Part of Advances in Neural Information Processing Systems 11 (NIPS 1998)
Bradley C. Love
Historically, connectionist systems have not excelled at represent(cid:173) ing and manipulating complex structures. How can a system com(cid:173) posed of simple neuron-like computing elements encode complex relations? Recently, researchers have begun to appreciate that rep(cid:173) resentations can extend in both time and space. Many researchers have proposed that the synchronous firing of units can encode com(cid:173) plex representations. I identify the limitations of this approach and present an asynchronous model of binding that effectively rep(cid:173) resents complex structures. The asynchronous model extends the synchronous approach. I argue that our cognitive architecture uti(cid:173) lizes a similar mechanism.