【报告摘要】Wavelet neural networks (WNNs) are powerful tools for learning unknown nonlinear mappings from data, widely used in signal processing, time-series analysis, and dynamic system control. However, their broader application is often limited by difficulties in constructing accurate wavelet bases and high computational costs. This study presents a constructive WNN framework that selects initial wavelet bases and incrementally introduces new ones to achieve predefined accuracy levels, all while reducing computational overhead. A novel contribution of this work is the frequency-based analysis of unknown nonlinear functions. By estimating the energy of spatial frequency components, we identify primary frequency elements and select suitable initial wavelets. This leads to a unique framework combining a frequency estimator and a wavelet-basis expansion mechanism, which prioritizes high-energy bases, significantly improving computational efficiency. The theoretical foundation defines the required time-frequency range for high-dimensional wavelets at a given accuracy. The framework’s applicability is demonstrated through four key use cases: (1) estimating static mappings from offline data, (2) combining two offline datasets, (3) identifying time-varying mappings from time-series data, and (4) extracting nonlinear mappings from online dynamic system measurements. These examples highlight the framework's versatility and its potential for practical, real-world applications.

【报告人简介】Dr. Ying Tan is a Professor in Mechanical Engineering at The University of Melbourne, Australia. She earned her bachelor's degree from Tianjin University, China, in 1995, and her PhD from the National University of Singapore in 2002. After a postdoctoral fellowship at McMaster University, she joined The University of Melbourne in 2004. Dr. Tan has received prestigious recognitions, including an Australian Postdoctoral Fellowship (2006-2008) and an ARC Future Fellowship (2009-2013). Currently, she serves on the ARC College of Experts (2024-2026) and holds several distinguished titles, including Fellow of IEEE (FIEEE), Engineers Australia (FIEAust), and the Asia-Pacific Artificial Intelligence Association. She is also a member of the IEEE Fellow Committee (2024-2025). Her research spans intelligent systems, nonlinear systems, data-driven optimization, rehabilitation robotics, human motor learning, wearable sensors, and model-guided machine learning.