Search research articles

Search research articles

Related Experiment Videos

A fast and convergent stochastic MLP learning algorithm.

¹Graduate School of Knowledge Science, Japan Advanced Institute of Science and Technology, Tatsunokuchi, Ishikawa923-1292, Japan. ASakurai@jaist.ac.jp

International Journal of Neural Systems

|February 20, 2002

Summary

This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Chimera Time-Crystalline Order in Quantum Spin Networks.

Physical review letters·2021

Same author

A clinical investigation of the association between perioperative oral management and prognostic nutritional index in patients with digestive and urinary cancers.

Current oncology (Toronto, Ont.)·2020

Same author

Whole-genome sequencing of Staphylococcus epidermidis bloodstream isolates from a prospective clinical trial reveals that complicated bacteraemia is caused by a limited number of closely related sequence types.

Clinical microbiology and infection : the official publication of the European Society of Clinical Microbiology and Infectious Diseases·2019

Same author

Effects of 3-Methylgibberellin Analogs on Gibberellin 3β-Hydroxylases and Plant Growth.

Bioscience, biotechnology, and biochemistry·2016

Same author

Effect of Uniconazole and Gibberellin on the Flowering of Pharbitis nil.

Bioscience, biotechnology, and biochemistry·2016

Same author

[Delayed augmentation effect of cytokine production after hyperthermia stimuli].

Molekuliarnaia biologiia·2015

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

A novel stochastic learning algorithm for multilayer perceptrons demonstrates rapid and reliable convergence for parity and classification tasks. This new method significantly outperforms existing algorithms in speed and efficiency.

Area of Science:

Artificial Intelligence
Machine Learning
Neural Networks

Background:

Multilayer perceptrons (MLPs) with linear-threshold units are fundamental in machine learning.
Efficient training algorithms are crucial for the practical application of MLPs.
Existing algorithms like backpropagation and Levenberg-Marquardt have limitations in speed and convergence.

Purpose of the Study:

To introduce a new stochastic learning algorithm for MLPs composed of linear-threshold units.
To analyze the theoretical convergence properties of the proposed algorithm.
To evaluate the algorithm's experimental performance against established methods.

Main Methods:

Development of a stochastic learning algorithm tailored for linear-threshold function units in MLPs.

Related Experiment Videos

Theoretical analysis to prove convergence with probability one.

Empirical testing on n-bit parity functions (n=2 to 12) and the Thyroid classification dataset from the UCI repository.

Main Results:

The algorithm achieved a 100% experimental convergence rate on tested parity problems.
It demonstrated significant speed improvements, being 5-10 times faster than the Levenberg-Marquardt algorithm for parity tasks.
For the Thyroid dataset, the algorithm showed comparable generalization accuracy to standard backpropagation and Levenberg-Marquardt, while being faster.

Conclusions:

The proposed stochastic learning algorithm offers a theoretically sound and experimentally efficient approach for training MLPs with linear-threshold units.
This method presents a viable and faster alternative to existing training algorithms for specific machine learning tasks.
The algorithm's speed and reliability make it promising for complex parity and classification problems.