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

Source: [Routing Networks: Adaptive Selection of Non-linear Functions for Multi-Task Learning](<https://www.sota2.com/research/paper/68fa63e91e209131cea01636>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-25T20:52:10.607Z

Authors: Clemens Rosenbaum, Tim Klinger, Matthew Riemer

Published: 2017-11-03T17:07:51.000Z

## 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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Multi-task Reinforcement Learning on Meta-World MT50 Conditioned v1](<https://www.sota2.com/research/sota/multi-task-reinforcement-learning-on-meta-world-mt50-conditioned-v1.md>): Multi-task reinforcement learning; Meta-World MT50 Conditioned v1; Average Success Rate: 29.1.

- [Multi-task Reinforcement Learning on Meta-World MT10 Conditioned v1](<https://www.sota2.com/research/sota/multi-task-reinforcement-learning-on-meta-world-mt10-conditioned-v1.md>): Multi-task reinforcement learning; Meta-World MT10 Conditioned v1; Average Success Rate: 27.

- [Multi-task Reinforcement Learning on Meta-World MT50 v1 (Fixed)](<https://www.sota2.com/research/sota/multi-task-reinforcement-learning-on-meta-world-mt50-v1-fixed.md>): Multi-task reinforcement learning; Meta-World MT50 v1 (Fixed); Success Rate: 22.9.

- [Multi-task Reinforcement Learning on Meta-World MT10 v1 (Fixed)](<https://www.sota2.com/research/sota/multi-task-reinforcement-learning-on-meta-world-mt10-v1-fixed.md>): Multi-task reinforcement learning; Meta-World MT10 v1 (Fixed); Success Rate: 20.8.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/1711.01239>)

[PDF](<https://arxiv.org/pdf/1711.01239.pdf>)
