{"id":30690,"date":"2025-10-13T11:17:35","date_gmt":"2025-10-13T15:17:35","guid":{"rendered":"https:\/\/www.bu.edu\/csmet\/?p=30690"},"modified":"2025-10-13T11:17:35","modified_gmt":"2025-10-13T15:17:35","slug":"babyvfm-data-efficient-pretraining-of-vision-llms-inspired-by-infant-learning","status":"publish","type":"post","link":"https:\/\/www.bu.edu\/csmet\/2025\/10\/13\/babyvfm-data-efficient-pretraining-of-vision-llms-inspired-by-infant-learning\/","title":{"rendered":"BabyVFM: Data-Efficient Pretraining of Vision LLMs Inspired by Infant Learning"},"content":{"rendered":"<p><iframe style=\"width: 400px; height: 285px; border: 0;\" id=\"kaltura_player\" src=\"https:\/\/cdnapisec.kaltura.com\/p\/2159741\/embedPlaykitJs\/uiconf_id\/54272102?iframeembed=true&amp;entry_id=1_r2z1k9dr&amp;config%5Bprovider%5D=%7B%22widgetId%22%3A%221_5ovhbei2%22%7D&amp;config%5Bplayback%5D=%7B%22startTime%22%3A0%7D\" allowfullscreen=\"allowfullscreen\" webkitallowfullscreen=\"webkitallowfullscreen\" mozallowfullscreen=\"mozallowfullscreen\" allow=\"autoplay *; fullscreen *; encrypted-media *\" sandbox=\"allow-downloads allow-forms allow-same-origin allow-scripts allow-top-navigation allow-pointer-lock allow-popups allow-modals allow-orientation-lock allow-popups-to-escape-sandbox allow-presentation allow-top-navigation-by-user-activation\" title=\"CS Research Seminar Oct-10-2025 Reza Rawassizadeh\"><\/iframe><\/p>\n<p><strong>BabyVFM: Data-Efficient Pretraining of Vision LLMs Inspired by Infant Learning<\/strong><\/p>\n<p><strong>Guest Speaker: <a href=\"https:\/\/www.bu.edu\/cs\/profiles\/boqing-gong\/\">Dr. Boqing Gong<\/a><\/strong><\/p>\n<p>Moderated by <a href=\"https:\/\/www.bu.edu\/csmet\/profile\/reza-rawassizadeh\/\">Dr. Reza Rawassizadeh<\/a>, Associate Professor of Computer Science<\/p>\n<p><strong>Friday, 10th October, 2025<\/strong><\/p>\n<p><strong>Abstract:<\/strong> Pretraining vision (large) foundation models (VFMs) is prohibitively expensive, making it a privilege for institutions with abundant resources and leaving independent researchers to downstream tasks, such as benchmarking, interpreting, and aligning VFMs. This situation is a crisis for computer vision research &#8212; \u201cWhat I cannot\u00a0create, I do not understand,\u201d quoted Richard Feynman. Independent researchers and the public cannot gain a true\u00a0understanding, trust, and safe use of VFMs passively from open weights or APIs. Meanwhile, the few privileged VFM creators could momentarily reach a plateau without the broad research community\u2019s nurturing.<\/p>\n<p>Hence, we propose democratizing VFM pretraining by scaling it down to a developmentally plausible framework that is scientifically reasonable and computationally friendly to university budgets, aiming to promote exploration rather than exploitation of the pretraining and enable independent researchers to build general-purpose VFMs that approach \u201cbaby intelligence\u201d to benefit efforts towards \u201cgrown-up\u201d AI.\u00a0This framework will closely mimic the minimal yet highly informative sensory experiences of human infants, encompassing:<\/p>\n<ol>\n<li><strong>Pretraining data <\/strong>curated from longitudinal, egocentric audiovisual recordings of babies.<\/li>\n<li>A suite of\u00a0<strong>developmentally aligned evaluation benchmarks <\/strong>assessing VFM capabilities against cognitive milestones like object permanence, social skills, and language acquisition.<\/li>\n<li>A user-friendly\u00a0<strong>pretraining codebase and baseline models<\/strong>.<\/li>\n<\/ol>\n<p><strong>Bio: <\/strong>Boqing Gong is a computer science faculty member at Boston University and a part-time research scientist at Google DeepMind. His research on machine learning and computer vision focuses on visual recognition, video, and AI models&#8217; generalization and efficiency.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>BabyVFM: Data-Efficient Pretraining of Vision LLMs Inspired by Infant Learning Guest Speaker: Dr. Boqing Gong Moderated by Dr. Reza Rawassizadeh, Associate Professor of Computer Science Friday, 10th October, 2025 Abstract: Pretraining vision (large) foundation models (VFMs) is prohibitively expensive, making it a privilege for institutions with abundant resources and leaving independent researchers to downstream tasks, [&hellip;]<\/p>\n","protected":false},"author":24516,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[10840],"tags":[],"_links":{"self":[{"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/posts\/30690"}],"collection":[{"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/users\/24516"}],"replies":[{"embeddable":true,"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/comments?post=30690"}],"version-history":[{"count":1,"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/posts\/30690\/revisions"}],"predecessor-version":[{"id":30693,"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/posts\/30690\/revisions\/30693"}],"wp:attachment":[{"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/media?parent=30690"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/categories?post=30690"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.bu.edu\/csmet\/wp-json\/wp\/v2\/tags?post=30690"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}