Last Friday, Dr Ron Greene of physics dept. of University of new Orleans gave us a very nice seminar on the efficient retrieval from sparse associative memory.

At first, Dr Greene introduced the motivation: memory-based reasoning, associate memory.

Then Dr Greene presented the sparse associate memory: feature vectors, sparse memory; and implementation: connectionist (neural network) approach, conventional approaches.

Then Dr Greene introduced the connectionist-Hashed associative memory (CHAM), connectionist network Hash function, fixed network example, CHAM with fixed network: operation (storage, retrieve), best fit search, simple generalization.

Then Dr Greene presented some tables of probability of retrieving an item from memory, etc.

Then Dr Greene introduced the enhancements to the basic CHAM: 1. increase the number of network outputs; 2. using multiple hashing network; 3. select search.

Then Dr Greene presented the results: random feature vectors, correlated feature vectors.

At last Dr Greene introduced the CHAM conclusion: 1. faster than the linear probabilistic best-match retrieves; 2, handles noise or partially specified data; 3. easily expanded capacity.

Last Update: 5/7/98
Web Author: Zizhong Wang
The report is for Dr Paprzycki@ marcin.paprzycki@ibspan.waw.pl or@ Home Page