dynamic gating fusion

This work addresses the lack of a unified computational interpretation for neural policy gating mechanisms. We propose GateMod, a theoretically grounded gating framework that couples task structure with neural circuit dynamics via the principle of free-energy minimization. GateMod comprises two core components: GateFlow—a continuous-time energy-flow model—and GateNet—a soft-competitive recurrent network—enabling emergent gating for skill composition and behavioral planning. We formally prove GateMod’s global exponential convergence and robustness under perturbations. Empirically, GateMod achieves significant performance gains over state-of-the-art methods in multi-agent cooperative tasks and human multi-armed bandit experiments. Crucially, it provides the first quantitative demonstration of how task structure modulates gating behavior through neural energy dynamics. By offering a computationally precise and empirically testable account, GateMod establishes a principled theoretical foundation for understanding strategy selection in prefrontal–basal ganglia circuits.

This work addresses the prevalent low-frequency bias in lightweight image classification models by systematically analyzing the impact of gating mechanisms on neural network training dynamics from a frequency-domain perspective. We establish, for the first time, a theoretical frequency-domain interpretation of gating operations—specifically, the coupled element-wise multiplication and nonlinear activation—revealing their collaborative modulation of multi-frequency components. Guided by this analysis, we propose GmNet, a lightweight architecture that minimizes low-frequency bias via a frequency-sensitive information flow control structure, overcoming the empirical limitations of conventional gating designs. Leveraging convolution theorem-based frequency-domain insights for principled model design, GmNet achieves superior accuracy and inference efficiency with fewer parameters on benchmarks including ImageNet, significantly outperforming state-of-the-art lightweight models such as MobileNetV3 and EfficientNet-Lite.

Traditional leaky integrate-and-fire (LIF) neuron models neglect the dynamic conductance mechanisms inherent in biological neurons, limiting the robustness and computational capability of spiking neural networks (SNNs) under noise perturbations and temporal variations. To address this, we propose the Dynamic Gating Neuron (DGN) model, which— for the first time—formulates dynamic conductance as a biologically interpretable, adaptive gating mechanism to enable selective information flow regulation and perturbation suppression. The DGN model ensures theoretical stability while maintaining neuroscientific plausibility and supports end-to-end training via backpropagation through time. Evaluated on standard temporal benchmarks—including TIDIGITS and the Spiking Heidelberg Digits (SHD) dataset—DGN consistently outperforms LIF across multiple dimensions: noise robustness, temporal precision, and generalization performance. This work establishes a novel paradigm for designing SNNs that are both highly robust and biologically grounded.

This work addresses the lack of a unified formulation among conventional activation functions and their hardware deployment limitations imposed by ADC/DAC bottlenecks. The authors propose Threshold Gating (TG) as a universal primitive for neural nonlinearities, employing an input-conditioned branch gating mechanism that subsumes major activation functions as special cases. Building on this unification, they introduce the “Minimal Branch Theorem,” enabling lossless activation function conversion and seamless model transfer across diverse architectures—such as CNNs, Transformers, and RNNs—without retraining. The framework supports both training from scratch and post-hoc adaptation, offering benefits in model compression, performance enhancement, and training acceleration. Furthermore, it maps efficiently onto analog in-memory computing architectures, substantially reducing power consumption and area overhead.

To address the challenge of simultaneously achieving biological plausibility, hardware efficiency, and competitive performance in spiking neural networks (SNNs) for general supervised classification, this paper proposes a columnar hierarchical SNN architecture tailored for classification. It employs intra-class-difference-driven columnar organization—each column represents a discriminative subcategory—and adopts an all-spiking signal flow with functionally specialized neurons. A biologically grounded learning mechanism is introduced, integrating local anti-Hebbian plasticity with dopamine neuromodulation to replace backpropagation entirely. The method unifies model-driven reinforcement learning with a state-proximity evaluation framework, enabling end-to-end training directly in the spike domain. Experiments demonstrate that the architecture achieves high accuracy and strong generalization across multiple benchmark classification tasks, while exhibiting exceptional compatibility with low-power neuromorphic hardware. This work establishes a novel paradigm for practical, deployable SNNs.

This work proposes a biologically plausible multilayer neuronal network model that moves beyond the simplified weighted-sum neuron paradigm of conventional artificial neural networks, which struggles to capture the true learning mechanisms of biological systems. The proposed architecture employs cascaded adaptive combiners to enable efficient online, streaming learning without relying on backpropagation. By integrating neuron-level computations that more closely mirror biological reality with practical machine learning principles, the model establishes a concise and scalable learning framework. Empirical evaluation on image classification tasks demonstrates competitive performance, confirming both the efficacy and practicality of the approach.

This work addresses the limitations of existing spiking neurons—namely insufficient performance, poor adaptability, and low training efficiency—in large-scale vision and language tasks. The authors propose Adaptive Spiking Neurons (ASN) and their normalized variant (NASN), the first neuron designs systematically guided by a functional perspective. By incorporating learnable membrane potential dynamics, an integer-based training–spiking inference paradigm, and normalization mechanisms, ASN significantly enhances training stability and model generalizability. Comprehensive experiments across five task categories and nineteen datasets spanning both vision and language domains demonstrate that the ASN family exhibits strong generalization capabilities and holds substantial promise as a universal spiking neuron model.

This work addresses the challenges of deploying conventional artificial neural networks on edge devices—namely, their dense global connectivity, catastrophic forgetting, and opaque decision-making—by introducing Decomposable Spiking Neural Networks (D-SNNs). Inspired by the modular organization of biological nervous systems, D-SNNs structurally isolate classification pathways into independent expert modules and employ a biologically inspired push-pull loss function for optimization. The proposed approach achieves accuracy comparable to dense networks on MNIST, Fashion-MNIST, and CIFAR-10/100 benchmarks while reducing model parameters by an order of magnitude and decreasing spike rates and synaptic operations by several orders of magnitude. Moreover, D-SNNs inherently resist catastrophic forgetting and enable auditable decision processes, offering a promising pathway toward efficient, interpretable, and robust neuromorphic computing on resource-constrained hardware.

研究通过在多层感知机中引入生物启发机制如稳态、结构稀疏化等，调节神经元活动和有效连接性，以促进从记忆到泛化的转变。

This work addresses the limitations of conventional deep learning—namely its reliance on backpropagation, fixed computational graphs, and synchronous inter-layer processing—by introducing a novel neural computing paradigm that does not require differentiable architectures. The proposed approach constructs an asynchronous, temporally structured graph network with shared neurons, where signals propagate through a shared neuron pool via delay modulation. Network topology, connection weights, and propagation delays are co-evolved using a genetic algorithm, eliminating the need for backpropagation. The model enables sample-adaptive computational depth and automatically discovers lateral interactions among pathways. Evaluated on a ten-class MNIST classification task using frozen ResNet18 features, the evolved architecture achieves 85.9% test accuracy after 10,000 generations with a model size of only 115 KB, demonstrating both the efficacy and representational capacity of the proposed framework.