A Silicon Primitive for Competitive Learning

Competitive learning is a technique for training classification and clustering networks. We have designed and fabricated an 11transistor primitive, that we term an automaximizing bump circuit, that implements competitive learning dynamics. The circuit performs a similarity computation, affords nonvolatile storage, and implements simultaneous local adaptation and computation. We show that our primitive is suitable for implementing competitive learning in VLSI, and demonstrate its effectiveness in a standard clustering task.

• Type: Conference
• Year: 2000
• Where: NIPS
• Authors: David Hsu, Miguel Figueroa, Chris Diorio
• Keywords: Affords Nonvolatile Storage, Competitive Learning, Competitive Learning Dynamics, NIPS 2000, NIPS 2007