Single Cortical Neurons as Deep Artificial Neural Networks

Beniaguev, David; Segev, Idan; London, Michael

doi:10.2139/ssrn.3717773

Cited by 13 publications

(7 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The present study adds to the growing literature of potential computational abilities of compartmentalized neurons (Poirazi et al, 2003;Guerguiev et al, 2017;Beniaguev et al, 2019;Gidon et al, 2020). The associative HD neuron used in this study is a coincidence detector, which serves to associate external and internal inputs arriving at different compartments of the cell.…”

Section: Discussionmentioning

confidence: 97%

Learning accurate path integration in a ring attractor model of the head direction system

Vafidis

Owald

D'Albis

et al. 2021

Preprint

View full text Add to dashboard Cite

Ring attractor models for angular path integration have recently received strong experimental support. To function as integrators, head-direction (HD) circuits require precisely tuned connectivity, but it is currently unknown how such tuning could be achieved. Here, we propose a network model in which a local, biologically plausible learning rule adjusts synaptic efficacies during development, guided by supervisory allothetic cues. Applied to the Drosophila HD system, the model learns to path-integrate accurately and develops a connectivity strikingly similar to the one reported in experiments. The mature network is a quasi-continuous attractor and reproduces key experiments in which optogenetic stimulation controls the internal representation of heading, and where the network remaps to integrate with different gains. Our model predicts that path integration requires supervised learning during a developmental phase. The model setting is general and also applies to architectures that lack the physical topography of a ring, like the mammalian HD system.

show abstract

Section: Discussionmentioning

confidence: 97%

Learning accurate path integration in a ring attractor model of the head direction system

Vafidis

Owald

D'Albis

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…The SM-STC model can also be used in neurons with biological morphology, which have hundreds or thousands of compartments. A single-neuron model with biological morphology is shown to have strong computational power (Gidon et al, 2020;Beniaguev et al, 2021). Moreover, single neurons can learn network-level computations simply by tuning synaptic weights (Bicknell and Häusser, 2021).…”

Section: Discussionmentioning

confidence: 99%

A Simplified Plasticity Model Based on Synaptic Tagging and Capture Theory: Simplified STC

Ding

Wang

Cao

2022

Front. Comput. Neurosci.

View full text Add to dashboard Cite

The formation and consolidation of memory play a vital role for survival in an ever-changing environment. In the brain, the change and stabilization of potentiated and depressed synapses are the neural basis of memory formation and maintenance. These changes can be induced by rather short stimuli (only a few seconds or even less) but should then be stable for months or years. Recently, the neural mechanism of conversion from rapid change during the early phase of synaptic plasticity into a stable memory trace in the late phase of synaptic plasticity is more and more clear at the protein and molecular levels, among which synaptic tagging and capture (STC) theory is one of the most popular theories. According to the STC theory, the change and stabilization of synaptic efficiency mainly depend on three processes related to calcium concentration, including synaptic tagging, synthesis of plasticity-related product (PRP), and the capture of PRP by tagged synapse. Based on the STC theory, several computational models are proposed. However, these models hardly take simplicity and biological interpretability into account simultaneously. Here, we propose a simplified STC (SM-STC) model to address this issue. In the SM-STC model, the concentration of calcium ion in each neuronal compartment and synapse is first calculated, and then the tag state of synapse and PRP are updated, and the coupling effect of tagged synapse and PRP is further considered to determine the plasticity state of the synapse, either potentiation or depression. We simulated the Schaffer collaterals pathway of the hippocampus targeting a multicompartment CA1 neuron for several hours of biological time. The results show that the SM-STC model can produce a broad range of experimental phenomena known in the physiological experiments, including long-term potentiation induced by high-frequency stimuli, long-term depression induced by low-frequency stimuli, and cross-capture with two stimuli separated by a delay. Thus, the SM-STC model proposed in this study provides an effective learning rule for brain-like computation on the premise of ensuring biological plausibility and computational efficiency.

show abstract

“…On the other hand, seminal work by Poirazi et al [13] suggests that in fact a 2-layer artificial neural network may be necessary to capture the input-output mapping of a single neuron, implying that the neuron's expressive power may be on a similar level. Whereas these earlier results analyzed the neuron in a static framework, Beniaguev et al [14] incorporate temporal dynamics as well, and instead conclude that a temporally convolutional deep * Both authors contributed equally.…”

Section: Introductionmentioning

confidence: 92%

Dendritic plateau potentials can process spike sequences across multiple time-scales

Leugering

Nieters

Pipa

2019

Preprint

View full text Add to dashboard Cite

Many behavioural tasks require an animal to integrate information on a slow timescale that can exceed hundreds of milliseconds. How this is realized by neurons with membrane time constants on the order of tens of milliseconds or less remains an open question. We show, how the interaction of two kinds of events within the dendritic tree, excitatory postsynaptic potentials and locally generated dendritic plateau potentials, can allow a single neuron to detect specific sequences of spiking input on such slow timescales. The morphology of a neuron's dendritic arbor determines which inputs the neuron responds to, which we express formally in terms of a temporal logic. Finally, we extend this model of dendritic computation to homogeneous populations of neurons and illustrate in a simulated navigation task, how it can be used to integrate and retain information on a behaviourally relevant timescale.1 Introduction 1 The ability to detect long-lasting sequences of neural activity is crucial for complex behavior, but poses a serious 2 challenge for most established neuron models. Consider a rodent navigating through an environment in search for 3 food. Receptive fields of place and grid cells tile a spatial map of the environment and encode the current position by 4 their respective population activities [1, 2]. But in order to find its way back, the animal needs to know not only its 5 present location, but also which path it took to get there. Decoding this path from the sequential activation of place and 6 grid cells requires the integration of information on behavioural timescales that can span hundreds of milliseconds or 7 more [3]. Relevant patterns on such long timescales may prove to be a ubiquitous phenomenon, and have already been 8 documented for a wide range of sensory processing tasks, such as olfaction [4, 5] or cortical auditory processing [6]. 9 This raises the puzzling question, how such long sequences of neural activity can be processed by volatile neurons with 10 membrane time constants on the timescale of tens of milliseconds or less [7]. While this problem is typically addressed 11 on a network level, e.g. by relying on effects of fast-acting synaptic plasticity [8] or slow emergent dynamics due to 12 recurrent connections [9], we argue that it can be solved on the level of individual neurons by active processes within 13 their dendritic trees. 14 From experimental neurobiological evidence and theoretical considerations we conclude that: 15 1. Localized, active dendritic processes endow neurons with internal memory traces on the timescale of hundreds 16 of milliseconds.172. The interaction of these processes allows the detection of complex and long-lasting temporal patterns. 18 3. The computation realized by a neuron is determined by the morphology of its dendritic arbor and can be 19 expressed by a temporal logic.20 4. Due to stochastic transmission of spikes, populations of neurons can realize a 'fuzzy' probabilistic extension 21 of this temporal logic. 22 * Both authors contributed equally. PREP...

show abstract

Single Cortical Neurons as Deep Artificial Neural Networks

Cited by 13 publications

References 0 publications

Learning accurate path integration in a ring attractor model of the head direction system

Learning accurate path integration in a ring attractor model of the head direction system

A Simplified Plasticity Model Based on Synaptic Tagging and Capture Theory: Simplified STC

Dendritic plateau potentials can process spike sequences across multiple time-scales

Contact Info

Product

Resources

About