Feature Learning in Video-Based Analysis of Animal Emotional States

There is a growing interest in animal-related disciplines and in ACI in applying AI-based approaches for studying animal behavior. Most efforts have so far been focused on automatic tracking of animal movement, and recognition of some basic postures, behaviors and activities. However, AI has great potential not only for automatically "seeing"the animal, but also for "understanding"and even "explaining"its behavior. More specifically, deep learning techniques have an increasingly efficient performance in a variety of tasks related to image and video classification. In this research we wish to take these techniques one step further, harnessing their key idea of feature learning in the context of animal behavior. Namely, by investigating the learnt features of deep learning classifiers, we provide a method for extracting insights on how classification is performed, hopefully leading to new insights into animal behavior.