A Large-Scale Model of the Functioning Brain

Science, Год журнала: 2012, Номер 338(6111), С. 1202 - 1205

Опубликована: Ноя. 29, 2012

A central challenge for cognitive and systems neuroscience is to relate the incredibly complex behavior of animals equally activity their brains. Recently described, large-scale neural models have not bridged this gap between biological function. In work, we present a 2.5-million-neuron model brain (called "Spaun") that bridges by exhibiting many different behaviors. The presented only with visual image sequences, it draws all its responses physically modeled arm. Although simplified, captures aspects neuroanatomy, neurophysiology, psychological behavior, which demonstrate via eight diverse tasks.

Язык: Английский

---

Nengo: a Python tool for building large-scale functional brain models

Frontiers in Neuroinformatics, Год журнала: 2014, Номер 7

Опубликована: Янв. 1, 2014

Neuroscience currently lacks a comprehensive theory of how cognitive processes can be implemented in biological substrate. The Neural Engineering Framework (NEF) proposes one such theory, but has not yet gathered significant empirical support, partly due to the technical challenge building and simulating large-scale models with NEF. Nengo is software tool that used build simulate based on NEF; currently, it primary resource for both teaching NEF used, doing research generates specific explain experimental data. 1.4, which was Java, create Spaun, world's largest functional brain model (Eliasmith et al., 2012). Simulating Spaun highlighted limitations 1.4's ability support construction simple syntax, large quickly, collect amounts data subsequent analysis. This paper describes 2.0, Python overcomes these limitations. It uses extendable simulates benchmark scale 50 times faster than flexible mechanism collecting simulation results.

Язык: Английский

---

Mechanisms of Persistent Activity in Cortical Circuits: Possible Neural Substrates for Working Memory

Annual Review of Neuroscience, Год журнала: 2017, Номер 40(1), С. 603 - 627

Опубликована: Июль 25, 2017

A commonly observed neural correlate of working memory is firing that persists after the triggering stimulus disappears. Substantial effort has been devoted to understanding many potential mechanisms may underlie memory-associated persistent activity. These rely either on intrinsic properties individual neurons or connectivity within circuits maintain Nevertheless, it remains unclear which are at play in brain areas involved memory. Herein, we first summarize palette different can generate We then discuss recent work asks activity areas. Finally, future studies might tackle this question further. Our goal bridge between communities researchers who study single-neuron biophysical, circuit, underlies

Язык: Английский

---

Nengo and Low-Power AI Hardware for Robust, Embedded Neurorobotics

Frontiers in Neurorobotics, Год журнала: 2020, Номер 14

Опубликована: Окт. 9, 2020

In this paper we demonstrate how the Nengo neural modeling and simulation libraries enable users to quickly develop robotic perception action networks for on neuromorphic hardware using tools they are already familiar with, such as Keras Python. We identify four primary challenges in building robust, embedded neurorobotic systems, including: 1) developing infrastructure interfacing with environment sensors; 2) processing task specific sensory signals; 3) generating explainable control 4) compiling run target hardware. helps address these by: providing NengoInterfaces library, which defines a simple but powerful API interact simulations hardware; NengoDL lets use TensorFlow models; implementing Neural Engineering Framework, provides white-box methods known functions circuits; multiple backend libraries, NengoLoihi, that compile same model different present two examples CPUs GPUs well Intel's chip, Loihi, variations workflow. The first example is an implementation of end-to-end spiking network controls rover simulated Mujoco. integrates deep convolutional processes visual input from cameras mounted track target, system steering drive connection weights guide target. second uses smaller component has addressed some not all those challenges. Specifically it used augment force-based operational space controller adaptive improve performance during reaching real-world Kinova Jaco 2 arm. code details provided, intent enabling other researchers build their own systems.

