Converting Data for Spiking Neural Network Training DOI Open Access

Erik Sádovský,

Maroš Jakubec, Roman Jarina

et al.

International Journal of Advanced Computer Science and Applications, Journal Year: 2023, Volume and Issue: 14(8)

Published: Jan. 1, 2023

The application of spiking neural networks (SNNs) for processing visual and auditory data necessitate the conversion traditional network datasets into a format suitable spike-based computations. Existing designed conventional are incompatible with SNNs due to their reliance on spike timing specific preprocessing requirements. This paper introduces comprehensive pipeline that enables common rate-coded spikes, meeting demands SNNs. proposed solution is evaluated Spike-CNN trained Time-to-First-Spike encoded MNIST compared similar system neuromorphic dataset (N-MNIST). Both systems have comparative precision; however more energy efficient than based computing. Since, not limited any form can be applied various types audio/visual content. By providing means adapt existing datasets, this research facilitates exploration advancement across different domains.

Language: Английский

On-Chip Spike Detection and Classification using Neural Networks and Approximate Computing DOI
Efstratios Zacharelos, Ciro Scognamillo, Ettore Napoli

et al.

2022 IEEE Biomedical Circuits and Systems Conference (BioCAS), Journal Year: 2023, Volume and Issue: unknown

Published: Oct. 19, 2023

Neural ensembles control sensory, motor, and cognitive functions. Action potentials of neuronal cells (spikes) may signify such functions, or the presence a pathology. In this paper we give circuital implementation an Artificial Network, able to sort (detect classify) spikes in real time. The system is synthesized targeting 14nm FinFET technology. To partially alleviate computational burden, approximate computing methods have been integrated during inference stage, yielding up 63% reduction dynamic power. different versions circuit reach accuracy range from 65% 93%, with silicon area power that 2000μm 2 , 0.1μW@30kHz 6000μm 0.7μW@30kHz. electrical performances proposed overcome state art spike detection circuits while providing additional feature sorting single solution.

Language: Английский

Citations

2
Tracking spatial patterns and nocturnal arousal in an undisturbed natural population of the pulse-type weakly electric fish Gymnotus omarorum DOI
Adriana Migliaro, Federico Pedraja, Stefan Mucha

et al.

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: July 2, 2024

SUMMARY Assessing animals’ locomotor and activity-rest patterns in natural populations is challenging. It requires individual identification behavioral tracking sometimes complex inaccessible environments. Weakly electric fish are advantageous models for remote monitoring due to their continuous emission of signals (EODs). Gymnotus omarorum a South American freshwater pulse-type weakly fish. Previous manual recordings restrained individuals the wild showed spatial distribution compatible with territoriality nocturnal increase EOD rate interpreted as arousal. This interdisciplinary study presents development low-cost amplifiers refinement algorithms that provide recognition wild. We describe daily spacing undisturbed territoriality, although heterogeneous across sampling sites, confirm all resident robust likely associated variations water temperature. HIGHLIGHTS Successful pulse type G. known nocturnality Residents keep diurnal resting sites move within small areas during night The arousal residents linked temperature peak

Language: Английский

Citations

0
A data compression algorithm with the improved SRLE for high-throughput neural signal acquisition device DOI

W. Quan,

Xudong Guo, Haipo Cui

et al.

Technology and Health Care, Journal Year: 2024, Volume and Issue: 32(6), P. 3955 - 3966

Published: July 26, 2024

BACKGROUND: Multi-channel acquisition systems of brain neural signals can provide a powerful tool with wide range information for the clinical application computer interfaces. High-throughput implantable are limited by size and power consumption, posing challenges to system design. OBJECTIVE: To acquire more comprehensive wirelessly transmit high-throughput signals, FPGA-based multi-channel nerve has been developed. And Bluetooth transmission low-power technology utilized. METHODS: large amount data bandwidth improve accuracy signal decoding, an improved sharing run length encoding (SRLE) is proposed compress spike efficiency system. The functional prototype developed, which consists chips, FPGA main control module SRLE, wireless transmitter, receiver upper computer. developed was tested detection animal experiments. RESULTS: From experiments, it shows that successfully collect signals. SRLE algorithm excellent compression effect average rate 5.94%, compared double run-length encoding, FDR traditional encoding. CONCLUSION: system, incorporating algorithm, capable capturing 1024 channels, thereby realizing

Language: Английский

Citations

0
Improving Spike Sorting Efficiency with Separability Index and Spectral Clustering DOI

Leila Ranjbar,

Hossein Parsaei, Mohammad Mehdi Movahedi

et al.

