Title :
MDL learning of probabilistic neural networks for discrete problem domains
Author :
Tirri, Henry ; Myliymaki, P.
Author_Institution :
Dept. of Comput. Sci., Helsinki Univ., Finland
fDate :
27 Jun-2 Jul 1994
Abstract :
Given a problem, a case-based reasoning (CBR) system will search its case memory and use the stored cases to find the solution, possibly modifying retrieved cases to adapt to the required input specifications. In discrete domains CBR reasoning can be based on a rigorous Bayesian probability propagation algorithm. Such a Bayesian CBR system can be implemented as a probabilistic feedforward neural network with one of the layers representing the cases. In this paper we introduce a minimum description length (MDL) based learning algorithm to obtain the proper network structure with the associated conditional probabilities. This algorithm together with the resulting neural network implementation provide a massively parallel architecture for solving the efficiency bottleneck in case-based reasoning
Keywords :
case-based reasoning; feedforward neural nets; learning (artificial intelligence); probability; case-based reasoning system; discrete domains; discrete problem domains; efficiency bottleneck; massively parallel architecture; minimum description length based learning algorithm; probabilistic feedforward neural network; probabilistic neural networks; rigorous Bayesian probability propagation algorithm; search; Bayesian methods; Computer architecture; Computer science; Concurrent computing; Feedforward neural networks; Indexing; Learning systems; Neural networks; Parallel architectures; Parallel processing;
Conference_Titel :
Neural Networks, 1994. IEEE World Congress on Computational Intelligence., 1994 IEEE International Conference on
Conference_Location :
Orlando, FL
Print_ISBN :
0-7803-1901-X
DOI :
10.1109/ICNN.1994.374508
