BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Cambridge MedAI Seminar Series - Dr. Agnieszka Słowik\, Microsof t Research Cambridge and Mariana Lindo\, Critical Techworks DTSTART:20231127T100000Z DTEND:20231127T110000Z UID:TALK208738@talks.cam.ac.uk CONTACT:Ines Machado DESCRIPTION:The *Cancer Research UK Cambridge Centre* and the *Department of Radiology at Addenbrooke's* are pleased to announce a seminar series on *Artificial Intelligence (AI) in Medicine*\, which aims to provide a comp rehensive overview of the latest developments in this rapidly evolving fie ld. As AI continues to revolutionize healthcare\, we believe it is essenti al to explore its potential and discuss the challenges and opportunities i t presents.\n\n\nThe seminar series will feature prominent experts in the field who will share their research and insights on a range of topics\, in cluding AI applications in disease diagnosis\, drug discovery\, and patien t care. Each seminar will involve two talks\, followed by an interactive d iscussion with coffee and pastries! We hope that this seminar series will be a valuable platform for researchers\, practitioners and students to lea rn about the latest trends and explore collaborations in the exciting fiel d of AI in Medicine.\n\n \nThe next seminar will be held on the *27th of N ovember 2023\, 10am at the Jeffrey Cheah Biomedical Centre (Main Lecture T heatre)\, University of Cambridge* and will feature two talks:\n\n\n\n*Tit le: "Turn and face the strange: Out-of-distribution generalisation in mach ine learning" – Dr. Agnieszka Słowik\, Microsoft Research Cambridge*\n\ n\nAgnieszka Słowik is a Postdoctoral Researcher working on R&D in robust and responsible AI at Microsoft Research Cambridge. Prior to joining Micr osoft\, she did her PhD in out-of-distribution generalisation in machine l earning at University of Cambridge. During her PhD\, she did several resea rch internships at Mila\, Meta AI and Microsoft Research\, and published a t top machine learning venues\, such as AISTATS and AAAI. \n\nAbstract: Wh en applied to a new data distribution\, machine learning algorithms have b een shown to deteriorate. Distribution shifts are caused by spurious corre lations that hold at training time but not at test time\, changes to the d omain\, as well as under- and over-representation of certain populations i n training data. In this talk\, I present two studies in the setting of le arning from multiple data sources. In the first study\, On Distributionall y Robust Optimization and Data Rebalancing\, multiple data sources are use d to minimise the error on the most challenging data source. In the second study\, Linear unit-tests for invariance discovery\, I present a set of ' unit tests' that validate whether a given algorithm ignores spurious\, uns table features that are unlikely to hold in the future\, while learning th e features that hold across all sources of training data. I conclude with a discussion of potential applications of this research to AI in medicine. \n\n \n\n*Title: "Development of a Natural Language Processing Multilingua l Model for Summarizing Radiology Reports" – Mariana Lindo\, Critical Te chworks*\n \n\nMariana Lindo is a Biomedical Engineer specialized in Medic al Informatics. She obtained her degree and master's degree at the Univers ity of Minho in Braga\, Portugal. During her academic career\, she had the opportunity to participate in projects related to AI and health\, includi ng the Scientific Talent Grant in Artificial Intelligence awarded by the C alouste Gulbenkian Foundation\, where she was able to develop a model base d on Generative Adversarial Networks for the generation of X-ray images of difficult-to-detect rib fractures. She also had the opportunity to comple te an internship at the Institute for Artificial Intelligence in Medicine in Essen\, Germany\, as part of the ERASMUS+ program\, where she focused o n developing her dissertation and learned a lot from the institute's resea rch team. She is currently working as a Data Mastermind at Critical Techwo rks\, a company responsible for the development and construction of BMW ve hicles software.\n\nAbstract: The impression section of a radiology report summarizes important radiology findings and plays a critical role in comm unicating these findings to physicians. However\, the preparation of these summaries is time-consuming and error-prone for radiologists. Recently\, numerous models for radiology report summarization have been developed. Ne vertheless\, there is currently no model that can summarize these reports in multiple languages. Such a model could greatly improve future research and the development of Deep Learning models that incorporate data from pat ients with different ethnic backgrounds. In this study\, the generation of radiology impressions in different languages was automated by fine-tuning a model\, publicly available\, based on a multilingual text-to-text Trans former to summarize findings available in English\, Portuguese\, and Germa n radiology reports. In a blind test\, two board-certified radiologists in dicated that for at least 70% of the system-generated summaries\, the qual ity matched or exceeded the corresponding human-written summaries\, sugges ting substantial clinical reliability. Furthermore\, this study showed tha t the multilingual model outperformed other models that specialized in sum marizing radiology reports in only one language\, as well as models that w ere not specifically designed for summarizing radiology reports\, such as ChatGPT.\n\n \n\nThis is a hybrid event so you can also join via Zoom: \n \n\nhttps://zoom.us/j/99050467573?pwd=UE5OdFdTSFdZeUtIcU1DbXpmdlNGZz09\n\n Meeting ID: 990 5046 7573 and Passcode: 617729\n\n \n\nWe look forward to your participation! If you are interested in getting involved and presenti ng your work\, please email Ines Machado at im549@cam.ac.uk LOCATION:Jeffrey Cheah Biomedical Centre\, Puddicombe Way\, Cambridge CB2 0AW END:VEVENT END:VCALENDAR