Summary of Code: Contrasting Self-generated Description to Combat Hallucination in Large Multi-modal Models, by Junho Kim et al.
CODE: Contrasting Self-generated Description to Combat Hallucination in Large Multi-modal Models
by Junho Kim, Hyunjun Kim, Yeonju Kim, Yong Man Ro
First submitted to arxiv on: 4 Jun 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- 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 Large Multi-modal Models (LMMs) have shown impressive capabilities in visual context understanding and coherent response generation, but the issue of hallucinations has emerged as a significant challenge. Hallucinations occur when LMMs produce erroneous responses unrelated to the visual contents. To address this issue, we introduce COuntering DEscription Contrastive Decoding (CODE), a novel decoding method that leverages self-generated descriptions as contrasting references during the decoding phase. CODE utilizes these descriptions as visual counterparts to correct and improve response alignment with actual visual content. By dynamically adjusting information flow and distribution of next-token predictions in the LMM’s vocabulary, CODE enhances coherence and informativeness of generated responses. Our experiments demonstrate that our method significantly reduces hallucinations and improves cross-modal consistency across various benchmarks and cutting-edge LMMs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine you’re talking to a computer that can understand pictures and give good answers. But sometimes, this computer makes up things that aren’t true! This is called hallucination. To fix this problem, we created a new way for the computer to generate answers. It uses its own ideas as clues to make sure the answer is related to what it’s looking at. We tested our method with many different pictures and computers, and it worked really well. Our new way of answering questions helps the computer be more accurate and reliable. |
Keywords
» Artificial intelligence » Alignment » Hallucination » Multi modal » Token
