Flexible Multi-task Networks by Learning Parameter Allocation
Krzysztof Maziarz, Efi Kokiopoulou, Andréa Gesmundo et al. · arXiv (Cornell University) · 2019
This paper proposes a novel learning method for multi-task applications. Multi-task neural networks can learn to transfer knowledge across different tasks by using parameter sharing. However, sharing parameters between unrelated tasks can hurt performance. To address this issue, we propose a framework to learn fine-grained patterns of parameter sharing. Assuming that the network is composed of several components across layers, our framework uses learned binary variables to allocate components to tasks in order to encourage more parameter sharing between related tasks, and discourage parameter sharing otherwise. The binary allocation variables are learned jointly with the model parameters by standard back-propagation thanks to the Gumbel-Softmax reparametrization method. When applied to the Omniglot benchmark, the proposed method achieves a 17% relative reduction of the error rate compared to state-of-the-art.
Cited by 8 works. doi:10.48550/arxiv.1910.04915 Open-access PDF
Authors
References
- Learning Multiple Layers of Features from Tiny Images
- Fast R-CNN
- Simple statistical gradient-following algorithms for connectionist reinforcement learning
- Neural Architecture Search with Reinforcement Learning
- Multitask learning
- Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms
- Human-level concept learning through probabilistic program induction
- Group Normalization
- Efficient Neural Architecture Search via Parameters Sharing
- DARTS: Differentiable Architecture Search