Medical Engineering & Physics, Journal Year: 2024, Volume and Issue: 135, P. 104265 - 104265

Published: Nov. 29, 2024

Language: Английский

Citations

0
A Novel Clustering Algorithm Integrating Gershgorin Circle Theorem and Nonmaximum Suppression for Neural Spike Data Analysis DOI Creative Commons
Sahaj Anilbhai Patel, Abidin Yildirim

Signals, Journal Year: 2024, Volume and Issue: 5(2), P. 402 - 416

Published: June 4, 2024

(1) Problem Statement: The development of clustering algorithms for neural recordings has significantly evolved, reaching a mature stage with predominant approaches including partitional, hierarchical, probabilistic, fuzzy logic, density-based, and learning-based clustering. Despite this evolution, there remains need innovative that can efficiently analyze spike data, particularly in handling diverse noise-contaminated recordings. (2) Methodology: This paper introduces novel algorithm named Gershgorin—nonmaximum suppression (G–NMS), which incorporates the principles Gershgorin circle theorem, deep learning post-processing method known as nonmaximum suppression. performance G–NMS was thoroughly evaluated through extensive testing on two publicly available, synthetic datasets. evaluation involved five distinct groups experiments, totaling eleven individual to compare against six established algorithms. (3) Results: results highlight superior three out group achieving high average accuracy minimal standard deviation (SD). Specifically, Dataset 1, experiment S1 (various SNRs) recorded an 99.94 ± 0.01, while 2 showed accuracies 99.68 0.15 E1 (Easy 1) 99.27 0.35 E2 2). slight decrease remaining D1 (Difficult D2 2) from 2, compared top-performing these categories, maintained lower SD, indicating consistent performance. Additionally, demonstrated robustness efficiency across various recordings, ranging low signal-to-noise ratios. (4) Conclusions: G–NMS’s integration techniques eigenvalue inclusion theorems proven highly effective, marking significant advancement domain. Its performance, characterized by variability, opens new avenues high-performing algorithms, contributing body research field.

Language: Английский

Citations

0
Algorithm and Validation Method for Spike Sorting Based on Wavelet Analysis and a Genetic Algorithm DOI

Federico Alscher,

Rocío A. Lenzi,

Pamela Pérez Escobar

et al.

IFMBE proceedings, Journal Year: 2024, Volume and Issue: unknown, P. 57 - 67

Published: Jan. 1, 2024

Language: Английский

Citations

0
Deep learning-based spike sorting: a survey DOI Creative Commons
Lynn Meyer, Majid Zamani, János Rokai

et al.

Journal of Neural Engineering, Journal Year: 2024, Volume and Issue: 21(6), P. 061003 - 061003

Published: Oct. 25, 2024

Abstract Objective. Deep learning is increasingly permeating neuroscience, leading to a rise in signal-processing applications for extracellular recordings. These signals capture the activity of small neuronal populations, necessitating ‘spike sorting’ assign action potentials (spikes) their underlying neurons. With publications delving into new methodologies and techniques deep learning-based spike sorting, it crucial synthesise these findings critically. This survey provides an in-depth evaluation approaches, outcomes presented recent articles, shedding light on current state-of-the-art. Approach. Twenty-four articles published until December 2023 sorting have been examined. The proposed methods are divided three sub-problems sorting: detection, feature extraction classification. Moreover, integrated systems, i.e. models that detect spikes extract features or do classification within single network, included. Main results. Although most algorithms developed single-channel recordings, utilising multi-channel data already shown promising results, with efficient hardware implementations running quantised application-specific circuits field programmable gate arrays. Convolutional neural networks used extensively detection as can be processed spatiotemporally while maintaining low-parameter increasing generalisation efficiency. Autoencoders mainly utilised dimensionality reduction, enabling subsequent clustering standard methods. Also, systems great potential solving problem from end end. Significance. explores highlights capabilities overcoming associated challenges, but also biases certain models. Serving resource both newcomers seasoned researchers field, this work insights latest advancements may inspire future model development.

Language: Английский

Citations

0
Converting Data for Spiking Neural Network Training DOI Open Access

Erik Sádovský,

Maroš Jakubec, Roman Jarina

et al.

International Journal of Advanced Computer Science and Applications, Journal Year: 2023, Volume and Issue: 14(8)

Published: Jan. 1, 2023

The application of spiking neural networks (SNNs) for processing visual and auditory data necessitate the conversion traditional network datasets into a format suitable spike-based computations. Existing designed conventional are incompatible with SNNs due to their reliance on spike timing specific preprocessing requirements. This paper introduces comprehensive pipeline that enables common rate-coded spikes, meeting demands SNNs. proposed solution is evaluated Spike-CNN trained Time-to-First-Spike encoded MNIST compared similar system neuromorphic dataset (N-MNIST). Both systems have comparative precision; however more energy efficient than based computing. Since, not limited any form can be applied various types audio/visual content. By providing means adapt existing datasets, this research facilitates exploration advancement across different domains.

Language: Английский

Citations

0