Paper: "Spatiotemporal Connectionist Networks: A Taxonomy and Review" (download)

Kremer, 2001, Neural Computation 13 :249-306.

While connectionist modeling has illuminated several areas of neuroscience e.g. hippocampal learning and memory and visual object recognition, most models still operate in 'batch mode,' receiving one input at a time. How can we model tasks such as processing of speech and language and planning, where the relevant inputs are time series and where short-term memory must play a critical role? To solve these problems, 'spatio-temporal connectionist networks' (STCNs) have been devised. What are STCNs, and can they solve some of the most difficult computational problems facing neural systems?

1:30pm -3:00pm, HSE 810.