Язык: Английский

---

Learning the pseudoinverse solution to network weights

Neural Networks, Год журнала: 2013, Номер 45, С. 94 - 100

Опубликована: Март 14, 2013

Язык: Английский

---

Learning to Select Actions with Spiking Neurons in the Basal Ganglia

Frontiers in Neuroscience, Год журнала: 2012, Номер 6

Опубликована: Янв. 1, 2012

We expand our existing spiking neuron model of decision making in the cortex and basal ganglia to include local learning on synaptic connections between striatum, modulated by a dopaminergic reward signal. then compare this animal data bandit task, which is used test rodent conditions involving forced choice under rewards. Our results indicate good match terms both behavioral spike patterns ventral striatum. The successfully generalizes utilities multiple actions, can learn choose different actions states. purpose provide high-level predictions low-level timing while respecting known neurophysiology neuroanatomy.

Язык: Английский

---

A spiking neural model of adaptive arm control

Proceedings of the Royal Society B Biological Sciences, Год журнала: 2016, Номер 283(1843), С. 20162134 - 20162134

Опубликована: Ноя. 30, 2016

We present a spiking neuron model of the motor cortices and cerebellum control system. The consists anatomically organized neurons encompassing premotor, primary motor, cerebellar cortices. proposes novel neural computations within these areas to nonlinear three-link arm that can adapt unknown changes in dynamics kinematic structure. demonstrate mathematical stability both forms adaptation, suggesting this is robust approach for common biological problems changing body size (e.g. during growth), unexpected dynamic perturbations when moving through different media, such as water or mud). To plausibility proposed mechanisms, we show accounts data across 19 studies These include mix behavioural activity, subjects performing adaptive static tasks. Given characterization processes involved arm, provide several experimentally testable predictions distinguish our from previous work.

Язык: Английский

---

Predicting non-linear dynamics by stable local learning in a recurrent spiking neural network

eLife, Год журнала: 2017, Номер 6

Опубликована: Ноя. 27, 2017

The brain needs to predict how the body reacts motor commands, but a network of spiking neurons can learn non-linear dynamics using local, online and stable learning rules is unclear. Here, we present supervised scheme for feedforward recurrent connections in heterogeneous neurons. error output fed back through fixed random with negative gain, causing follow desired dynamics. rule Feedback-based Online Local Learning Of Weights (FOLLOW) local sense that weight changes depend on presynaptic activity signal projected onto postsynaptic neuron. We provide examples linear, chaotic dynamics, as well two-link arm. Under reasonable approximations, show, Lyapunov method, FOLLOW uniformly stable, going zero asymptotically.

Язык: Английский

---

Neurocognitive networks: Findings, models, and theory

Neuroscience & Biobehavioral Reviews, Год журнала: 2012, Номер 36(10), С. 2232 - 2247

Опубликована: Авг. 17, 2012

Язык: Английский

---

Large-Scale Synthesis of Functional Spiking Neural Circuits

Proceedings of the IEEE, Год журнала: 2014, Номер 102(5), С. 881 - 898

Опубликована: Март 27, 2014

In this paper, we review the theoretical and software tools used to construct Spaun, first (and so far only) brain model capable of performing cognitive tasks. This tool set allowed us configure 2.5 million simple nonlinear components (neurons) with 60 billion connections between them (synapses) such that resulting can perform eight different perceptual, motor, To reverse-engineer in way, a method is needed shows how large numbers components, each which receives thousands inputs from other be organized desired computations. We achieve through neural engineering framework (NEF), mathematical theory provides methods for systematically generating biologically plausible spiking networks implement linear dynamical systems. On top this, propose semantic pointer architecture (SPA), hypothesis regarding some aspects organization, function, representational resources mammalian brain. conclude by discussing an example uses SPA implemented using NEF. Throughout, discuss Neural ENGineering Objects (Nengo), allows synthesis simulation models efficiently on scale support constructing NEF SPA. The NEF/SPA/Nengo combination general both evaluating hypotheses about works, building systems compute particular functions neuron-like components.

Язык: Английский

---