报告地点：行健楼学术活动室526
邀请人：朱建栋教授
报告摘要：
This talk covers two aspects of the use of recurrent neural networks (RNNs) in engineering signal processing. The first part focuses on using RNNs for controller design in nonlinear systems. The system states are stacked in time while the controller is represented by the RNN output. A novel activation function is imposed on RNNs to achieve finite-time convergence and improve control performance, with a brief stability proof provided. The second part presents a learning-based multiscale feature engineering (LMFE) framework for detecting power disturbances in a three-phase system. The three-phase measurements are preprocessed and features are extracted at both the global and local scale, including phase-level and measurement-level aggregations and waveform information. The extracted features are fed into a classifier trained by an RNN. The LMFE approach is evaluated on the VSB ENET dataset and outperforms existing solutions for PD detection.
报告人简介：
刘熔洁博士，现任佛罗里达州立大学统计系终身制助理教授，IEEE senior member，于2010年在东南大学数学系获得信息与计算科学学士学位，2016年在美国德州大学圣安东尼奥分校获得电子与计算机工程博士学位，2020年在美国莱斯大学George R. Brown工程学院获得统计学博士学位。研究方向包括：非线性系统最优控制，计算机视觉，机器/深度学习，强化学习等。在高水平国际最优控制，机器学习，图像模式识别的期刊和顶会（Journal of Optimization Theory and Application，IEEE Transactions on Neural Networks and Learning Systems，Neurocomputing，NeuroImage，CVPR，ACC，CDC）发表学术论文40多篇。
