Professor Rama Chellappa, AI pioneer and Bloomberg Distinguished Professor, and
Professor Ed Schlesinger, Dean of the Johns Hopkins Whiting School of Engineering
This was followed by an excellent discussion by Dr. Brian Hasselfeld, Senior Medical Director, Digital Health and Innovation at Johns Hopkins Medicine.
Below is the AI-generated summary of the November 10, 2023 seminar, with minor edits made by the AI Workgroup Leadership:
The meeting discussed the importance of technology, particularly AI, in addressing challenges in healthcare. The Business Health Initiative's working group on AI and healthcare was introduced, and its efforts to foster collaboration across universities were highlighted. The role of data in training AI models was questioned, and the potential impact of AI on medicine was discussed. The university's commitment to involve everyone in the project and its focus on translational efforts were confirmed. The challenges of implementing data science and AI in pathology were acknowledged, and the release of autonomous AI agents for paid subscribers was mentioned as an opportunity for fine-tuning.
Summary
AI and Healthcare: Investment and Collaboration
Dan highlighted the importance of technology in addressing challenges in their system and introduced the Business Health Initiative's working group on AI and healthcare, led by Rosa Wolf and Tinglong. The group meets monthly to discuss issues and foster collaboration across universities. Risa then emphasized the need to create a community around AI and healthcare and encouraged new collaborations. Tinglong introduced the featured speaker, Professor Rama Chellappa, who was set to discuss AI and its potential impact on medicine. The discussion was centered around the university's investment in AI, with Rama expressing his views and Dan sharing his perspectives as well. Ed showed interest and was open to learning more about the topic.
Integrating AI at Hopkins: Data, Collaboration, and Translation
In the meeting discussing the integration of AI at Hopkins, Ed emphasized the university's commitment to involve everyone in the project. Dan questioned the role of data in training AI models and suggested the possibility of developing a national health data infrastructure. Ed affirmed that data maintenance and curation would be a crucial aspect of the project and hinted at the library system's potential involvement. Therese showed interest in the collaboration between research, commercialization, and turning scientific discoveries into patient impact. Ed confirmed Hopkins' focus on encouraging translational efforts and aimed to create a new Kendall Square in Baltimore. Rama mentioned ongoing efforts at Hopkins, including a 5-year NIA-funded project, before experiencing technical difficulties sharing his slide deck.
National Institute of Aging's AI Research Project
Rama discussed the National Institute of Aging's award and the research conducted as part of the program. He explained that the project involves inviting proposals from companies and research groups, reviewing and selecting them, and monitoring their progress. Rama highlighted the importance of domain knowledge in medicine and the role of AI in analyzing and understanding dementia, fall prevention, and other health-related issues. He also shared some of the project's objectives, such as promoting AI technologies for analyzing dementia and fall prevention. Rama also mentioned some of the challenges faced in designing AI systems, including domain shift, adversarial attacks, bias, and interoperability. Finally, he shared plans to obtain data from up to 1,000 patients for training AI systems.
Parsons, Computing, Collaboration, Pre-Trained Model, Self-Supervised Learning, Trustworthy AI.
Rama discussed the transition of the Department of Defense to a company called Parsons, and the importance of top-class computing infrastructure for training large models. He mentioned the trend of companies like Meta and Google building larger models and the potential for collaboration with these entities. Rama also introduced a large pre-trained model for image and text recognition, which has been tested and is available on Github. He discussed self-supervised learning, a method that uses multiple variations of an original image to train a neural network. Rama highlighted the significance of human-AI interaction and trustworthy AI.
Data Science and AI in Pathology: Potential and Challenges
Rama discussed the potential and challenges of implementing data science and AI in pathology, a field that deals with vast amounts of data but often struggles with inconsistencies between labs. He highlighted the need for foundational models that can be fine-tuned and the importance of addressing domain shift, bias mitigation, and data privacy issues. Rama also emphasized the role of federated learning and generative models, and the need to balance performance and bias in AI systems. He further talked about the use of AI in explaining medical conditions to patients and improving diagnoses, and the potential of combining human and AI abilities. Lastly, he mentioned the evolution of learning techniques, now encompassing a spectrum of methodologies that depend on the nature of the data and annotations.
Exploring AI Potential and Data Privacy in Healthcare
Rama expressed his enthusiasm for the potential of AI in diverse fields, including engineering, medicine, and public health. Brian, a practicing internist and pediatrician with a computer science background, sought clarification on Federated versus centralized data and the role of data privacy in healthcare. Rama emphasized the importance of data privacy and introduced federated learning as a solution to address privacy concerns. Rama also discussed the challenges of nonlinear models in deep learning and the need for better data privacy regulations. He cited the example of Clear View, a company that collected 30 billion face images without permission, raising ethical questions. Rama highlighted the need for a solution that is equitable for everyone, and the possibility of a joint effort between him, Whiting, and the school of medicine was discussed. Stuart emphasized the need for AI experts to bridge the gap between the Whiting School and the Medical Center, which Rama confirmed is happening organically through projects like NIA. The discussion ended with Phil mentioning OpenAI's release of autonomous AI agents for paid subscribers as an opportunity for fine-tuning.
