【学术讲座】：On AI3.0 Internal LearningParadigm for Advanced Deep
报告题目： On AI3.0 Internal Learning Paradigm for Advanced Deep
主讲嘉宾： Prof. S.Y. Kung, Dept. EE, Princeton University, USA
邀请人 ： 陈佳义 博士
Back-propagation (BP) is known to be an external learning paradigm since it receives its supervision via the external (i.e. input/output) interfacing nodes of a neural net (NN). Consequently, the BP-based Deep Learning (NN/AI 2.0) has been applied to the training of NN’s parameters exclusively, leaving the question of finding optimal structure to merely trial and error. Arguably, structural training has to be a most vital task facing our next-generation NN/AI technology. To this end, we propose an internal learning paradigm to facilitate direct evaluation/learning of hidden nodes by means of (1) internal teacher labels (ITL); and (2) internal optimization metrics (IOM). Subsequently, via external/internal hybrid learning, we propose a parameter/structure training scheme of optimal Deep Learning/Compression Nets. In our comparative studies, the hybrid learning methodology appears to show both reduced nets and better accuracy at the same time. In fact, it consistently outperforms the existing AI2.0 deep learning/compression techniques. It is important to note that, the internal learning paradigm is, conceptually at least, one step beyond the notion of Internal Neuron's Explainablility, championed by DARPA's XAI (or NN/AI3.0).
S.Y. Kung (email@example.com) is a professor in the Department of Electrical Engineering at Princeton University, New Jersey. His research areas include machine learning, compressive privacy, data mining and analysis, statistical estimation, system identification, wireless communication, very-large-scale integration (VLSI) array processors, genomic signal processing, and multimedia information processing. He was a founding member of several technical committees of the IEEE Signal Processing Society. He served as a member of the Board of Governors of the IEEE Signal Processing Society (1989–1991). He has been the editor-in-chief of Journal of VLSI Signal Processing Systems since 1990. He has received multiple awards and recognitions. He is a Life Fellow of the IEEE.