Dynamic Neural Network for Multi-Task Learning Searching across Diverse Network Topologies

TL;DR

This paper introduces a novel multi-task learning framework that searches for optimized, task-specific neural network structures with diverse topologies, achieving state-of-the-art results efficiently.

Contribution

It proposes a restricted DAG-based search space and a three-stage training process for discovering compact, task-adaptive networks with diverse topologies.

Findings

Achieves state-of-the-art performance on multiple MTL datasets.

Validates the effectiveness of the proposed search and training schemes.

Demonstrates the efficiency of the flow-based reduction algorithm.

Abstract

In this paper, we present a new MTL framework that searches for structures optimized for multiple tasks with diverse graph topologies and shares features among tasks. We design a restricted DAG-based central network with read-in/read-out layers to build topologically diverse task-adaptive structures while limiting search space and time. We search for a single optimized network that serves as multiple task adaptive sub-networks using our three-stage training process. To make the network compact and discretized, we propose a flow-based reduction algorithm and a squeeze loss used in the training process. We evaluate our optimized network on various public MTL datasets and show ours achieves state-of-the-art performance. An extensive ablation study experimentally validates the effectiveness of the sub-module and schemes in our framework.