Related Experiment Videos
The backpropagation algorithm in J, a fast prototyping tool for researching neural networks.
1Department of Computing Science, University College of the Cariboo, Kamloops, British Columbia, Canada. rbrouwer@cariboo.bc.ca
International Journal of Neural Systems
|December 10, 1999
Summary
This study demonstrates how the J language simplifies neural network simulation, particularly for multilayer perceptrons using gradient descent. J offers flexibility and reduces development time for complex algorithms like back-propagation.
Area of Science:
- Computer Science
- Artificial Intelligence
- Computational Neuroscience
Background:
- Neural network simulations are crucial for understanding brain function and developing AI.
- Existing tools like MATLAB can be restrictive for complex neural network architectures.
- The back-propagation algorithm is a fundamental method for training multilayer neural networks.
Purpose of the Study:
- To illustrate the utility of the J programming language for simulating neural networks.
- To demonstrate J's application in training multilayer perceptrons via gradient descent.
- To show the ease of generalizing back-propagation within the J environment.
Main Methods:
- Utilizing the J programming language for its powerful array notation and operators.
- Applying J to implement a gradient descent training method for a multilayer perceptron.
- Expressing the back-propagation algorithm in J's array notation for direct execution.
Main Results:
- The J language provides a powerful and flexible environment for neural network simulation.
- Gradient descent training of multilayer perceptrons is effectively demonstrated using J.
- Back-propagation algorithm generalization to multilayer networks is achieved with simplicity in J.
- J's succinct code and array notation significantly decrease development time.
Conclusions:
- The J language is highly suitable for neural network research and development.
- J offers a flexible alternative to specialized simulators and software packages like MATLAB.
- The use of J can lead to substantial reductions in the time and effort required for developing neural network models.