报告1简介：

Data-driven modelling provides a powerful way to construct mathematical representations directly from input–output observations. It is especially useful when first-principles models are difficult to obtain. Focusing on nonlinear system identification, this report introduces the NARMAX modelling framework, which can describe nonlinear dynamics involving past inputs, outputs, and noise terms. Particular attention is given to the challenge of model structure selection, where the goal is to obtain a model that is accurate, sparse, interpretable, and generalizable. Classical methods such as Orthogonal Least Squares and the Error Reduction Ratio are also discussed as efficient tools for selecting significant model terms.

报告2简介：

Feature selection in pattern recognition and structure selection in system identification can be viewed as the same fundamental problem: selecting a compact and informative subset from many correlated candidates. Based on this unified perspective, the report introduces a generalized attribute selection framework for sparse modelling. A key focus is the two-dimensional learning strategy, which simultaneously determines which attributes should be selected and how many should be retained. The report also addresses the bias–variance dilemma in model selection and discusses how information criteria and multi-objective optimization can help balance prediction accuracy, model simplicity, and interpretability.

报告人简介：

Akshya Swain is an associate professor of Electrical, Computer and Software Engineering at the University of Auckland. His research advances nonlinear system identification, control, and intelligent optimization, with applications in power and energy systems. He leads the Applied Control and System Identification group and is known for impactful contributions bridging theory and engineering practice.