Automating Behavior Labeling in Video

The Center for Biological and Computational Learning (CBCL) is developing systems for automated recognition of behavior. From a computational model of the primate visual cortex, we developed a computer vision for recognizing 8 stereotypical behaviors in mice including grooming, eating, and walking (see cbcl.mit.edu/software-datasets/hueihan/. ) Building this system required over 200 hours of human annotation time and a painstaking process of managing and training the annotators.

Starting in IAP 2011, we will have a position available for a UROP student to develop a new system for automating behavioral analysis. We would like to build a Mechanical Turk interface to automate the labeling of behavior in video. The research challenge is to build a robust annotation system on top of crowdsourced annotators.

Prerequisites: The ideal candidate will have initiative, strong web programming skills, and interest in crowdsourced applications.

Contact: Please send questions or application with resume/CV to Prof. Tomaso Poggio(tp_at_csail.mit.edu).