Local Responsibility Allocation in Multi-Layer Perceptrons

Read full text Summarise with VIRAT
Authors: Jia Fujie
Cited by: 0
References: 6

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Traditional multi-layer perceptrons use global backpropagation, applying the same update rule to all connections. This paper proposes an optimization method introducing a local responsibility allocation mechanism. For each connection, a ``responsibility factor'' is calculated based on the distance between its input and a learnable center. This factor modulates both forward signal and backward gradient. Normalization of these factors turns them into a normalized allocation across connections, which serves as a general-purpose limiter: it bounds the neuron's output, stabilizes gradient flow, and allocates responsibility in a deterministic, mathematically derived manner. The rational form of the responsibility function ensures gradual decay, facilitating integration of local and global features. Operating at the connection level, this differs from neuron-level methods like those in RBF networks. This approach provides a novel perspective on credit assignment in neural networks, offering a mechanism for structured and interpretable learning.

Fields of Study & Subjects

Neural Networks and Applications Machine Learning and ELM Stochastic Gradient Optimization Techniques

Publication details

Journal / source
Research Square
Publication date
31 March 2026
Language
EN
Document type
Preprint