Organizations: Department of Computer Science and Engineering, Sister Nivedita University, Kolkata, India · Department of Computer Science and Engineering, UPES Dehradun, Dehradun, India
Implementing intelligent learning systems on low- power edge devices is challenging due to the high compu- tational complexity, training latency, and energy demands of traditional Artificial Neural Networks (ANNs). Spiking Neural Networks (SNNs), biologically plausible and spike-driven, come as an alternative. However, SNN training commonly relies on backpropagation through time (BPTT), which is computationally expensive, inherently global, and unsuitable for battery-driven devices. This work proposes a novel local learning framework for multi-layered SNNs inspired by Neural Activity Synchronization (NAS). Each layer updates its own weights using locally com- puted gradients. No gradient backpropagation is needed. This significantly reduces computational complexity, training latency, and energy consumption. The proposed SNN model is trained and evaluated on the MNIST dataset, achieving a classification accuracy of 97.78%. The model is also analyzed with Gaussian and salt-and-pepper noise, demonstrating robust performance across different noise levels.
Figures & tables
Fig. 1: Neural Activity Synchronization based Spiking Neural Network architecture. This figure illustrates the processing of image data at a single time-step after Poisson rate encoding. The data is encoded for T time steps. Here, I represents the input vector dimension, Dn denotes the number of neurons in the n -th hidden layer, and C represents the number of classes. The Red Tick mark indicates the firing of a neuron at that instant.
Fig. 2: Implementation of the Local Neuronal Synchronous Model. The figure illustrates the formation of the basis matrix using periodic function in the proposed architecture. A distinct spatial frequency fc is assigned to each class c , thus they become class-specific periodic basis vectors. For a layer containing D neurons and C classes, the basis matrix B∈RC×D is formed by stacking these basis vectors. The spike count vector SC(ℓ) of the layer, with dimension 1×D , is projected onto the transposed basis matrix B⊤∈RD×C to compute the logits, resulting in a logit vector of dimension 1×C .
Figure 3
Dataset
# Hidden layers
# Neurons
# Parameters
Learning rate
Basis type
Accuracy (%)
MNIST
3
1024-256-64
1,083,338
0.1
Periodic
97.78
Fashion-MNIST
2
2048-1024
3,716,106
0.1
Sinusoidal
88.06
EMNIST
4
1024-512-256-128
1,498,927
0.05
Sinusoidal
73.62
TABLE I: Best performance of Proposed SNN model with corresponding parameters across different Datasets.
Table 5
Paper
Training Method
# Hidden Layers (# Neurons)
Time-Steps
Epochs
Local Learning
Accuracy (%)
CoLaNET [ 14 ]
Supervised STDP with dopamine-modulated plasticity
1 (150)
20
-
-
96.87
CDNA-SNN [ 25 ]
Supervised STDP with CDNA-Weighted STDP
2 (55-38)
300
100
Yes (unsupervised)
98.91
HM2-BP [ 10 ]
Hybrid macro/micro level backpropagation
1 (800)
400
100
No (supervised)
98.93
STDBP [ 27 ]
Spike-Timing-Dependent Backpropagation
1 (800)
-
150
No (supervised)
98.50
BS4NN [ 13 ]
BPTT + Binary synaptic weights
1 (600)
256
500
No (supervised)
97.00
PSAC [ 20 ]
STDP with Reinforcement learning
1 (7010)
-
6
Yes (unsupervised)
98.10
TABLE IV: Comparative analysis of the Proposed SNN architecture and Method against Existing State-of-the-Art SNN approaches on MNIST Dataset.
