Speaker: Jingchao Ni

Date: Jan 31, 11:45am–12:45pm

Abstract: The proliferation of data acquisition technologies such as wearable devices, sensors, data logging, imaging, and IoT, has produced vast amounts of data across domains. This influx of data has empowered AI applications in scientific domains and for social good. In healthcare, for instance, AI has unleashed the potential of data such as Electronic Health Records (EHRs) for predicting disease progression, medical time series for recognizing patient activities, medical images for diagnosing lesions, and medical articles for creating clinical QA chatbots. However, with the increasing deployment of AI systems in complex and changing environments where the data may be imperfect and constantly evolving, the challenge has shifted from model development to model deployment, raising concerns about the reliability of these systems.

In this talk, with healthcare as the area of focus, I will begin by examining the major limitations of health data that impede the deployment of machine learning models in dynamic open-world scenarios. Then I will introduce a model personalization framework for evolving data streams, which aligns with the principles of personalization in precision medicine, aiming to address the challenges of generalization caused by limited health data. Within the framework, I will present our series of works that provide insights on key research problems: 1) robust learning on irregular time series that are integrated from heterogeneous sources; 2) learning adaptable representations in the face of shifting label domains; and 3) detecting out-of-distribution tasks from task streams in meta-learning scenarios. Finally, I will conclude the talk by outlining potential future research directions.

Biographical Sketch: Dr. Jingchao Ni is an applied scientist at AWS AI Labs, Amazon. Previously, he was a researcher at NEC Laboratories America. He received his Ph.D. degree from the Pennsylvania State University in 2018. His research is centered around machine learning and data science, with a focus on the development of robust and adaptable machine learning models in data-constrained scenarios, for reliable inference in environments and tasks that are subject to change. His recent research revolves around time-varying data and graph-structured data, and their extension to areas emphasizing AI for science (e.g., biomedicine, neuroscience) and AI for social good (e.g., healthcare, cyber-physical systems, AIOps, e-commerce, finance). His research outcomes have been published in refereed conferences (e.g., ICLR, ICML, NeurIPS, AAAI, CVPR, KDD, WWW) and journals (e.g., IEEE TKDE, ACM TKDD), and delivered to products with 24 patents filed or granted. One of his work was recognized as the best paper finalist in ICDM. Also, he actively contributes to the academic community as PC/SPC members and reviewers for prestigious venues including ICLR, ICML, NeurIPS, AAAI, TKDE,

Location and Zoom link: 307 Love, or https://fsu.zoom.us/j/98339125576