Routing Networks: Adaptive Selection of Non-linear Functions for Multi-Task Learning
Anonymous Clemens Rosenbaum College of Information and Computer Sciences University of Massachusetts Amherst 140 Governors Dr., Amherst, MA 01003、Tim Klinger、Matthew Riemer IBM Research AI 1101 Kitchawan Rd, Yorktown Heights, NY 10598
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 converge
中文速览
多任务学习中，不同任务之间的相互干扰（负迁移）常常会削弱共享表示所带来的好处。为解决这一问题，研究者提出了"路由网络"（Routing Network）：该架构由一个路由器和一组可复用的功能模块（function block）组成，路由器根据每个输入动态地选择并递归组合不同的功能模块，从而让相似任务共享模块、让差异任务走独立路径，自适应地平衡正负迁移。由于路由决策是离散的硬选择、不可微分，训练时采用协作式多智能体强化学习（Multi-Agent RL）来联合优化路由器和功能模块。在 MNIST、Mini-ImageNet、CIFAR-100（20个任务）等多任务图像分类基准上，路由网络在精度和收敛速度上均显著优于交叉缝合网络（cross-stitch network）和共享层基线，且训练耗时几乎不随任务数增加而线性增长——在 CIFAR-100 上达到同等精度时，训练时间减少了 85%。这项工作的重要意义在于，它首次将自动化架构搜索思路引入多任务学习，无需人工指定哪些参数共享、哪些任务专用，为大规模多任务场景提供了一种高效且通用的解决方案。
原文 arXiv:1711.01239;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1711.01239v2