Summary of Identifying Task Groupings For Multi-task Learning Using Pointwise V-usable Information, by Yingya Li et al.
Identifying Task Groupings for Multi-Task Learning Using Pointwise V-Usable Information
by Yingya Li, Timothy Miller, Steven Bethard, Guergana Savova
First submitted to arxiv on: 16 Oct 2024
Categories
- Main: Computation and Language (cs.CL)
- Secondary: Artificial Intelligence (cs.AI)
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 tackle the challenge of identifying the best task grouping in multi-task learning. Naive approaches can lead to negative transfer and worse performance than single-task models. The authors propose a new metric for task relatedness based on pointwise V-usable information (PVI), which measures the amount of usable information in a dataset given a model. They hypothesize that tasks with similar PVI estimates are suitable for joint learning. Experiments were conducted on 15 NLP datasets across three domains, comparing joint learners to single learners and existing methods like Llama 2 and GPT-4. The results show that grouping tasks by similar PVI estimates achieves competitive performance with fewer parameters. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about finding the best way to combine multiple learning tasks together. When you try to learn many things at once, it’s easy to make mistakes if some of the tasks are very different from each other. The authors came up with a new way to measure how similar two tasks are based on how hard they are to solve. They tested their idea on 15 datasets and found that when they grouped similar tasks together, they got better results than doing things separately. |
Keywords
» Artificial intelligence » Gpt » Llama » Multi task » Nlp
