Every teenager knows how quickly a ‘discussion’ with a parent can escalate. Such asymmetric interactions are ubiquitous. Even machines are at it, through high-frequency algorithmic trading. Capturing and understanding the key statistical features of these interacting systems is tricky. Most approaches assume some aspect of the dynamics changes slowly in time – or at least, that they do not show black swan type singularities. However the process of learning, even in its broadest terms, has a definite arrow of time with causality (or not).
In this talk, I summarize some approaches we have taken, and features we have uncovered, in both human and machine systems that are learning in a broad sense. Coming from a physical science background, our focus is not just on developing models that reproduce the non-stationary statistics of such interactions (e.g. infant-parent) but also on developing a generative understanding that can be related back to potential physiological and/or psychological causes.
Neil Johnson heads up a new inter-disciplinary research group in Complexity at the University of Miami looking at collective behavior and emergent properties in a wide range of real-world Complex Systems: from physical, biological and medical domains through to social and financial domains. He is also a Professor of Physics. He obtained his PhD from Harvard University and his MA and BA from Cambridge University. Prior to coming to Miami, he was a professor at Oxford University for 15 years.
VASC Seminar is sponsored by Disney Research.
Host: Kris Kitani
smatvey [atsymbol] cs.cmu.edu