Summary of Levi: Generalizable Fine-tuning Via Layer-wise Ensemble Of Different Views, by Yuji Roh et al.
LEVI: Generalizable Fine-tuning via Layer-wise Ensemble of Different Views
by Yuji Roh, Qingyun Liu, Huan Gui, Zhe Yuan, Yujin Tang, Steven Euijong Whang, Liang Liu, Shuchao Bi, Lichan Hong, Ed H. Chi, Zhe Zhao
First submitted to arxiv on: 7 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- 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 This paper addresses the limitations of fine-tuning foundation models on downstream tasks, particularly in out-of-distribution (OOD) scenarios. While fine-tuning has been successful in various tasks, recent studies have observed challenges in generalizing to unseen distributions. The authors contend that overly relying on pre-trained representations may hinder fine-tuning from learning essential representations for new tasks, leading to poor OOD generalization. To address this issue, they propose a novel fine-tuning method called LEVI (Layer-wise Ensemble of different VIews), which combines the pre-trained model with a small task-specific model layer-wise. This approach effectively suppresses problematic features in both fine-tuning data and pre-trained models while preserving useful features for new tasks. The authors demonstrate the effectiveness of LEVI through broad experiments using large language and vision models. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making sure that machines learning to do one job can also do a similar job they’ve never seen before. When we fine-tune these machines, or “foundation models,” to do a new task, it’s like giving them a new set of instructions. But sometimes this doesn’t work well when the new task is very different from what they were trained on. The authors think that this might be because we’re relying too much on what the machine learned beforehand, and not letting it learn enough about the new task. To fix this, they came up with a new way to fine-tune these machines called LEVI (Layer-wise Ensemble of different VIews). It’s like combining the old instructions with some new ones that are specific to the new task. This helps the machine learn what’s important and ignore what’s not, so it can do a better job on the new task. |
Keywords
* Artificial intelligence * Fine tuning * Generalization
