Keynote speech: Keynote 1: It's all about AI

Abstract

In this talk, I will cover two topics which are closely related to AI. The first one is ``spatiotemporal learning of basketball offensive strategies’’ and the second one is ``learning to classify shot types.’’ Video-based group behavior analysis is drawing attention to its rich application in sports, military, surveillance and biological observations. Focusing specifically on the analysis of basketball offensive strategies, in the first topic we introduce a systematic approach to establishing unsupervised modeling of group behaviors and then use it to perform tactics classification. In the second topic, a deep-net based fusion strategy is proposed to classify shots in concert videos. Varying types of shots are fundamental elements in the language of film, commonly used by a visual storytelling director to convey the emotion, ideas, and art. To classify such types of shots from images, we present a new framework that facilitates the intriguing task by addressing two key issues. First, we learn more effective features by fusing the layer-wise outputs extracted from a deep convolutional neural network (CNN). We then introduce a probabilistic fusion model, termed error weighted deep cross-correlation model, to boost the classification accuracy. We provide extensive experiment results on a dataset of live concert videos to demonstrate the advantage of the proposed approach.