An optimized recursive learning algorithm for three-layer feedforward neural networks for mimo nonlinear system identifications

快权重稳态重入网络中的递归动力学：迈向反思性智能

Recursive Dynamics in Fast-Weights Homeostatic Reentry Networks: Toward Reflective Intelligence

本文提出快权重稳态重入层FH-RL，通过快权重联想记忆、稳态正则化与学习到的重入反馈实现神经网络中的自指计算，实验发现中等重入增益下可涌现稳定且可表达的反思性内部动力学。

An Automated News Bias Classifier Using Caenorhabditis Elegans Inspired Recursive Feedback Network Architecture

Traditional approaches to classify the political bias of news articles have failed to generate accurate, generalizable results. Existing networks premised on CNNs and DNNs lack a model to identify and extrapolate subtle indicators of bias like word choice, context, and presentation. In this paper, we propose a network architecture that achieves human-level accuracy in assigning bias classifications to articles. The underlying model is based on a novel Mesh Neural Network (MNN),this structure enables feedback and feedforward synaptic connections between any two neurons in the mesh. The MNN onta…

SpikingGamma：无替代梯度和时间精确的在线训练spiking神经网络的 smoothed delays

SpikingGamma: Surrogate-Gradient Free and Temporally Precise Online Training of Spiking Neural Networks with Smoothed Delays

SpikingGamma通过内部递归记忆结构和sigma-delta spike编码，实现无替代梯度的在线训练，支持细时间模式学习和高效硬件映射。

Spike time displacement based error backpropagation in convolutional spiking neural networks

We recently proposed the STiDi-BP algorithm, which avoids backward recursive gradient computation, for training multi-layer spiking neural networks (SNNs) with single-spike-based temporal coding. The algorithm employs a linear approximation to compute the derivative of the spike latency with respect to the membrane potential and it uses spiking neurons with piecewise linear postsynaptic potential to reduce the computational cost and the complexity of neural processing. In this paper, we extend the STiDi-BP algorithm to employ it in deeper and convolutional architectures. The evaluation results…