Nonstationary modeling of neural population dynamics

Convolutional Networks Outperform Linear Decoders in Predicting EMG From Spinal Cord Signals

Advanced algorithms are required to reveal the complex relations between neural and behavioral data. In this study, forelimb electromyography (EMG) signals were reconstructed from multi-unit neural signals recorded with multiple electrode arrays (MEAs) from the corticospinal tract (CST) in rats. A six-layer convolutional neural network (CNN) was compared with linear decoders for predicting the EMG signal. The network contained three session-dependent Rectified Linear Unit (ReLU) feature layers and three Gamma function layers were shared between sessions. Coefficient of determination (R²) values over 0.2 and correlations over 0.5 were achieved for reconstruction within individual sessions in multiple animals, even though the forelimb position was unconstrained for most of the behavior duration. The CNN performed visibily better than the linear decoders and model responses outlasted the activation duration of the rat neuromuscular system. These findings suggest that the CNN model implicitly predicted short-term dynamics of skilled forelimb movements from neural signals. These results are encouraging that similar problems in neural signal processing may be solved using variants of CNNs defined with simple analytical functions. Low powered firmware can be developed to house these CNN solutions in real-time applications.

ASIC Implementation of a Nonlinear Dynamical Model for Hippocampal Prosthesis

A hippocampal prosthesis is a very large scale integration (VLSI) biochip that needs to be implanted in the biological brain to solve a cognitive dysfunction. In this letter, we propose a novel low-complexity, small-area, and low-power programmable hippocampal neural network application-specific integrated circuit (ASIC) for a hippocampal prosthesis. It is based on the nonlinear dynamical model of the hippocampus: namely multi-input, multi-output (MIMO)-generalized Laguerre-Volterra model (GLVM). It can realize the real-time prediction of hippocampal neural activity. New hardware architecture, a storage space configuration scheme, low-power convolution, and gaussian random number generator modules are proposed. The ASIC is fabricated in 40 nm technology with a core area of 0.122 mm[Formula: see text] and test power of 84.4 [Formula: see text]W. Compared with the design based on the traditional architecture, experimental results show that the core area of the chip is reduced by 84.94% and the core power is reduced by 24.30%.

Volterra representation enables modeling of complex synaptic nonlinear dynamics in large-scale simulations

Chemical synapses are comprised of a wide collection of intricate signaling pathways involving complex dynamics. These mechanisms are often reduced to simple spikes or exponential representations in order to enable computer simulations at higher spatial levels of complexity. However, these representations cannot capture important nonlinear dynamics found in synaptic transmission. Here, we propose an input-output (IO) synapse model capable of generating complex nonlinear dynamics while maintaining low computational complexity. This IO synapse model is an extension of a detailed mechanistic glutamatergic synapse model capable of capturing the input-output relationships of the mechanistic model using the Volterra functional power series. We demonstrate that the IO synapse model is able to successfully track the nonlinear dynamics of the synapse up to the third order with high accuracy. We also evaluate the accuracy of the IO synapse model at different input frequencies and compared its performance with that of kinetic models in compartmental neuron models. Our results demonstrate that the IO synapse model is capable of efficiently replicating complex nonlinear dynamics that were represented in the original mechanistic model and provide a method to replicate complex and diverse synaptic transmission within neuron network simulations.

An Application Specific Instruction Set Processor (ASIP) for Adaptive Filters in Neural Prosthetics

Neural coding is an essential process for neuroprosthetic design, in which adaptive filters have been widely utilized. In a practical application, it is needed to switch between different filters, which could be based on continuous observations or point process, when the neuron models, conditions, or system requirements have changed. As candidates of coding chip for neural prostheses, low-power general purpose processors are not computationally efficient especially for large scale neural population coding. Application specific integrated circuits (ASICs) do not have flexibility to switch between different adaptive filters while the cost for design and fabrication is formidable. In this research work, we explore an application specific instruction set processor (ASIP) for adaptive filters in neural decoding activity. The proposed architecture focuses on efficient computation for the most time-consuming matrix/vector operations among commonly used adaptive filters, being able to provide both flexibility and throughput. Evaluation and implementation results are provided to demonstrate that the proposed ASIP design is area-efficient while being competitive to commercial CPUs in computational performance.

