Chitta Ranjan, Ph.D.
Director of Science
Abstract: Theory of Pooling in Deep Learning
Pooling is essential in most convolutional networks. It makes them invariant, and computationally and statistically efficient. In fact, pooling has been pivotal in making convolutional networks the workhorse of deep learning. Despite the successes it has brought, little is known about the theory behind pooling. In absence of an overarching pooling theory, research and development in it has lagged behind other topics in deep learning. The talk draws the theory of pooling to show directions for advanced research in the field.
The talk will also briefly cover the Rare Event Prediction problem and research directions in it.
Tune in via Zoom:
Meeting ID: 936 9569 8478
Department of Industrial & Systems Engineering