Summary of Dmtg: One-shot Differentiable Multi-task Grouping, by Yuan Gao and Shuguo Jiang and Moran Li and Jin-gang Yu and Gui-song Xia
DMTG: One-Shot Differentiable Multi-Task Grouping
by Yuan Gao, Shuguo Jiang, Moran Li, Jin-Gang Yu, Gui-Song Xia
First submitted to arxiv on: 6 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (stat.ML)
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 The proposed Multi-Task Grouping (MTG) method tackles the challenge of Multi-Task Learning (MTL) with a large number of tasks by simultaneously identifying task groups and training model weights. Unlike existing methods, MTG employs one-shot optimization to fully exploit high-order task-affinity, improving training efficiency and mitigating objective bias. The approach is formulated as a differentiable pruning problem on an adaptive network architecture determined by a categorical distribution. The method categorizes N tasks into K groups using encoder branches and initially sets up KN task heads that are gradually pruned down to N. This ensures each task is exclusively categorized into one branch. The authors demonstrate the effectiveness of MTG on CelebA and Taskonomy datasets with ablations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary MTL aims to learn multiple tasks simultaneously. To achieve this, researchers have proposed various methods. One such method is called Multi-Task Grouping (MTG). It’s a way to group similar tasks together and train models accordingly. This approach is more efficient than previous methods that did the grouping step by step. |
Keywords
* Artificial intelligence * Encoder * Multi task * One shot * Optimization * Pruning
