"Generative Modeling of Space, Time, and Objects: A First Step to Endowing Common Sense to Machines", Sungjin Ahn (RUTGERS, Assistant Prof., Dept of Comp. Sci.)
Tuesday, November 19, 2019, 01:00pm - 02:30pm
Busch Campus, Psych 105
Abstract: Human intelligence is based on the ability to build the model of the world in our brain. Learning such world model is mostly unsupervised or self-supervised process. To obtain human-level general intelligence, an artificial agent should also be able to build the model of its environment. However, contemporary AI based on deep learning is lacking this ability and heavily relies on supervised pattern association. As a result, it misses one of the most critical abilities: common sense. In this talk, I present how an artificial agent can build the world model and its structured representation in an unsupervised fashion. In particular, I focus on how to model and represent space, time, and objects, the most critical aspects of the physical world. The presented models are based on probabilistic generative latent-variable modeling and thus can be used in AI as the engine for future imagination. I also discuss how this world model can be incorporated to an artificial agent in order to enable model-based reinforcement learning
Note: If you would like to receive email announcements about the colloquium series, please contact the Business Office to have your name added to our announce lists at firstname.lastname@example.org.