluate model robustness.

Contribution

It proposes a novel task discovery method that identifies tasks aligned with neural network biases, offering a new perspective on model generalization and interpretability.

Findings

Many tasks can be derived from a single set of images where models generalize well.

Discovered tasks reflect the inductive biases and data patterns of neural networks.

Tasks can be used to create adversarial train-test splits that cause models to fail without changing data labels.

Abstract

When developing deep learning models, we usually decide what task we want to solve then search for a model that generalizes well on the task. An intriguing question would be: what if, instead of fixing the task and searching in the model space, we fix the model and search in the task space? Can we find tasks that the model generalizes on? How do they look, or do they indicate anything? These are the questions we address in this paper. We propose a task discovery framework that automatically finds examples of such tasks via optimizing a generalization-based quantity called agreement score. We demonstrate that one set of images can give rise to many tasks on which neural networks generalize well. These tasks are a reflection of the inductive biases of the learning framework and the statistical patterns present in the data, thus they can make a useful tool for analysing the neural networks and their biases. As an example, we show that the discovered tasks can be used to automatically create adversarial train-test splits which make a model fail at test time, without changing the pixels or labels, but by only selecting how the datapoints should be split between the train and test sets. We end with a discussion on human-interpretability of the discovered tasks.