# Attentive Single-Tasking of Multiple Tasks

## TL;DR

This paper introduces a single-tasking multi-task learning approach where a shared network performs one task at a time with task-specific feature attention, reducing interference and parameter count while maintaining or improving accuracy.

## Contribution

It proposes a novel single-tasking multi-task training method with task attention and adversarial gradient alignment to reduce task interference and parameter usage.

## Key findings

- Significant parameter reduction while maintaining or improving performance.
- Effective task attention mechanism improves task-specific feature utilization.
- Smooth trade-off between computation cost and multi-task accuracy.

## Abstract

In this work we address task interference in universal networks by considering that a network is trained on multiple tasks, but performs one task at a time, an approach we refer to as "single-tasking multiple tasks". The network thus modifies its behaviour through task-dependent feature adaptation, or task attention. This gives the network the ability to accentuate the features that are adapted to a task, while shunning irrelevant ones. We further reduce task interference by forcing the task gradients to be statistically indistinguishable through adversarial training, ensuring that the common backbone architecture serving all tasks is not dominated by any of the task-specific gradients. Results in three multi-task dense labelling problems consistently show: (i) a large reduction in the number of parameters while preserving, or even improving performance and (ii) a smooth trade-off between computation and multi-task accuracy. We provide our system's code and pre-trained models at http://vision.ee.ethz.ch/~kmaninis/astmt/.