A lecture on “Machine Learning with Connectionist Models – A Developmental Perspective”was held by RIIT

On Mar 22, Prof. Biing Hwang (Fred) Juang, a member of the Advisory Committee on Speech and Language Technology R&D Center and a professor from Georgia Institute of Technology, was invited to give a lecture on “Machine Learning with Connectionist Models – A Developmental Perspective” in Room 1-315 in FIT Building.

In the lecture, Prof. Juang, with his profound knowledge, explained in detail the probability and the learning mechanism of Deep Neural Networks (DNN) from the perspective of the whole machine learning development. Prof. Juang pointed out that the big success of DNN was rooted in sixty years of long-term research on machine learning and statistical pattern recognition, especially the epistemological leap in collaborative learning of feature and mode. The core of DNN method is not the increase of the pattern’s depth, but the selection of distinct features by the neural model and the combination of memory functions and unsupervised learning methods to generate strong learning abilities. Based on this, DNN has solved an important problem which confused the academia for a long time: how to learn feature extraction to make descriptions and classifications by simple models. In fact, it blurs the boundaries between features and modeling, which has a profound impact on basic concepts of machine learning.