Efficient parameter calibration and real-time simulation of large-scale spiking neural networks with GeNN and NEST DOI Creative Commons
Felix Johannes Schmitt, Vahid Rostami, Martin Paul Nawrot

и другие.

Frontiers in Neuroinformatics, Год журнала: 2023, Номер 17

Опубликована: Фев. 10, 2023

Spiking neural networks (SNNs) represent the state-of-the-art approach to biologically realistic modeling of nervous system function. The systematic calibration for multiple free model parameters is necessary achieve robust network function and demands high computing power large memory resources. Special requirements arise from closed-loop simulation in virtual environments real-time robotic application. Here, we compare two complementary approaches efficient large-scale SNN simulation. widely used NEural Simulation Tool (NEST) parallelizes across CPU cores. GPU-enhanced Neural Network (GeNN) simulator uses highly parallel GPU-based architecture gain speed. We quantify fixed variable costs on single machines with different hardware configurations. As a benchmark model, use spiking cortical attractor topology densely connected excitatory inhibitory neuron clusters homogeneous or distributed synaptic time constants comparison random balanced network. show that scales linearly simulated biological and, networks, approximately size as dominated by number connections. Additional GeNN are almost independent size, while NEST increase size. demonstrate how can be simulating up 3.5 · 10 6 neurons (> 3 12 synapses) high-end GPU, 250, 000 (25 9 low-cost GPU. Real-time was achieved 100, neurons. parameter grid search efficiently using batch processing. discuss advantages disadvantages both cases.

Язык: Английский

Thoughtful faces: inferring internal states across species using facial features DOI Creative Commons
Alejandro Tlaie, Muad Y Abd El Hay,

Berkutay Mert

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Янв. 29, 2024

Animal behaviour is shaped to a large degree by internal cognitive states, but it unknown whether these states are similar across species. To address this question, we developed virtual reality setup in which mice and macaques engage the same naturalistic visual foraging task. We exploited richness of wide range facial features extracted from video recordings during task, train Markov-Switching Linear Regression (MSLR). By doing so, identified, on single-trial basis, set that reliably predicted when animals were going react presented stimuli. Even though model was trained purely reaction times, could also predict task outcome, supporting behavioural relevance inferred states. The identified comparable between monkeys. Furthermore, each state corresponded characteristic pattern features, highlighting importance expressions as manifestations

Язык: Английский

Процитировано

2
Hierarchical behavior control by a single class of interneurons DOI Creative Commons
Jing Huo, Tianqi Xu, Qi Liu

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2023, Номер unknown

Опубликована: Март 13, 2023

Animal behavior is organized into nested temporal patterns that span multiple timescales. This hierarchy believed to arise from a hierarchical neural architecture: neurons near the top of are involved in planning, selecting, initiating, and maintaining motor programs, whereas those bottom act concert produce fine spatiotemporal activity. In Caenorhabditis elegans , on long timescale emerges ordered flexible transitions between different behavioral states, such as forward, reversal, turn. On short timescale, parts animal body coordinate fast rhythmic bending sequences directional movements. Here, we show SAA, class interneurons enable cross-communication dorsal ventral head neurons, play dual role shaping dynamics SAA regulate stabilize activity during forward same suppress spontaneous reversals facilitate reversal termination by inhibiting RIM, an integrating neuron helps maintain state. These results suggest feedback lower-level cell assembly higher-level command center essential for bridging at levels.

Язык: Английский

Процитировано

5
Exploring transgenerational inheritance in epigenotypes of DAT heterozygous rats: Circadian anomalies and attentional vulnerability DOI Creative Commons
Concetto Puzzo, Fabiana Festucci, Giuseppe Curcio

и другие.

Behavioural Brain Research, Год журнала: 2024, Номер 464, С. 114921 - 114921

Опубликована: Фев. 24, 2024

Dopamine (DA) is mainly involved in locomotor activity, reward processes and maternal behaviors. Rats with KO gene for dopamine transporter (DAT) coding a truncated DAT protein are hyperdopaminergic conditions develop stereotyped behaviors hyperactivity. Our aim was to test the prior transgenerational modulation of wild allele as expressed heterozygous rats: specifically we addressed possible sequelae due genotype gender ancestors, regard behavioral differences F1, F2, F3 rats. We studied non-classical heterozygotes based on two specular lines, putative grand-maternal vs. grand-paternal imprinting. MAT females (F1; offspring male WT female) mated generate MIX (F2). Specularly, PAT female male) PIX Similarly PAT, obtained MUX (F2; HET sire dam); also observed (MYX: female, thus grandmother like PIX). their circadian activity behavior elevated-plus-maze (EPM). Locomotor hyper-activity occurs opposite MYX rats appearing undistinguishable from ones. Open-arm preference emerged MIX. Only showed significant vulnerability ADHD-like inattentive symptoms (duration rearing EPM; Viggiano et al., 2002). A risk-taking profile evident F2 phenotype while inattentiveness F1 progeny tends be transferred F3. hypothesize that DAT-related phenotypes result effective inheritance through pedigree dependent grandparents, suggesting protective role gestation future dam uterus. For major features, similar odd (F1, F3) generations appear opposed even (F2) ones; minor specific transfer may affect progenies but not DAT-KO ancestor.

Язык: Английский

Процитировано

1
Probabilistic modeling reveals coordinated social interaction states and their multisensory bases DOI Creative Commons
Sarah J. Stednitz,

Andrew Lesak,

Adeline Fecker

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2024, Номер unknown

Опубликована: Авг. 6, 2024

Social behavior across animal species ranges from simple pairwise interactions to thousands of individuals coordinating goal-directed movements. Regardless the scale, these are governed by interplay between multimodal sensory information and internal state each animal. Here, we investigate how animals use multiple modalities guide social in highly zebrafish (

Язык: Английский

Процитировано

1
Efficient parameter calibration and real-time simulation of large-scale spiking neural networks with GeNN and NEST DOI Creative Commons
Felix Johannes Schmitt, Vahid Rostami, Martin Paul Nawrot

и другие.

Frontiers in Neuroinformatics, Год журнала: 2023, Номер 17

Опубликована: Фев. 10, 2023

Spiking neural networks (SNNs) represent the state-of-the-art approach to biologically realistic modeling of nervous system function. The systematic calibration for multiple free model parameters is necessary achieve robust network function and demands high computing power large memory resources. Special requirements arise from closed-loop simulation in virtual environments real-time robotic application. Here, we compare two complementary approaches efficient large-scale SNN simulation. widely used NEural Simulation Tool (NEST) parallelizes across CPU cores. GPU-enhanced Neural Network (GeNN) simulator uses highly parallel GPU-based architecture gain speed. We quantify fixed variable costs on single machines with different hardware configurations. As a benchmark model, use spiking cortical attractor topology densely connected excitatory inhibitory neuron clusters homogeneous or distributed synaptic time constants comparison random balanced network. show that scales linearly simulated biological and, networks, approximately size as dominated by number connections. Additional GeNN are almost independent size, while NEST increase size. demonstrate how can be simulating up 3.5 · 10 6 neurons (> 3 12 synapses) high-end GPU, 250, 000 (25 9 low-cost GPU. Real-time was achieved 100, neurons. parameter grid search efficiently using batch processing. discuss advantages disadvantages both cases.

Язык: Английский

Процитировано

2