Paper
Training Method
# Hidden Layers (# Neurons)
Time-Steps
Epochs
Local Learning
Accuracy (%)
CDNA-SNN [ 25 ]
Supervised STDP with CDNA-Weighted STDP
2 (53-46)
300
100
Yes (unsupervised)
90.12
STDBP [ 27 ]
Spike-Timing-Dependent Backpropagation
1 (1000)
-
150
No (supervised)
88.10
BS4NN [ 13 ]
BPTT + Binary synaptic weights
1 (1000)
256
500
No (supervised)
87.30
This work
Local Learning + NAS
2 (2048-1024)
20
100
Yes (supervised)
88.06
TABLE V: Comparative analysis of the Proposed SNN architecture and Method against Existing State-of-the-Art SNN approaches on Fashion-MNIST Dataset.
Spiking neural networks process information through discrete, time-dependent neural events, aiming for efficient computation and biologically inspired learning.
Deploying adaptive intelligence at the edge remains challenging due to the high computational and energy cost of training neural models. Spiking Neural Networks (SNNs) offer a promising alternative, but enabling on-device learning requires hardware-algorithm co-design. This paper presents SPIKER-LL, an FPGA-based SNN accelerator that extends the open-source Spiker+ inference architecture with efficient support for the STSF local learning rule. Through targeted microarchitectural extensions, SPIKER-LL performs inference and online learning with minimal overhead. Across MNIST, F-MNIST, and DIGITS, it achieves up to 93% accuracy, sub-millisecond latency, and less than 0.1 mJ per inference, while remaining DSP-free and highly scalable for edge-FPGA deployments.
Edge intelligence requires models to sense continuously in real time and to keep adapting on-device, all under tight compute, energy, and memory budgets. Although spiking neural networks (SNNs) enable efficient event-driven inference, standard surrogate-gradient backpropagation (BP) serializes updates and blocks ongoing inference. We investigate Bidirectional Spike-Based Distillation (BSD) as an on-device learning principle that lets edge SNNs learn while inferring. BSD couples a stimulus-driven forward network with an independent target-driven reverse network and aligns their intermediate representations through local objectives. Because the two pathways have disjoint computation graphs until alignment, forward inference, reverse inference, and stage-wise updates can run concurrently. On 25 benchmarks spanning the five sensing modalities of SOUL, BSD stays within 3.8 percentage points of matched BP baselines on average, while only its forward branch is needed at deployment. The learned representations also transfer well to few-shot class-incremental learning without replaying past data. By removing the dense floating-point backward chain, BSD reduces the projected training latency to 0.72× and the estimated training energy to 0.36× that of BP.
Yanxun Zhang, Yifei Wang, Changze Lv +6
College of Computer Science and Artificial Intelligence, Fudan University · Shanghai Key Laboratory of Intelligent Information Processing · School of Software Technology, Zhejiang University
Spiking Neural Networks (SNNs) provide a promising framework for energy-efficient and biologically grounded computation; however, scalable learning in deep recurrent architectures with sparse connectivity remains a major challenge. In this work, we propose a structured multi-layer recurrent SNN architecture composed of locally dense recurrent layers augmented with sparse small-world long-range projections to a readout population. The long-range connectivity is largely fixed, preserving routing efficiency and hardware scalability, while synaptic adaptation is performed using strictly local plasticity mechanisms. To enable supervised learning without backpropagation or surrogate gradients, we introduce a biologically motivated learning framework that combines: (i) population-based winner-take-all (WTA) teaching signals at the output layer, (ii) fixed random broadcast alignment feedback pathways, and (iii) low-dimensional modulatory neuron populations that gate synaptic updates through three-factor learning rules with eligibility traces. This design supports deep recurrent computation with sparse global communication and purely local synaptic updates. We analyze the algorithmic properties, computational complexity, and hardware feasibility of the proposed approach, and demonstrate stable learning and competitive performance on benchmark classification tasks. The results highlight the potential of structured recurrence and neuromodulatory learning to enable scalable, hardware-compatible SNN training beyond gradient-based methods.
Bo Tang, Weiwei Xie
Department of Electrical and Computer Engineering Worcester Polytechnic Institute Worcester, USA