Task Selection Policies for Multitask Learning
arXiv:1907.06214
Abstract
One of the questions that arises when designing models that learn to solve multiple tasks simultaneously is how much of the available training budget should be devoted to each individual task. We refer to any formalized approach to addressing this problem (learned or otherwise) as a task selection policy. In this work we provide an empirical evaluation of the performance of some common task selection policies in a synthetic bandit-style setting, as well as on the GLUE benchmark for natural language understanding. We connect task selection policy learning to existing work on automated curriculum learning and off-policy evaluation, and suggest a method based on counterfactual estimation that leads to improved model performance in our experimental settings.
References in corpus (15)
- Distilling the Knowledge in a Neural Network
- Theoretical Models of Learning to Learn
- SQuAD: 100,000+ Questions for Machine Comprehension of Text
- GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding
- Recurrent World Models Facilitate Policy Evolution
- The Natural Language Decathlon: Multitask Learning as Question Answering
- Doubly Robust Policy Evaluation and Learning
- Sentence Encoders on STILTs: Supplementary Training on Intermediate Labeled-data Tasks
- Improving Multi-Task Deep Neural Networks via Knowledge Distillation for Natural Language Understanding
- Learning General Purpose Distributed Sentence Representations via Large Scale Multi-task Learning
- Doubly Robust Off-policy Value Evaluation for Reinforcement Learning
- Learning from Scarce Experience
- Counterfactual Reasoning and Learning Systems
- A comparative study of counterfactual estimators
- Off-the-Shelf Unsupervised NMT