- This event has passed.
Condensed Matter/Biophysics Seminars – BingKan Xue (UF Physics)
Associative memory, attractor dynamics, and neural networks
I will describe some basic ideas in the theoretical study of neural networks, originating from the classic work of Hopfield. The Hopfield model provides an abstract characterization of associative memory, the ability of a neural network to retrieve stored patterns. In this model, memories are encoded as attractors of the network dynamics. The analogy to statistical mechanics allows us to understand the behavior of the network, whereas the biological context brings new types of questions, such as the network’s capacity for storing different memories. The idea has since then been developed and generalized in many ways. I will describe some recent developments that incorporate temporal structures, such as transition between attractors and recognition of dynamic patterns, or spatial structures, such as spatial maps of environments and assembly of macro-molecules.
Host: Steve Hagen