Special Title: Building Next Generation Edge Vision System through Automatic AI Model Compression and Self-Learning
Abstract: This talk addresses the key factors in building the next generation edge vision system through a systematic design approach incorporating automatic AI model compression and self-learning methodology. To compress an AI model in a systematic way, what we proposed include the AI model pruning tool (P-Craft) and AI quantization tool (Q-Craft), that allow users to compress the AI models via user-defined setting to achieve the goal of fitting in the provided processing power via the selected AI SoC to achieve the accuracy of the target applications. Some examples for the AI model compression via the proposed P-Craft and Q-Craft will be discussed, which covers the applications of smart transportation, smart manufacturing, smart healthcare, etc. In addition to AI model compression, some important tools to assist the training of compressed AI models will also be introduced, including self-training tool (eSL-Craft) for AI model training and fine tuning and the corner case dataset generation tool (GenAI-Craft) to provide more diverse datasets used for AI model training in a click. With all these AI model compression and self-learning tools, users can quickly design the slim fit AI model inferenced on the target AI SoC for applications in an efficient way.

Biodata: Prof. Jiun-In Guo received the B.S. and Ph.D. degrees in Electronics Engineering from National Chiao Tung University (NCTU), Hsinchu, Taiwan, in 1989 and 1993, respectively. He is currently a Distinguished Professor of the Institute of Electronics, National Yang Ming Chiao Tung University (NYCU), Hsinchu, Taiwan, and the Founder and CTO of his start-up, eNeural Technologies, Inc, founded in March 2022. His research interests include images, multimedia, and digital signal processing, VLSI algorithm/architecture design, digital SIP design, SOC design, and intelligent vision processing applications including ADAS/Self-driving vehicles. Prof. Guo received the outstanding electrical engineering professor award from the Chinese Institute of Electrical Engineering in 2010, the outstanding engineering professor award from the Chinese Institute of Engineers in 2014, the outstanding research award of Minister of Science (MOST) in 2017, as well as the outstanding technology transferring award of MOST in 2018 and 2020 with the topic of deep learning ADAS systems. Prof. Guo was also selected as the Elsevier 1960-2020, 1960-2021, 1960-2022, 1960-2023 top 2% Scientist in Life-long Impact by Stanford University in consecutive four years. Prof. Guo is the author of 273 technical papers on the research areas and has served as the PI/Co-PI of 117 research projects, 133 industrial projects, and 65 industrial technology transfer projects. Prof. Guo is also the inventor of 73 invention patents and the receiptent of 129 awards in the research areas. With all the cumulated research outcome, Prof. Guo starts up a company called eNeural Technologies, Inc. since March 2022, where eNeural Technologies Inc. is an embedded AI design service house in the area of automotive and AIOT applications to help customers to solve the pain points in developing quality light weight AI models on embedded computing platforms. For more information about eNeural Technologies Inc., please visit the official website below: https://www.eneural.ai/.

Speech Title: From Centralized AI to Distributed Intelligence: Edge Language Models for Ubiquitous Intelligent Systems
Abstract: Recent advances in foundation models have dramatically expanded the capability of artificial intelligence, while also reinforcing a cloud-centric paradigm. In ubiquitous intelligent systems, however, intelligence should not only become more capable but also more widely distributed across users, sensors, devices, and physical environments. This talk discusses edge language models as a key enabler of distributed intelligence. I first review recent advances in connecting sensor data with language through multimodal representation learning and semantic grounding, enabling language-based understanding and reasoning over physical environments. I then present our recent research on local intelligence, trustworthy context understanding, and cooperative device agents, illustrating how edge language models can support intelligent behavior beyond cloud-centric AI. Finally, I discuss future research challenges toward distributed intelligence—including semantic communication among edge agents, open-world context understanding, trustworthy local reasoning, and adaptive cooperation across heterogeneous IoT devices—and outline our vision through a newly launched research project on distributed IoT intelligence toward collaborative situation understanding, from smart homes to smart cities.

Biodata: de received the B.E., M.E., and Ph.D. degrees in communications engineering from Osaka University in 1988, 1990, and 1997, respectively. From 1991 to 2008, he was an Assistant Professor and an Associate Professor with Osaka University. He has been a Professor with the Department of Computer Science and Intelligent Systems, Graduate School of Engineering, Osaka Prefecture University since 2008, which is renewed as Graduate School of Informatics, Osaka Metropolitan University from 2022. His current research interests include architectures and controls for optical networks, wireless LAN and sensor networks, network security, and future Internet. He is a Fellow of the Institute of Electronics Information and Communication Engineers, Japan.