Presented By: Biomedical Engineering
Biomedical Engineering Seminar Series
Early Detection of Dementia: Bridging Data and Knowledge Through AI, with Jiayu Zhou
Abstract:
The global incidence of dementia is increasing, with over 9.9 million new cases each year, translating to a new diagnosis every 3.2 seconds. This upward trend is expected to intensify due to the growing aging population worldwide. Research into modeling dementia encounters unique challenges, including sparse data and complex non-linear interactions between variables and outcomes. In this talk, I will share my recent work on employing artificial intelligence to unravel the mechanisms behind aging and dementia. This work leverages combined insights from data analysis and domain knowledge. Specifically, for imaging-based analysis, we have developed a novel "subspace network" approach. This technique employs efficient deep learning models for non-linear multi-task learning, even with limited data. The subspace network incrementally refines predictions by sketching a low-dimensional subspace that facilitates knowledge transfer across tasks. Our empirical findings show that this method efficiently identifies the correct parameter subspaces and surpasses existing models in predicting clinical scores of dementia from brain imaging data. Additionally, we explore the use of language markers in diagnosing dementia early. We introduce a novel reinforcement learning framework for training a dialogue agent that efficiently interacts with elderly individuals to detect signs of dementia. This agent optimizes conversations by generating a disease-specific lexical probability distribution, aiming to increase diagnostic precision while minimizing the number of conversational turns needed. We have also developed a cross-modality learning framework that synergizes language markers with brain imaging data. By using a contrastive loss technique, this framework aligns linguistic and imaging data, enhancing the predictive accuracy of language-based markers with the help of auxiliary imaging variables derived from language.
Bio:
Jiayu Zhou is an Associate Professor in the Department of Computer Science and Engineering at Michigan State University. He received his Ph.D. degree in computer science from Arizona State University in 2014. He has a broad research interest in large-scale machine learning and data mining, and biomedical informatics. He served as a technical program committee member of premier conferences such as NIPS, ICML, and SIGKDD.
Dr. Zhou's research is supported by the National Science Foundation, the National Institutes of Health, and the Office of Naval Research. He is a recipient of the National Science Foundation CAREER Award (2018). His papers received the Best Student Paper Award at the 2014 IEEE International Conference on Data Mining (ICDM), the Best Student Paper Award at the 2016 International Symposium on Biomedical Imaging (ISBI), the Best Paper Award at the 2016 IEEE International Conference on Big Data (BigData), and Best Paper Award in Health Track, the 2022 SIGKDD Conference on Knowledge Discovery and Data Mining. He was one of the final winners of the NSF/NIST Privacy-Enhancing Technologies Challenges, where his privacy-preserving machine learning innovation was showcased at the Summit of Democracy as a testament to reinforcing democratic values.
Zoom:
https://umich.zoom.us/j/94801149707
The global incidence of dementia is increasing, with over 9.9 million new cases each year, translating to a new diagnosis every 3.2 seconds. This upward trend is expected to intensify due to the growing aging population worldwide. Research into modeling dementia encounters unique challenges, including sparse data and complex non-linear interactions between variables and outcomes. In this talk, I will share my recent work on employing artificial intelligence to unravel the mechanisms behind aging and dementia. This work leverages combined insights from data analysis and domain knowledge. Specifically, for imaging-based analysis, we have developed a novel "subspace network" approach. This technique employs efficient deep learning models for non-linear multi-task learning, even with limited data. The subspace network incrementally refines predictions by sketching a low-dimensional subspace that facilitates knowledge transfer across tasks. Our empirical findings show that this method efficiently identifies the correct parameter subspaces and surpasses existing models in predicting clinical scores of dementia from brain imaging data. Additionally, we explore the use of language markers in diagnosing dementia early. We introduce a novel reinforcement learning framework for training a dialogue agent that efficiently interacts with elderly individuals to detect signs of dementia. This agent optimizes conversations by generating a disease-specific lexical probability distribution, aiming to increase diagnostic precision while minimizing the number of conversational turns needed. We have also developed a cross-modality learning framework that synergizes language markers with brain imaging data. By using a contrastive loss technique, this framework aligns linguistic and imaging data, enhancing the predictive accuracy of language-based markers with the help of auxiliary imaging variables derived from language.
Bio:
Jiayu Zhou is an Associate Professor in the Department of Computer Science and Engineering at Michigan State University. He received his Ph.D. degree in computer science from Arizona State University in 2014. He has a broad research interest in large-scale machine learning and data mining, and biomedical informatics. He served as a technical program committee member of premier conferences such as NIPS, ICML, and SIGKDD.
Dr. Zhou's research is supported by the National Science Foundation, the National Institutes of Health, and the Office of Naval Research. He is a recipient of the National Science Foundation CAREER Award (2018). His papers received the Best Student Paper Award at the 2014 IEEE International Conference on Data Mining (ICDM), the Best Student Paper Award at the 2016 International Symposium on Biomedical Imaging (ISBI), the Best Paper Award at the 2016 IEEE International Conference on Big Data (BigData), and Best Paper Award in Health Track, the 2022 SIGKDD Conference on Knowledge Discovery and Data Mining. He was one of the final winners of the NSF/NIST Privacy-Enhancing Technologies Challenges, where his privacy-preserving machine learning innovation was showcased at the Summit of Democracy as a testament to reinforcing democratic values.
Zoom:
https://umich.zoom.us/j/94801149707
Explore Similar Events
-
Loading Similar Events...
