Routing Networks: Adaptive Selection of Non-linear Functions for Multi-Task Learning

Abstract

Multi-task learning (MTL) with neural networks leverages commonalities in tasks to improve performance, but often suffers from task interference which reduces the benefits of transfer. To address this issue we introduce the routing network paradigm, a novel neural network and training algorithm. A routing network is a kind of self-organizing neural network consisting of two components: a router and a set of one or more function blocks. A function block may be any neural network - for example a fully-connected or a convolutional layer. Given an input the router makes a routing decision, choosing a function block to apply and passing the output back to the router recursively, terminating when a fixed recursion depth is reached. In this way the routing network dynamically composes different function blocks for each input. We employ a collaborative multi-agent reinforcement learning (MARL) approach to jointly train the router and function blocks. We evaluate our model against cross-stitch networks and shared-layer baselines on multi-task settings of the MNIST, mini-imagenet, and CIFAR-100 datasets. Our experiments demonstrate a significant improvement in accuracy, with sharper convergence. In addition, routing networks have nearly constant per-task training cost while cross-stitch networks scale linearly with the number of tasks. On CIFAR-100 (20 tasks) we obtain cross-stitch performance levels with an 85% reduction in training time.

Under Review at ICLR 2018

References in corpus (6)

• Adam: A Method for Stochastic Optimization
• Designing Neural Network Architectures using Reinforcement Learning
• SMASH: One-Shot Model Architecture Search through HyperNetworks
• Deep Networks with Internal Selective Attention through Feedback Connections
• AdaNet: Adaptive Structural Learning of Artificial Neural Networks
• Deep Sequential Neural Network

Cited by in corpus (19)

• Multi-Task Learning with Deep Neural Networks: A Survey
• Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity
• AdaShare: Learning What To Share For Efficient Deep Multi-Task Learning
• Multi-Task Reinforcement Learning with Soft Modularization
• Text as Neural Operator: Image Manipulation by Text Instruction
• Meta-Consolidation for Continual Learning
• Deep Transformers with Latent Depth
• Exploring Shared Structures and Hierarchies for Multiple NLP Tasks
• Object Files and Schemata: Factorizing Declarative and Procedural Knowledge in Dynamical Systems
• Discrete-Valued Neural Communication
• A Meta-MDP Approach to Exploration for Lifelong Reinforcement Learning
• Feature Partitioning for Efficient Multi-Task Architectures
• Routing Networks with Co-training for Continual Learning
• Navigating the Trade-Off between Multi-Task Learning and Learning to Multitask in Deep Neural Networks
• Transformers with Competitive Ensembles of Independent Mechanisms
• Decentralized Reinforcement Learning: Global Decision-Making via Local Economic Transactions
• Modular meta-learning in abstract graph networks for combinatorial generalization
• Multi-path Neural Networks for On-device Multi-domain Visual Classification
• Fast Line Search for Multi-Task Learning