Deep Decision Trees for Discriminative Dictionary Learning with Adversarial Multi-agent Trajectories

Abstract

With the explosion in the availability of spatio-temporal tracking data in modern sports, there is an enormous opportunity to better analyse, learn and predict important events in adversarial group environments. In this paper, we propose a deep decision tree architecture for discriminative dictionary learning from adversarial multi-agent trajectories. We first build up a hierarchy for the tree structure by adding each layer and performing feature weight based clustering in the forward pass. We then fine tune the player role weights using back propagation. The hierarchical architecture ensures the interpretability and the integrity of the group representation. The resulting architecture is a decision tree, with leaf-nodes capturing a dictionary of multiagent group interactions. Due to the ample volume of data available, we focus on soccer tracking data, although our approach can be used in any adversarial multi-agent domain. We present applications of proposed method for simulating soccer games as well as evaluating and quantifying team strategies.

Memory Augmented Deep Generative models for Forecasting the Next Shot Location in Tennis

“…However recent studies in the sports prediction field [15] have demonstrated the importance of learning the underlying feature distribution in an automatic fashion. For instance in [15] the authors learn a dictionary of player formations in soccer for classifying the outcome of a shot.…”

“…However recent studies in the sports prediction field [15] have demonstrated the importance of learning the underlying feature distribution in an automatic fashion. For instance in [15] the authors learn a dictionary of player formations in soccer for classifying the outcome of a shot. Even though they achieve comprehensive advancement towards automatic feature learning with player trajectories, those systems cannot be directly applied to model player strategies in tennis.…”

“…Current state-of-the-art methodologies try to capture the underlying semantics through a handful of handcrafted features, without paying attention to essential mechanisms in the human brain, where the expertise and observations are stored and knowledge is derived. It is a broadly established fact that handcrafted features only capture abstract level semantics in a given environment [11], [12], [13], [14] and it is proven that these ill represent the • T. Fernando context in several data mining and knowledge discovery tasks [15]. The goal of this paper is to derive a deep learning model to anticipate the next shot location in tennis, given current match context and a short history of player and ball behaviour from Hawk-eye ball tracking data [16].…”

Center point to pose: Multiple views 3D human pose estimation for multi-person

“…3D human pose estimation in crowded scenes is an important component of computer vision applications such as autonomous driving [ 1 ], surveillance [ 2 ], robotics [ 3 ] and human-computer interaction [ 4 ]. When single view is used for human pose estimation, although it is widely applicated in other object-detection, the lack of depth information and occlusion directly results the incomplete pose estimation.…”