Related Experiment Videos
On the capabilities of higher-order neurons: a radial basis function approach
1Lehrstuhl Mathematik und Informatik, Fakultät für Mathematik, Ruhr-Universität Bochum, D-44780 Bochum, Germany. mschmitt@lmi.ruhr-uni-bochum.de
Neural Computation
|April 2, 2005
Summary
Higher-order neurons with k monomials in n variables achieve a Vapnik-Chervonenkis (VC) dimension of at least nk + 1. This finding improves upon previous bounds and indicates a greater VC dimension than k-term monotone disjunctive normal form (DNF) formulas.
Area of Science:
- Machine Learning
- Computational Learning Theory
- Neural Networks
Background:
- The Vapnik-Chervonenkis (VC) dimension is a key measure of a model's capacity in statistical learning theory.
- Previous research established lower bounds for the VC dimension of higher-order neurons using k-term monotone disjunctive normal form (DNF) formulas.
Purpose of the Study:
- To establish a tighter lower bound for the VC dimension of higher-order neurons with k monomials in n variables.
- To compare the VC dimension of higher-order neurons with that of k-term monotone DNF formulas.
Main Methods:
- An exponential approach was introduced to analyze the VC dimension.
- Gaussian radial basis function neural networks were employed for point classification.
Main Results:
- A new lower bound of nk + 1 for the VC dimension of higher-order neurons with k monomials in n variables was proven.
- The established lower bound supersedes previously known bounds derived from k-term monotone DNF formulas.
- It was demonstrated that the VC dimension of higher-order neurons with k monomials is strictly greater than that of k-term monotone DNF.
Conclusions:
- The study provides a significant advancement in understanding the capacity of higher-order neurons.
- The findings have implications for the design and analysis of neural networks and learning algorithms.
- The established VC dimension bound offers a more accurate assessment of model complexity for higher-order neurons.