Summary of Fair Resource Allocation in Multi-task Learning, by Hao Ban et al.
Fair Resource Allocation in Multi-Task Learning
by Hao Ban, Kaiyi Ji
First submitted to arxiv on: 23 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Multi-task learning (MTL) can improve data efficiency and generalization performance by leveraging shared knowledge across tasks. However, conflicting gradients can hinder the optimization of some tasks, impeding overall performance. Our paper formulates MTL as a utility maximization problem, inspired by fair resource allocation in communication networks. We propose FairGrad, a novel MTL optimization method that achieves flexible emphasis on certain tasks and theoretical convergence guarantee. Extensive experiments demonstrate state-of-the-art performance among gradient manipulation methods on multi-task benchmarks in supervised learning and reinforcement learning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Multi-task learning is a way to improve how machines learn by using shared knowledge across different tasks. Right now, this approach has a problem: it can be hard to optimize some tasks without affecting others. We came up with a new idea that helps solve this problem by thinking about fairness in communication networks. Our method, called FairGrad, allows us to focus on certain tasks and still achieve good results. We tested our method on many different problems and showed that it works better than other methods. |
Keywords
* Artificial intelligence * Generalization * Multi task * Optimization * Reinforcement learning * Supervised
