The Time Dimension for Neural Computation
Abstract
The binding problem refers to how sensory elements organize into perceived objects. The issue of binding is hotly debated in recent years in neuroscience and related communities. Much of the debate, however, gives little attention to computational considerations a rather curious status asthe problem is originally formulated from the computational perspective. This article starts withtwo problems considered by Rosenblatt to be the most challenging to the development of perceptron theory 40 years ago, and argues that the main challenge is the figure-ground separation problem, which is intrinsically related to the binding problem. The central claim of the article is that introducing the time dimension is essential for systematically attacking Rosenblatts challenge.The temporal correlation theory as well as its special form - oscillatory correlation theory, is discussed as an adequate representation theory to address the binding problem in neural computation. A computational mechanism for the oscillatory correlation theory - LEGIONdynamics - provides a solution to the Minsky-Papert connectedness problem, which is an important example of the binding problem, and the mechanism is successfully applied to a variety of scene segmentation tasks. The plausibility and implication of the oscillatory correlation theory are discussed at the physiological, perceptual, and cognitive levels. A number of controversial issues regarding oscillatory correlation are considered and clarified. Finally, the time dimension is argued to be necessary for versatile computation.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2002
- Accession Number
- AD1001149
Entities
People
- DeLiang Wang
Organizations
- Ohio State University