Summary of Continual Sft Matches Multimodal Rlhf with Negative Supervision, by Ke Zhu and Yu Wang and Yanpeng Sun and Qiang Chen and Jiangjiang Liu and Gang Zhang and Jingdong Wang
Continual SFT Matches Multimodal RLHF with Negative Supervision
by Ke Zhu, Yu Wang, Yanpeng Sun, Qiang Chen, Jiangjiang Liu, Gang Zhang, Jingdong Wang
First submitted to arxiv on: 22 Nov 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Computer Vision and Pattern Recognition (cs.CV)
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 abstract presents a novel approach to continually improve the comprehension of vision-language models (VLMs) through negative supervised finetuning (nSFT). The authors argue that conventional multimodal reinforcement learning with human feedback (RLHF) overestimates its benefits, instead attributing its value to the logit of rejected responses. They propose a simple SFT loss that aligns VLMs more efficiently than RLHF, requiring fewer large models. The effectiveness of nSFT is demonstrated through comparisons with various multimodal RLHF approaches across different datasets and evaluation metrics. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine learning how computers can better understand what they see and hear. A team of researchers has discovered a new way to make these machines smarter by using something called “negative supervision”. They found that this approach is more efficient than the usual method, which requires many large models working together. The new method, called negative supervised finetuning (nSFT), helps computers understand better by giving them small corrections when they get things wrong. This breakthrough could lead to even smarter computers in the future. |
Keywords
» Artificial intelligence » Reinforcement learning » Rlhf » Supervised
