{"author":{"name":"Ming Li, Jie Wu, Justin Cui, Xiaojie Li, Rui Wang, Chen Chen","slug":"ming-li-jie-wu-justin-cui-xiaojie-li-rui-wang-chen-chen","article_count":1,"latest_published_at":"2026-04-30T13:20:08.468+00:00","profile_url":"https://vff.ai/authors/ming-li-jie-wu-justin-cui-xiaojie-li-rui-wang-chen-chen","api_url":"https://vff.ai/api/authors/ming-li-jie-wu-justin-cui-xiaojie-li-rui-wang-chen-chen"},"articles":[{"slug":"vipo-visual-preference-optimization-at-scale","title":"Poly-DPO and ViPO: Scaling Visual Preference Optimization","url":"https://vff.ai/article/2026/04/30/vipo-visual-preference-optimization-at-scale","content_type":"research_summary","summary":"Researchers introduced Poly-DPO, an algorithmic extension to preference optimization that adds a polynomial term to handle noisy preference data, and ViPO, a large-scale dataset of 1M image pairs and 300K video pairs with balanced distributions and high-quality signals. The work addresses a core bottleneck in scaling visual generative models: existing preference datasets contain conflicting patterns and quality issues that prevent effective learning. When applied to high-quality data, Poly-DPO converges to standard DPO, validating both the dataset quality and the algorithm's adaptive design.","published_at":"2026-04-30T13:20:08.468+00:00","updated_at":"2026-05-07T02:19:52.375108+00:00","source":{"url":"https://arxiv.org/abs/2604.24953","name":"ArXiv (cs.AI)"},"featured_image":{"url":"https://cdn.prod.website-files.com/65e5ee2bf77d7d26c3acfcc7/67117ccd5209ed2748ecf026_storj-blog-ai-bottleneck.webp","alt":null},"categories":[{"name":"Research","slug":"research"},{"name":"Multimodal","slug":"multimodal"},{"name":"Data & Training","slug":"data-training"},{"name":"Generative AI","slug":"generative-ai"}]}]}