The Impact of Interference on Generalization in Feedforward Neural Networks

Title: The Impact of Interference on Generalization in Feedforward Neural Networks

Research Question: How does the interference of signals between different inputs in feedforward neural networks affect the generalization ability of the network?

Methodology: The study uses a feedforward artificial neural network as a model to investigate the generalization process in both artificial and biological neural networks. The interference of signals, or the mutual processing of different inputs, is analyzed and its impact on the generalization ability of the network is examined.

Results: The research found that the interference of signals can lead to random generalization, where the neural network's ability to generalize to unknown data is hindered. This random generalization can be caused by the mutual processing of different inputs, which can result in highly random changes to the network's generalizing function.

Implications: The study suggests that the interference of signals can pose a challenge to the generalization ability of feedforward neural networks. This finding has implications for the design and optimization of neural networks, as it highlights the importance of reducing interference between different inputs to improve generalization. Additionally, the research provides insights into the generalization process in both artificial and biological neural networks, contributing to a better understanding of how these networks process and learn from information.

Link to Article: https://arxiv.org/abs/0310009v2 Authors: arXiv ID: 0310009v2