Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cyclic Neural Network

Liangwei Yang, Hengrui Zhang, Zihe Song, Jiawei Zhang, Weizhi Zhang, Jing Ma, Philip S. Yu

TL;DR

This work proposes Cyclic Neural Networks, a graph-structured alternative to traditional layer-by-layer ANNs, inspired by biological connectomes that form complex, cyclic graphs. By introducing computational neurons, synapses, and localized optimization with a forward-forward training regime, the authors develop Graph Over Multi-layer Perceptron (GOMLP) as the first detailed model under this paradigm. Experiments across MNIST, NewsGroup, and IMDB demonstrate that FF-trained, graph-structured networks can surpass or match BP-trained baselines, validating the viability of localized learning and cyclic connectivity. The approach broadens the design space for AI, offering potential gains in biological plausibility, privacy, and parallelism, with practical impact for systems that benefit from graph-based, locally optimized neural architectures.

Abstract

This paper answers a fundamental question in artificial neural network (ANN) design: We do not need to build ANNs layer-by-layer sequentially to guarantee the Directed Acyclic Graph (DAG) property. Drawing inspiration from biological intelligence (BI), where neurons form a complex, graph-structured network, we introduce the groundbreaking Cyclic Neural Networks (Cyclic NNs). It emulates the flexible and dynamic graph nature of biological neural systems, allowing neuron connections in any graph-like structure, including cycles. This offers greater adaptability compared to the DAG structure of current ANNs. We further develop the Graph Over Multi-layer Perceptron, which is the first detailed model based on this new design paradigm. Experimental validation of the Cyclic NN's advantages on widely tested datasets in most generalized cases, demonstrating its superiority over current BP training methods through the use of a forward-forward (FF) training algorithm. This research illustrates a totally new ANN design paradigm, which is a significant departure from current ANN designs, potentially leading to more biologically plausible AI systems.

Cyclic Neural Network

TL;DR

This work proposes Cyclic Neural Networks, a graph-structured alternative to traditional layer-by-layer ANNs, inspired by biological connectomes that form complex, cyclic graphs. By introducing computational neurons, synapses, and localized optimization with a forward-forward training regime, the authors develop Graph Over Multi-layer Perceptron (GOMLP) as the first detailed model under this paradigm. Experiments across MNIST, NewsGroup, and IMDB demonstrate that FF-trained, graph-structured networks can surpass or match BP-trained baselines, validating the viability of localized learning and cyclic connectivity. The approach broadens the design space for AI, offering potential gains in biological plausibility, privacy, and parallelism, with practical impact for systems that benefit from graph-based, locally optimized neural architectures.

Abstract

This paper answers a fundamental question in artificial neural network (ANN) design: We do not need to build ANNs layer-by-layer sequentially to guarantee the Directed Acyclic Graph (DAG) property. Drawing inspiration from biological intelligence (BI), where neurons form a complex, graph-structured network, we introduce the groundbreaking Cyclic Neural Networks (Cyclic NNs). It emulates the flexible and dynamic graph nature of biological neural systems, allowing neuron connections in any graph-like structure, including cycles. This offers greater adaptability compared to the DAG structure of current ANNs. We further develop the Graph Over Multi-layer Perceptron, which is the first detailed model based on this new design paradigm. Experimental validation of the Cyclic NN's advantages on widely tested datasets in most generalized cases, demonstrating its superiority over current BP training methods through the use of a forward-forward (FF) training algorithm. This research illustrates a totally new ANN design paradigm, which is a significant departure from current ANN designs, potentially leading to more biologically plausible AI systems.
Paper Structure (28 sections, 10 equations, 3 figures, 3 tables, 1 algorithm)

This paper contains 28 sections, 10 equations, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Cyclic Neural Network
  3. Computational Neuron
  4. Synapse
  5. Local Optimization
  6. Inference
  7. Advantages of Cyclic NN
  8. Graph Over Multi-layer Perceptron
  9. Input Construction
  10. Computation Graph
  11. Neuron Update
  12. Synapse Propagation
  13. Readout Layer
  14. Local Optimization
  15. Computational Neuron Optimization
  16. ...and 13 more sections

Figures (3)

  • Figure 1: Neuron connection between Biology Neural Network and Artificial Neural Network
  • Figure 2: Comparison between different types of MLP structure.
  • Figure 3: Parameter sensitivity of $T$ and $\theta$