{"688478":{"#nid":"688478","#data":{"type":"news","title":"Student Getting Research Boost Through Google Ph.D. Fellowship","body":[{"value":"\u003Cp\u003EA Georgia Tech Ph.D. candidate is getting a boost to his research into developing more efficient multi-tasking artificial intelligence (AI) models without fine-tuning.\u003C\/p\u003E\u003Cp\u003EGeorgia Stoica is one of 38 Ph.D. students worldwide researching machine learning who were named a\u003Ca href=\u0022https:\/\/research.google\/programs-and-events\/phd-fellowship\/recipients\/\u0022\u003E\u003Cstrong\u003E 2025 Google Ph.D. Fellow\u003C\/strong\u003E\u003C\/a\u003E.\u003C\/p\u003E\u003Cp\u003EStoica is designing AI training methods that bypass fine-tuning, which is the process of adapting a large pre-trained model to perform new tasks. Fine-tuning is one of the most common ways engineers update large-language models like ChatGPT, Gemini, and Claude to add new capabilities.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EIf an AI company wants to give a model a new capability, it could create a new model from scratch for that specific purpose. However, if the model already has relevant training and knowledge of the new task, fine-tuning is cheaper.\u003C\/p\u003E\u003Cp\u003EStoica argues that fine-tuning still uses large amounts of data, and that other methods can help models learn more effectively and efficiently.\u003C\/p\u003E\u003Cp\u003E\u201cFull fine-tuning yields strong performance, but it can be costly, and it risks catastrophic forgetting,\u201d Stoica said. \u201cMy research asks if we can extend a model\u2019s capabilities by imbuing it with the expertise of others, without fine-tuning?\u003C\/p\u003E\u003Cp\u003E\u201cReducing cost and improving efficiency is more important than ever. We have so many publicly available models that have been trained to solve a variety of tasks. It\u2019s redundant to train a new model from scratch. It\u2019s much more efficient to leverage the information that already exists to get a model up to speed.\u201d\u003C\/p\u003E\u003Cp\u003EStoica said the solution is a cost-effective method called model merging. This method combines two or more AI models into a single model, improving performance without fine-tuning.\u003C\/p\u003E\u003Cp\u003EOn a basic level, Stoica said an example would be combining a model that is efficient at classifying cats with one that works well at dogs.\u003C\/p\u003E\u003Cp\u003E\u201cMerging is cheap because you just take the parameters, the weights of your existing models, and combine them,\u201d he said. \u201cYou could take the average of the weights to create a new model, but that sometimes doesn\u2019t work. My work has aimed to rearrange the weights so they can communicate easily with each other.\u201d\u003C\/p\u003E\u003Cp\u003EThrough his Google fellowship, Stoica seeks to apply model merging to create a cutting-edge vision encoder. A vision encoder converts image or video data into numerical representations that computers can understand. This enables tasks such as image or facial recognition and generative image captioning.\u003C\/p\u003E\u003Cp\u003E\u201cI want to be at the frontier of the field, and Google is clearly part of that,\u201d Stoica said. \u201cThe vision encoder is very large-scale, and Google has the infrastructure to accommodate it.\u201d\u003C\/p\u003E","summary":"","format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003EGeorgia Stoica is one of 38 Ph.D. students worldwide researching machine learning who were named a\u003Ca href=\u0022https:\/\/research.google\/programs-and-events\/phd-fellowship\/recipients\/\u0022\u003E\u003Cstrong\u003E 2025 Google Ph.D. Fellow\u003C\/strong\u003E\u003C\/a\u003E.\u003C\/p\u003E\u003Cp\u003EStoica is designing AI training methods that bypass fine-tuning, which is the process of adapting a large pre-trained model to perform new tasks. Fine-tuning is one of the most common ways engineers update large-language models like ChatGPT, Gemini, and Claude to add new capabilities.\u0026nbsp;\u003C\/p\u003E","format":"limited_html"}],"field_summary_sentence":[{"value":"Georgia Stoica is one of 38 Ph.D. students worldwide researching machine learning who were named a 2025 Google Ph.D. Fellow."}],"uid":"36530","created_gmt":"2026-02-23 17:43:54","changed_gmt":"2026-03-20 12:53:05","author":"Nathan Deen","boilerplate_text":"","field_publication":"","field_article_url":"","location":"Atlanta, GA","dateline":{"date":"2026-02-23T00:00:00-05:00","iso_date":"2026-02-23T00:00:00-05:00","tz":"America\/New_York"},"extras":[],"hg_media":{"679394":{"id":"679394","type":"image","title":"IMG_2942-copy-2.jpg","body":null,"created":"1771868657","gmt_created":"2026-02-23 17:44:17","changed":"1771868657","gmt_changed":"2026-02-23 17:44:17","alt":"George Stoica","file":{"fid":"263553","name":"IMG_2942-copy-2.jpg","image_path":"\/sites\/default\/files\/2026\/02\/23\/IMG_2942-copy-2.jpg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2026\/02\/23\/IMG_2942-copy-2.jpg","mime":"image\/jpeg","size":112361,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2026\/02\/23\/IMG_2942-copy-2.jpg?itok=KCVheh-u"}}},"media_ids":["679394"],"groups":[{"id":"47223","name":"College of Computing"},{"id":"1188","name":"Research Horizons"},{"id":"50876","name":"School of Interactive Computing"}],"categories":[{"id":"194606","name":"Artificial Intelligence"},{"id":"153","name":"Computer Science\/Information Technology and Security"}],"keywords":[{"id":"3165","name":"google"},{"id":"9143","name":"Graduate Research Fellowship"},{"id":"192863","name":"go-ai"},{"id":"9153","name":"Research Horizons"},{"id":"187915","name":"go-researchnews"}],"core_research_areas":[{"id":"193655","name":"Artificial Intelligence at Georgia Tech"}],"news_room_topics":[{"id":"71871","name":"Campus and Community"}],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[],"email":[],"slides":[],"orientation":[],"userdata":""}}}