A Hopfield network model of neuromodulatory arousal state DOI Creative Commons
Mohammed Abdal Monium Osman, Kai J Fox, Joshua Isaac Stern

и другие.

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

Опубликована: Сен. 16, 2024

Abstract Neural circuits display both input-driven activity that is necessary for the real-time control of behavior and internally generated memory, planning, other cognitive processes. A key mediator between these intrinsic evoked dynamics arousal, an internal state variable determines animal’s level engagement with its environment. It has been hypothesized arousal acts through neuromodulatory gain mechanisms suppress recurrent connectivity amplify bottom-up input. In this paper, we instantiate longstanding idea in a continuous Hopfield network embellished parameter mimics by suppressing interactions network’s units. We show capturing some essential effects at neural levels emerge simple model as single parameter—recurrent gain—is varied. Using model’s formal connections to Boltzmann machine Ising model, offer functional interpretations rooted Bayesian inference statistical physics. Finally, liken neuromodulator release annealing schedule facilitates adaptive ever-changing environments. summary, present minimal exhibits rich but analytically tractable emergent reveals conceptually clarifying parallels seemingly unrelated phenomena.

Язык: Английский

Deciphering neuronal variability across states reveals dynamic sensory encoding DOI Creative Commons
Shailaja Akella, Peter Ledochowitsch, Joshua H. Siegle

и другие.

Nature Communications, Год журнала: 2025, Номер 16(1)

Опубликована: Фев. 19, 2025

Язык: Английский

Процитировано

0
Structured flexibility in recurrent neural networks via neuromodulation DOI Creative Commons
Julia C. Costacurta, Shaunak Bhandarkar, David M. Zoltowski

и другие.

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

Опубликована: Июль 26, 2024

Abstract The goal of theoretical neuroscience is to develop models that help us better understand biological intelligence. Such range broadly in complexity and detail. For example, task-optimized recurrent neural networks (RNNs) have generated hypotheses about how the brain may perform various computations, but these typically assume a fixed weight matrix representing synaptic connectivity between neurons. From decades research, we know weights are constantly changing, controlled part by chemicals such as neuromodulators. In this work explore computational implications gain scaling, form neuromodulation, using low-rank RNNs. our neuromodulated RNN (NM-RNN) model, neuromodulatory subnetwork outputs low-dimensional signal dynamically scales an output-generating RNN. empirical experiments, find structured flexibility NM-RNN allows it both train generalize with higher degree accuracy than RNNs on set canonical tasks. Additionally, via analyses show scaling endows gating mechanisms commonly found artificial We end analyzing dynamics trai ned NM-RNNs, task computations distributed.

Язык: Английский

Процитировано

1
A Hopfield network model of neuromodulatory arousal state DOI Creative Commons
Mohammed Abdal Monium Osman, Kai J Fox, Joshua Isaac Stern

и другие.

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

Опубликована: Сен. 16, 2024

Abstract Neural circuits display both input-driven activity that is necessary for the real-time control of behavior and internally generated memory, planning, other cognitive processes. A key mediator between these intrinsic evoked dynamics arousal, an internal state variable determines animal’s level engagement with its environment. It has been hypothesized arousal acts through neuromodulatory gain mechanisms suppress recurrent connectivity amplify bottom-up input. In this paper, we instantiate longstanding idea in a continuous Hopfield network embellished parameter mimics by suppressing interactions network’s units. We show capturing some essential effects at neural levels emerge simple model as single parameter—recurrent gain—is varied. Using model’s formal connections to Boltzmann machine Ising model, offer functional interpretations rooted Bayesian inference statistical physics. Finally, liken neuromodulator release annealing schedule facilitates adaptive ever-changing environments. summary, present minimal exhibits rich but analytically tractable emergent reveals conceptually clarifying parallels seemingly unrelated phenomena.

Язык: Английский

Процитировано

0