This is an online seminar. Registration is required.
【Structured Learning Team】
【Date】2024/June/17(Mon) 15:00-16:00(JST)
*【Speaker】Yikfoong Tham, Kyushu University *
Title: Adaptive Learning with Self-Organizing Dynamical Equations and Self-Oscillatory Networks
Abstract:
Current backpropagation models, despite their outstanding achievements, still face significant limitations in adaptation and robustness. Perhaps, switching to a paradigm of learning with dynamical systems could offer a better approach to lifelong learning. In response, we have proposed models such as Self-Organizing Dynamical Equations (SODE) and Self-Oscillatory Networks (SONs). The former employs dynamical equations with bottom-up self-organizing properties to solve the chunking problem, and the latter uses a reservoir network to model oscillation behaviors. SODE has proven to be more adaptable compared to models like Word2Vec and BERT, and SONs present a more biologically plausible approach to modeling biological circuits.
Public events of RIKEN Center for Advanced Intelligence Project (AIP)
Join community