Laguerre-Volterra identification of spike-timing-dependent plasticity from spiking activity: a simulation study

This paper presents a Laguerre-Volterra methodology for identifying a plasticity learning rule from spiking neural data with four components: 1) By analyzing input-output spiking data, the effective contribution of an input on the output firing probability can be quantified with weighted Volterra kernels. 2) The weight of these Volterra kernels can be tracked over time using the stochastic state point processing filtering algorithm (SSPPF) 3) Plasticity system Volterra kernels can be estimated by treating the tracked change in weight over time as the plasticity system output and the spike timing data as the input. 4) Laguerre expansion of all Volterra kernels allows for minimization of open parameters during estimation steps. A single input spiking neuron with Spike-timing-dependent plasticity (STDP) and prolonged STDP induction is simulated. Using the spiking data from this simulation, the amplitude of the STDP learning rule and the time course of the induction is accurately estimated. This framework can be applied to identify plasticity for more complicated plasticity paradigms and is applicable to in vivo data.

A customizable stochastic state point process filter (SSPPF) for neural spiking activity

Stochastic State Point Process Filter (SSPPF) is effective for adaptive signal processing. In particular, it has been successfully applied to neural signal coding/decoding in recent years. Recent work has proven its efficiency in non-parametric coefficients tracking in modeling of mammal nervous system. However, existing SSPPF has only been realized in commercial software platforms which limit their computational capability. In this paper, the first hardware architecture of SSPPF has been designed and successfully implemented on field-programmable gate array (FPGA), proving a more efficient means for coefficient tracking in a well-established generalized Laguerre-Volterra model for mammalian hippocampal spiking activity research. By exploring the intrinsic parallelism of the FPGA, the proposed architecture is able to process matrices or vectors with random size, and is efficiently scalable. Experimental result shows its superior performance comparing to the software implementation, while maintaining the numerical precision. This architecture can also be potentially utilized in the future hippocampal cognitive neural prosthesis design.

Nonlinear dynamic modeling of synaptically driven single hippocampal neuron intracellular activity

A high-order nonlinear dynamic model of the input-output properties of single hippocampal CA1 pyramidal neurons was developed based on synaptically driven intracellular activity. The purpose of this study is to construct a model that: 1) can capture the nonlinear dynamics of both subthreshold activities [postsynaptic potentials (PSPs)] and suprathreshold activities (action potentials) in a single formalism; 2) is sufficiently general to be applied to any spike-input and spike-output neurons (point process input and point process output neural systems); and 3) is computationally efficient. The model consisted of three major components: 1) feedforward kernels (up to third order) that transform presynaptic action potentials into PSPs; 2) a constant threshold, above which action potentials are generated; and 3) a feedback kernel (first order) that describes spike-triggered after-potentials. The model was applied to CA1 pyramidal cells, as they were electrically stimulated with broadband Poisson random impulse trains through the Schaffer collaterals. The random impulse trains used here have physiological properties similar to spiking patterns observed in CA3 hippocampal neurons. PSPs and action potentials were recorded from the soma of CA1 pyramidal neurons using whole-cell patch-clamp recording. We evaluated the model performance separately with respect to PSP waveforms and the occurrence of spikes. The average normalized mean square error of PSP prediction is 14.4%. The average spike prediction error rate is 18.8%. In summary, although prediction errors still could be reduced, the model successfully captures the majority of high-order nonlinear dynamics of the single-neuron intracellular activity. The model captures the general biophysical processes with a small set of open parameters that are directly constrained by the intracellular recording, and thus, can be easily applied to any spike-input and spike-output neuron.

Tracking the changes of hippocampal population nonlinear dynamics in rats learning a memory-dependent task

Neurobiological processes associated with learning are known to be highly nonlinear, dynamical, and time-varying. Characterizing the time-varying functional input-output properties of neural systems is a critical step to understand the neurobiological basis of learning. In this paper, we present a study on tracking of the changes of neural dynamics in rat hippocampus during learning of a memory-dependent delayed nonmatch-to-sample (DNMS) task. The rats were first trained to perform the DNMS task without a delay between the sample and response events. After reaching a performance level, they were subjected to the DNMS task with variable delays with a 5s mean duration. Spike trains were recorded from hippocampal CA3 (input) and CA1 (output) regions during all training sessions and constitute the input-output data for modeling. We applied the time-varying Generalized Laguerre-Volterra Model to study the changes of the CA3-CA1 nonlinear dynamics using these data. Result showed significant changes in the Volterra kernels after the introduction of delays. This result suggests that the CA3-CA1 nonlinear dynamics established in the initial training sessions underwent a functional reorganization as animals were learning to perform the task that now requires delays.