论文信息 - Errors correction with optimised Hopfield neural networks

Errors correction with optimised Hopfield neural networks

We present in this paper a method of increasing both the storage capacity of Hopfield neural networks and their capability of error correction. The presented method uses the general principles of generating error-correcting codes in information theory combined with a gradient - an heuristic algorithm. Using this method, there are registered improvements in the growth of network storage capacity (number of words memorized), and also in the increase of the correct answers' probability. These types of networks can be used in several applications which use associativity, including the correction errors in communication, the image reconstruction and the object recognition.

Laurentiu Ionescu | Gheorghe Serban | Alin Mazare | Constantin Anton | Ion Tutanescu

[1] Matthias Löwe,et al. The storage capacity of the Hopfield model and moderate deviations , 2005 .

[2] Patrick Simen,et al. Hebbian learning in linear-nonlinear networks with tuning curves leads to near-optimal, multi-alternative decision making , 2011, Neural Networks.

[3] Raúl Rojas,et al. Neural Networks - A Systematic Introduction , 1996 .

[4] Adriano Barra,et al. On the equivalence of Hopfield networks and Boltzmann Machines , 2011, Neural Networks.

[5] Jose F. Monserrat,et al. Convergence and Stability of Quantized Hopfield Networks Operating in a Fully Parallel Mode , 2010, Neural Computation.

[6] Yuko Osana,et al. Kohonen feature map associative memory with area representation for sequential analog patterns , 2006, 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence).