第73回統計的機械学習セミナー / The 73rd Statistical Machine Learning Seminar (Hybrid)

【Date & Time】
28 April, 2026 (Monday)　11:00 - 12:00
Admission Free, No Booking Necessary
【Place】
Seminar Room 4 (D312-B), The Institute of Statistical Mathematics
Online : Zoom
【Speaker】
Justin Dauwels (Associate Professor, the TU Delft)
【Title】
Efficient Brain-Inspired Learning via Predictive Coding
【Abstract】
This presentation explores predictive coding as a brain-inspired framework for learning in artificial neural networks. In contrast to backpropagation, predictive coding relies on local prediction errors and iterative inference, offering a more biologically grounded view of how neural systems may process sensory information and update internal representations. Beyond its biological motivation, predictive coding provides a principled connection between learning, inference, and uncertainty modeling. The talk first introduces the core ideas behind predictive coding, including hierarchical generative models, free-energy minimization, and local learning rules. It then discusses key limitations of standard predictive coding, such as slow iterative inference, delayed error propagation, and reduced scalability in deeper networks. To address these challenges, the presentation highlights recent advances that incorporate direct feedback pathways into the predictive coding framework, enabling faster and more effective error transmission across layers while preserving locality. Overall, the presentation positions predictive coding as a promising route toward efficient, scalable, and biologically inspired learning, with potential implications for both neuroscience and next-generation artificial intelligence systems.
【Biography】
Dr. Justin Dauwels is an Associate Professor at the TU Delft (Signals and Systems, Department of Microelectronics) and serves as co-Director of the Safety and Security Institute at the TU Delft. He is also the scientific lead of the Model-Driven Decisions Lab (MoDDL), a first lab for the Knowledge Building program between the Netherlands police and the TU Delft. His research interests are in data analytics with applications to prediction problems (e.g., nowcasting of precipitation, remaining useful lifetime prediction of electronic components), intelligent transportation systems, autonomous systems, and analysis of human behavior and physiology. His academic lab has spawned four startups across a range of industries, ranging from AI for healthcare to autonomous vehicles.
【Contact】
E-mail : shiroism.ac.jp