ELLIS Distinguished Lecture: YingLi Tian

On August 20, 2026, YingLi Tian (City College of New York, CUNY) will give a lecture: Compressing Complexity: Knowledge Distillation and Model Quantization for Efficient and Accessible AI. Hosted by Guoying Zhao (University of Oulu and ELLIS Institute Finland).

Date and time 

August 20, 2026, 9:30-10:30 EEST

Location

TU1 Saab Auditorium, Aalto University (Maarintie 8, Espoo)
Online (Zoom)

Abstract

Modern deep learning architectures, such as Transformers, hybrid CNN–Transformer models, and iterative diffusion models, have achieved remarkable success across a wide range of visual intelligence tasks. However, these advances often come at the cost of massive parameters, high memory requirements, and substantial computational overhead, creating significant barriers to real-world deployment and resource-constrained academic research. In this talk, I will present our recent efforts to address the fundamental accuracy–efficiency trade-off through two paradigms: 1) Knowledge-Centric Efficiency: We develop knowledge distillation frameworks that transfer rich structural representations from powerful teacher models to lightweight student networks. Our work spans spectral decoupling for spatiotemporal forecasting, cross-image relational knowledge transfer for semantic segmentation, and geometric structure distillation for 3D point-cloud understanding. 2) Resource-Centric Efficiency: We directly reduce inference costs through model compression techniques, including structural pruning to eliminate redundant prediction heads in multimodal tracking systems and adaptive mixed-precision quantization strategies that mitigate outlier effects while preserving temporal consistency in low-bit diffusion models. Across diverse applications, these approaches substantially reduce latency, memory consumption, and computational demands while maintaining near-lossless performance relative to their full-scale counterparts, enabling the development of efficient, scalable, and accessible AI systems for advanced visual intelligence applications.
 

Bio

Dr. YingLi Tian is a Distinguished Professor in the Department of Electrical Engineering at the City College of New York (CCNY) and in the Department of Computer Science at the Graduate Center of the City University of New York (CUNY). She is a Fellow of the Institute of Electrical and Electronics Engineers (IEEE), the International Association of Pattern Recognition (IAPR), and the American Association for the Advancement of Science (AAAS), recognized for her pioneering contributions to automatic facial expression analysis, human activity understanding, and assistive technology. Her research on facial expression analysis and database development has had a lasting impact on the field and was honored with the Test of Time Award at the IEEE International Conference on Automatic Face and Gesture Recognition in 2019. Before joining CCNY in 2008, Dr. Tian was a Research Staff Member at the IBM T. J. Watson Research Center, where she led the video analytics team. Her current research interests include computer vision, machine learning, artificial intelligence, assistive technologies, medical image analysis, and remote sensing. Dr. Tian has authored more than 280 peer-reviewed journal and conference publications and holds 29 issued U.S. patents.