Automotive Object Detection via Learning Sparse Events by Spiking Neurons DOI Creative Commons
Hu Zhang, Yanchen Li, Luziwei Leng

et al.

arXiv (Cornell University), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 1, 2023

Event-based sensors, distinguished by their high temporal resolution of 1$\mathrm{μs}$ and a dynamic range 120$\mathrm{dB}$, stand out as ideal tools for deployment in fast-paced settings like vehicles drones. Traditional object detection techniques that utilize Artificial Neural Networks (ANNs) face challenges due to the sparse asynchronous nature events these sensors capture. In contrast, Spiking (SNNs) offer promising alternative, providing representation is inherently aligned with event-based data. This paper explores unique membrane potential dynamics SNNs ability modulate events. We introduce an innovative spike-triggered adaptive threshold mechanism designed stable training. Building on insights, we present specialized spiking feature pyramid network (SpikeFPN) optimized automotive detection. Comprehensive evaluations demonstrate SpikeFPN surpasses both traditional advanced ANNs enhanced attention mechanisms. Evidently, achieves mean Average Precision (mAP) 0.477 {GEN1 Automotive Detection (GAD)} benchmark dataset, marking significant increase 9.7\% over previous best SNN. Moreover, efficient design ensures robust performance while optimizing computational resources, attributed its innate computation capabilities.

Language: Английский

SIBoLS: Robust and Energy-Efficient Learning for Spike-Based Machine Intelligence in Information Bottleneck Framework DOI
Shuangming Yang, Haowen Wang, Badong Chen

et al.

IEEE Transactions on Cognitive and Developmental Systems, Journal Year: 2023, Volume and Issue: 16(5), P. 1664 - 1676

Published: Nov. 6, 2023

Spike-based machine intelligence has recently attracted increasing research attention, and been considered as a promising approach towards artificial general (AGI). It applied in energy-efficient neuromorphic computing systems. One of the most critical questions for spike-based learning is how to leverage powerful information-theoretic theories derive algorithms improving robustness energy efficiency spiking neural networks (SNNs). In this study, we first present an efficient effective information bottleneck (IB) framework training SNN, named Information Bottleneck with Learnable State (SIBoLS). We thoroughly explore design space concerning IB by using membrane potential state hidden representation learnable variable. Comprehensive test conducted, which two types background noise five input are considered. shows SIBoLS can improve both static image event-based dataset processor. Furthermore, induces less rate, resulting lower power consumption compared other techniques. advantages terms applications, give insights development AGI.

Language: Английский

Citations

40
A Spatial–Channel–Temporal-Fused Attention for Spiking Neural Networks DOI

Wuque Cai,

Hongze Sun, Rui Liu

et al.

IEEE Transactions on Neural Networks and Learning Systems, Journal Year: 2023, Volume and Issue: 35(10), P. 14315 - 14329

Published: May 31, 2023

Spiking neural networks (SNNs) mimic brain computational strategies, and exhibit substantial capabilities in spatiotemporal information processing. As an essential factor for human perception, visual attention refers to the dynamic process selecting salient regions biological vision systems. Although mechanisms have achieved great success computer applications, they are rarely introduced into SNNs. Inspired by experimental observations on predictive attentional remapping, we propose a new spatial-channel–temporal-fused (SCTFA) module that can guide SNNs efficiently capture underlying target utilizing accumulated historical spatial–channel present study. Through systematic evaluation three event stream datasets (DVS Gesture, SL-Animals-DVS, MNIST-DVS), demonstrate SNN with SCTFA (SCTFA-SNN) not only significantly outperforms baseline (BL-SNN) two other models degenerated modules, but also achieves competitive accuracy existing state-of-the-art (SOTA) methods. Additionally, our detailed analysis shows proposed SCTFA-SNN model has strong robustness noise outstanding stability when faced incomplete data, while maintaining acceptable complexity efficiency. Overall, these findings indicate incorporating appropriate cognitive of may provide promising approach elevate

Language: Английский

Citations

14
SSEFusion: Salient semantic enhancement for multimodal medical image fusion with Mamba and dynamic spiking neural networks DOI
Shiqiang Liu, Weisheng Li, Dan He

et al.

