Bidirectional Clustering of MLP Weights for Finding Nominally Conditioned Polynomials

Authors:
Yusuke Tanahashi;Ryohei Nakano
Affiliations:
Nagoya Institute of Technology, Nagoya, Japan 466-8555;Chubu University, Kasugai, Japan 487-8501
Venue:
ICANN '09 Proceedings of the 19th International Conference on Artificial Neural Networks: Part II
Year:
2009

Citing 7
Cited 1

Integrating quantitative and qualitative discovery in the ABACUS system

Machine learning
C4.5: programs for machine learning

C4.5: programs for machine learning
Hierarchical mixtures of experts and the EM algorithm

Neural Computation
Extracting regression rules from neural networks

Neural Networks
X-means: Extending K-means with Efficient Estimation of the Number of Clusters

ICML '00 Proceedings of the Seventeenth International Conference on Machine Learning
Model selection and weight sharing of multi-layer perceptrons

KES'05 Proceedings of the 9th international conference on Knowledge-Based Intelligent Information and Engineering Systems - Volume Part IV
Extraction of rules from artificial neural networks for nonlinear regression

IEEE Transactions on Neural Networks

Nominally conditioned linear regression

ICANN'10 Proceedings of the 20th international conference on Artificial neural networks: Part III

Quantified Score

Hi-index	0.00

Visualization

Abstract

We present a method for finding nominally conditioned polynomials to fit multivariate data containing both numeric and nominal variables. Here a polynomial is accompanied with a nominal condition stating when the polynomial is applied. Our method employs a four-layer perceptron (MLP) having shared weights. To get succinct polynomials, we employ weight sharing method called BCW, where each weight is allowed to be one of common weights, and a near-zero common weight can be eliminated. BCW performs bidirectional search to obtain an excellent set of common weights. Moreover, we employ the Bayesian Information Criterion (BIC) to efficiently select the optimal model parameters. In our experiments the proposed method successfully restored the original polynomials for artificial data, and found succinct polynomials for real data sets, showing excellent generalization.