# Flexible Multi-task Networks by Learning Parameter Allocation

arXiv: https://arxiv.org/abs/1910.04915 · HTML: https://mlsanity.com/paper/1910.04915/

## Key facts

- Cited by 7 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 10 Oct 2019.

## Built on by (most cited first, 7 of 7)

- [DSelect-k: Differentiable Selection in the Mixture of Experts with Applications to Multi-Task Learning](https://mlsanity.com/paper/2106.03760/) (2021)
- [Sparse is Enough in Scaling Transformers](https://mlsanity.com/paper/2111.12763/) (2021)
- [An Evolutionary Approach to Dynamic Introduction of Tasks in Large-scale Multitask Learning Systems](https://mlsanity.com/paper/2205.12755/) (2022)
- [muNet: Evolving Pretrained Deep Neural Networks into Scalable Auto-tuning Multitask Systems](https://mlsanity.com/paper/2205.10937/) (2022)
- [GDOD: Effective Gradient Descent using Orthogonal Decomposition for Multi-Task Learning](https://mlsanity.com/paper/2301.13465/) (2023)
- [Compositional Models: Multi-Task Learning and Knowledge Transfer with Modular Networks](https://mlsanity.com/paper/2107.10963/) (2021)
- [DTN: Deep Multiple Task-specific Feature Interactions Network for Multi-Task Recommendation](https://mlsanity.com/paper/2408.11611/) (2024)

## Abstract (arXiv)

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.