Information Fusion, Journal Year: 2025, Volume and Issue: unknown, P. 103031 - 103031

Published: Feb. 1, 2025

Language: Английский

Citations

0
TR-SNN: a lightweight spiking neural network based on tensor ring decomposition DOI Creative Commons
Shifeng Mao, Baoxin Yang, Hongze Sun

et al.

Brain-Apparatus Communication A Journal of Bacomics, Journal Year: 2025, Volume and Issue: 4(1)

Published: Feb. 27, 2025

Language: Английский

Citations

0
BISNN: bio-information-fused spiking neural networks for enhanced EEG-based emotion recognition DOI
Hongze Sun, Shifeng Mao,

Wuque Cai

et al.

Cognitive Neurodynamics, Journal Year: 2025, Volume and Issue: 19(1)

Published: March 22, 2025

Language: Английский

Citations

0
Memristive leaky integrate-and-fire neuron and learnable straight-through estimator in spiking neural networks DOI
Tao Chen, Chunyan She, Lidan Wang

et al.

Cognitive Neurodynamics, Journal Year: 2024, Volume and Issue: 18(5), P. 3075 - 3091

Published: June 20, 2024

Language: Английский

Citations

2
Reliable object tracking by multimodal hybrid feature extraction and transformer-based fusion DOI
Hongze Sun, Rui Liu,

Wuque Cai

et al.

Neural Networks, Journal Year: 2024, Volume and Issue: 178, P. 106493 - 106493

Published: June 28, 2024

Language: Английский

Citations

2
Automotive Object Detection via Learning Sparse Events by Spiking Neurons DOI
Hu Zhang, Yanchen Li, Luziwei Leng

et al.

IEEE Transactions on Cognitive and Developmental Systems, Journal Year: 2024, Volume and Issue: unknown, P. 1 - 15

Published: Jan. 1, 2024

Event-based sensors, distinguished by their high temporal resolution of 1 µs and a dynamic range 120 dB, stand out as ideal tools for deployment in fast-paced settings like vehicles drones.Traditional object detection techniques that utilize Artificial Neural Networks (ANNs) face challenges due to the sparse asynchronous nature events these sensors capture.In contrast, Spiking (SNNs) offer promising alternative, providing representation is inherently aligned with event-based data.This paper explores unique membrane potential dynamics SNNs ability modulate events.We introduce an innovative spike-triggered adaptive threshold mechanism designed stable training.Building on insights, we present specialized spiking feature pyramid network (SpikeFPN) optimized automotive detection.Comprehensive evaluations demonstrate SpikeFPN surpasses both traditional advanced ANNs enhanced attention mechanisms.Evidently, achieves mean Average Precision (mAP) 0.477 GEN1 Automotive Detection (GAD) benchmark dataset, marking significant increases over selected SNN baselines.Moreover, efficient design ensures robust performance while optimizing computational resources, attributed its innate computation capabilities.Source codes are publicly accessible at https://github.com/EMI-Group/spikefpn.

Language: Английский

Citations

2
Surrogate gradient scaling for directly training spiking neural networks DOI
Tao Chen, Shu Wang, Yu Gong

et al.

Applied Intelligence, Journal Year: 2023, Volume and Issue: 53(23), P. 27966 - 27981

Published: Sept. 20, 2023

Language: Английский

Citations

3
Improving Graph Collaborative Filtering via Spike Signal Embedding Perturbation DOI
Ying Ma, Gang Chen, Guoqi Li

et al.

IEEE Transactions on Cognitive and Developmental Systems, Journal Year: 2023, Volume and Issue: 16(5), P. 1688 - 1697

Published: Dec. 4, 2023

Nowadays, graph collaborative filtering has proven to be a highly effective method in recommendation systems. It learns user preferences through interactions between users and items. During the training process of filtering, introducing suitable perturbations is crucial model training. commonly used prevent overfitting enhance robustness. Perturbation widely adopted as data augmentation technique systems extensively contrastive learning. Contrastive learning involves multi-task aimed at various views from diverse augmentations. However, these tasks can sometimes interfere with each other, impacting their effectiveness. In contrast methods that focus on achieve better embedding representations, we propose straightforward yet approach directly incorporate Spike Signal Embedding (SEP) into process.Unlike many other approaches introduce Gaussian-distributed noise, spike signals generated by Poisson encoder typically maintain positive relationship original embeddings. Our experimental results demonstrate this proposed significantly enhances performance when compared LightGCN. leads substantial improvements efficiency.

Language: Английский

Citations

2