Title: An exact learning algorithm for autoassociative neural networks with binary couplings

Abstract: Exact solutions for the learning problem of autoassociative networks with binary couplings are determined by a new method. The use of a branch-and-bound algorithm leads to a substantial saving of computational time compared with complete enumeration. As a result, fully connected networks with up to 40 neurons could be investigated. The network capacity is found to be close to 0.83.