报告人：邓波 教授（美国内布拉斯加大学林肯分校）

时间： 2026年05月25日 10:00-

地点：理科楼LA103

摘要：By definition, training an artificial neural network is finding the global minimum of its loss function. The Gradient Descent Tunneling method  solves the training problem in theory. In practice, for training problems with large data sizes, the method is very slow, or not always working. In this talk, we introduce a new model architecture for which the training problem can always be solved quickly. The key difference from the conventional deep ANNs lies in that our new architecture is a parallel web of shallow neural networks, which allows parallel training in short amount of time.

邀请人：穆春来

欢迎广大师生积极参与！