A Subtler Silicon Cell for Neural Networks

See allHide authors and affiliations

Science  26 Sep 1997:
Vol. 277, Issue 5334, pp. 1935
DOI: 10.1126/science.277.5334.1935

You are currently viewing the summary.

View Full Text

Log in to view the full text

Log in through your institution

Log in through your institution


Neural nets traditionally consist of so-called sigmoidal neurons, circuits that add up incoming signals until they reach a fixed threshold and then fire themselves. But a University of Pennsylvania research has now developed so-called bifurcation neurons, which switch between different modes of operation depending on subtler factors, including the interaction between many incoming signals and the neuron's recent history. This work could yield neural nets with more complex behavior than has been seen in networks to date, such as the ability to see, recognize, and even react to the world in real time.