{"673976":{"#nid":"673976","#data":{"type":"event","title":"PhD Defense | Less is More: Accelerating Vision by Eliminating Redundancy","body":[{"value":"\u003Cp\u003EDaniel Bolya - Machine Learning PhD Student - School of Interactive Computing\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EDate:\u0026nbsp;\u003C\/strong\u003EApril 12th\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003ETime: \u003C\/strong\u003E4:00 PM \u2013 5:30 PM ET\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003ELocation\u003C\/strong\u003E: Coda C1115 Druid Hills\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EMeeting Link\u003C\/strong\u003E: \u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Ca href=\u0022https:\/\/gatech.zoom.us\/j\/96608837820?pwd=cGtSOXZMaHRVL0g0ZGN2aE9QeTNaZz09\u0022\u003Ehttps:\/\/gatech.zoom.us\/j\/96608837820?pwd=cGtSOXZMaHRVL0g0ZGN2aE9QeTNaZz09\u003C\/a\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003ECommittee\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003EJudy Hoffman (Advisor), School of Interactive Computing\u003C\/p\u003E\r\n\r\n\u003Cp\u003EJames Hays, School of Interactive Computing\u003C\/p\u003E\r\n\r\n\u003Cp\u003EZsolt Kira, School of Interactive Computing\u003C\/p\u003E\r\n\r\n\u003Cp\u003EDhruv Batra, School of Interactive Computing\u003C\/p\u003E\r\n\r\n\u003Cp\u003EChristoph Feichtenhofer, FAIR, Meta\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EAbstract\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThe massive models that power today\u2019s state-of-the-art in computer vision require trillions of floating-point operations to compute. But how much of these operations do we really need? Given how well techniques like pruning or quantization work, it\u2019s clear that a lot of this computation is redundant. My work focuses on speeding up vision models by reducing redundancy with simple but powerful techniques. In this thesis defense, I\u2019ll give a brief overview of all of my work and then hone in on discussing Token Merging to merge redundant tokens in vision transformers for classification and diffusion and Hiera, which removes redundant modules in modern vision architectures by explicitly teaching spatial bias. Then, I\u0027ll show that you can combine these and other approaches for a multiplicative effect (for e.g., ~10x speed-up on video).\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n","summary":"","format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003ELess is More: Accelerating Vision by Eliminating Redundancy\u003C\/p\u003E\r\n","format":"limited_html"}],"field_summary_sentence":[{"value":"Daniel Bolya - Machine Learning PhD Student - School of Interactive Computing"}],"uid":"36518","created_gmt":"2024-04-03 19:31:00","changed_gmt":"2024-04-03 19:34:00","author":"shatcher8","boilerplate_text":"","field_publication":"","field_article_url":"","field_event_time":{"event_time_start":"2024-04-12T16:00:00-04:00","event_time_end":"2024-04-12T17:30:00-04:00","event_time_end_last":"2024-04-12T17:30:00-04:00","gmt_time_start":"2024-04-12 20:00:00","gmt_time_end":"2024-04-12 21:30:00","gmt_time_end_last":"2024-04-12 21:30:00","rrule":null,"timezone":"America\/New_York"},"location":"CODA C1115 Druid Hills","extras":[],"groups":[{"id":"576481","name":"ML@GT"}],"categories":[],"keywords":[],"core_research_areas":[],"news_room_topics":[],"event_categories":[{"id":"1788","name":"Other\/Miscellaneous"}],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[],"email":[],"slides":[],"orientation":[],"userdata":""}}}