Introducing Influence Based Learning, A New Algorithm for Neural Networks

Share Article

Based On A Simple and Intuitive Concept, Influence Learning Resolves Many of the Most Important Limitations of Existing Neural Network Learning Algorithms, Including the Elimination of All Restrictions On Neural Network Feedback

A new learning method called Influence Learning is among a variety of innovations described in an upcoming book about neural networks. The book is titled: "Netlab Loligo: New Approaches To Neural Network Simulation". The book rejects the idea of simply adding more layers of computational complexity to cover over shortcomings of existing solutions. It distinguishes itself from past books on artificial intelligence (a.i.) by presenting a set of unique "new ideas" for the design and construction of neural networks. This underlying philosophy has lead to some advantageous new tools.

Influence learning, for example, is one of two new learning methods introduced in the book. It is based on the simple notion that our own attraction to people who exercise influence (and sometimes our repulsion by it) might have its origins within the inter-cellular communications mechanisms employed by our brains' neurons. Good or bad, there is a clear human tendency to trust influential people and use them as role-models for adapting our own behavior. Influence Learning extrapolates this observation about human behavior back, in order to explain some of the observed behaviors of individual neurons. This notion is then employed as a model for how neurons are able to independently adapt their connection-maps within deep structures of the brain.

There are many benefits of this influence based approach to learning, one of which is simplicity. The method is not merely a new layer of computational complexity, which has been added to correct for deficiencies in an existing formula. Instead, it provides a self-evident mechanism with easily perceived reasons for how and why it works.

Its intrinsic simplicity leads to many other advantages as well. The bane of many existing neural network learning models has been the severe restrictions they place on how feedback may be specified within neural network designs. Such restrictions are completely eliminated in designs that use influence learning. Also, influence learning has no dependence on network-global calculations, and so may be used for neural networks that are spread over many, asynchronously running processors.

There are many other benefits afforded by this simply but powerful new concept of neural learning based on attraction and aversion to influence exercised by other neurons. More details about Influence Learning can be found in the description posted at the author's blog for the book.

This new learning algorithm, which is patent pending, is described in detail in the book: "Netlab Loligo: New Approaches to Neural Network Simulation", which is available now from, or from the book-description page at the publisher's web site. (

Title: Netlab Loligo: New Approaches to Neural Network Simulation
ISBN: 978-0984425600
Paperback: 332 pages


Share article on social media or email:

View article via:

Pdf Print

Contact Author

John Repici
Visit website