Summary of Robust Multi-task Learning with Excess Risks, by Yifei He et al.
Robust Multi-Task Learning with Excess Risks
by Yifei He, Shiji Zhou, Guojun Zhang, Hyokun Yun, Yi Xu, Belinda Zeng, Trishul Chilimbi, Han Zhao
First submitted to arxiv on: 3 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 In this paper, researchers propose a new approach to multi-task learning (MTL) that addresses the challenge of optimizing task weights when label noise is present. Existing MTL methods use adaptive weight updating schemes, but these can be overwhelmed by noisy tasks with large Bayes optimal errors. The proposed method, ExcessMTL, updates task weights based on their distances to convergence, assigning higher weights to worse-trained tasks that are further from convergence. To estimate excess risks, the authors develop an efficient and accurate method using Taylor approximation. Theoretically, they show that ExcessMTL achieves convergence guarantees and Pareto stationarity. Empirically, the algorithm outperforms existing methods on various MTL benchmarks in the presence of label noise. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary ExcessMTL is a new way to do multi-task learning that helps when there’s noise in the labels. Right now, most MTL methods get stuck because they prioritize tasks with big errors, even if those errors are just from noisy data. The ExcessMTL algorithm fixes this by giving more weight to tasks that need more work to get right. This is important because it makes MTL work better when there’s noise in the labels. |
Keywords
* Artificial intelligence * Multi task
