Summary of Direct Preference Optimization For Suppressing Hallucinated Prior Exams in Radiology Report Generation, by Oishi Banerjee et al.
Direct Preference Optimization for Suppressing Hallucinated Prior Exams in Radiology Report Generation
by Oishi Banerjee, Hong-Yu Zhou, Subathra Adithan, Stephen Kwak, Kay Wu, Pranav Rajpurkar
First submitted to arxiv on: 10 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 research proposes a novel method for modifying generative vision-language models (VLMs) in radiology report generation, specifically targeting the suppression of unwanted hallucinations and nonsensical text. Building on recent work on direct preference optimization (DPO), the approach aims to prevent the production of prior exam hallucinations in chest X-ray report generation, a long-standing problem behavior in VLMs. The study demonstrates that DPO fine-tuning achieves a significant reduction in lines hallucinating prior exams (3.2-4.8x) while maintaining model performance on clinical accuracy metrics. |
Low | GrooveSquid.com (original content) | Low Difficulty Summary In simple terms, this research is about making computer models better at generating reports for X-rays and other medical images. These models can sometimes produce unrealistic or nonsensical text, which can be misleading for doctors and harm patients. The scientists developed a new way to fine-tune these models so they generate more accurate and helpful reports, while reducing the amount of unwanted text. This breakthrough could have significant implications for radiology and patient care. |
Keywords
» Artificial intelligence » Fine tuning » Optimization