Summary of Self-supervised Visual Preference Alignment, by Ke Zhu and Zheng Ge and Liang Zhao and Xiangyu Zhang
Self-Supervised Visual Preference Alignment
by Ke Zhu, Zheng Ge, Liang Zhao, Xiangyu Zhang
First submitted to arxiv on: 16 Apr 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
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 presents the first attempt at unsupervised preference alignment in Vision-Language Models (VLMs), allowing for robust and powerful answers without relying on supervision from GPT-4 or human involvement. The authors generate chosen and rejected responses to original and augmented image pairs, then conduct preference alignment using direct preference optimization. By properly designing augmentations to the image input, VLMs learn to produce more accurate answers. The pipeline achieves 90% relative score to GPT-4 on complex reasoning in LLaVA-Bench with only 8k unsupervised data and improves scores on MM-Vet by 6.7%/5.6%. Visualizations show improved ability to align with user-intentions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper helps computers better understand images and words without being told what’s right or wrong. It makes a new way for machines to learn from mistakes, making them more accurate and helpful. The authors test their idea on some big datasets and it works really well! They show that the computer can now give better answers on complex problems. This is important because it could help computers be more useful in our daily lives. |
Keywords
» Artificial intelligence » Alignment » Gpt » Optimization » Unsupervised
