Summary of Cpt: Consistent Proxy Tuning For Black-box Optimization, by Yuanyang He et al.
CPT: Consistent Proxy Tuning for Black-box Optimization
by Yuanyang He, Zitong Huang, Xinxing Xu, Rick Siow Mong Goh, Salman Khan, Wangmeng Zuo, Yong Liu, Chun-Mei Feng
First submitted to arxiv on: 1 Jul 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 address the limitations of proxy-tuning, a technique used to fine-tune black-box language models. They introduce Consistent Proxy Tuning (CPT), a novel method that exploits both a large frozen black-box model and a small frozen white-box model to ensure consistency between training-stage optimization and test-time proxies. This approach improves model performance by applying the difference of output logits before and after tuning a smaller proxy model. The authors demonstrate the effectiveness of CPT in fine-tuning Large Language Models (LLMs) and Vision-Language Models (VLMs) across various datasets, outperforming existing methods like proxy-tuning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary In simple terms, this paper is about finding ways to improve how we adjust or “tune” language models without knowing the inner workings of these complex algorithms. The authors propose a new method called Consistent Proxy Tuning (CPT) that uses two different models to make adjustments and ensure better results. They test CPT on various language tasks and show that it outperforms other methods. |
Keywords
* Artificial intelligence * Fine tuning * Logits * Optimization
