Summary of Mosld: An Extremely Parameter-efficient Mixture-of-shared Loras For Multi-task Learning, by Lulu Zhao et al.
MoSLD: An Extremely Parameter-Efficient Mixture-of-Shared LoRAs for Multi-Task Learning
by Lulu Zhao, Weihao Zeng, Xiaofeng Shi, Hua Zhou
First submitted to arxiv on: 12 Dec 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
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 paper proposes a novel mixture-of-shared-LoRAs (MoSLD) model to address the challenges of fine-tuning large pre-trained models in multi-task learning scenarios. MoSLD shares the upper projection matrix among experts, promoting general knowledge across tasks while allowing lower matrices to focus on task-specific features. A dropout strategy alleviates imbalanced updates and mitigates overfitting. The model demonstrates excellent performance in both single- and multi-task settings with robust out-of-domain generalization capabilities. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper creates a new way to fine-tune big models for many tasks at once. It solves two main problems: how different types of data interact and how the model forgets what it learned from one task when moving to another. The solution is called MoSLD, which shares important information among different parts of the model. This helps the model learn general things that apply across all tasks, while still focusing on unique details for each task. The model also includes a “dropout” feature to prevent overfitting and ensure it generalizes well to new situations. |
Keywords
» Artificial intelligence » Domain generalization » Dropout » Fine tuning » Multi task » Overfitting
