Nosrat Mohammadi

@nosrat.bsky.social

873 Followers 1129 Following

Postdoctoral researcher in computational neuroscience, university of Geneva. Studying the compositional properties of syntax. Neural manifolds and emotions! webpage: nosratullah.github.io

Haft-sin
Zotero collection folder
A figure on spike variablity across different input characteristics from Gerstner's book.
Neuronal dynamics; a book on "neuronal dynamics"!
Ai generated scientific figure layout
A function to create custom color gradient
Plot configuration for matplotlib
A python function that conver cm to inch and can be used in matplotlib.
A scientific figure blueprint guide!
Saufla waterfall in Champéry, Switzerland during winter.
A demonstration of self-organized criticality by a tilted hourglass! A pile of sand is being enlarged and followed by a seemingly random collapse.
Dynamics of neuronal-network NH-PDMP models fitted to human neuronal population recordings: (a) Speech processing task. Left: speech waveforms, and phonemic content for five example words used in the semantic categorization task. Right: Microelectrode array recordings located in the anterior superior temporal gyrus (aSTG). Brain picture in panel A adapted from NIH BIOART Source. (b) Estimation of NH-PDMP models. The input to the NH-PDMP models consisted of the phonemes present in the heard word, here “mouse” (top). Spike train recordings of five neurons for one trial for the word “mouse” (middle). Corresponding simulation of the estimated NH-PDMP models (bottom). (c) Parameters for the cross-history kernels (left) and inputs (right). (d) The 15 fixed points in the 10-dimensional diffusion SDE state space that corresponded to each vowel input were projected into the first two PCs of the state space. Left: a separation of the fixed points is observed on the basis of the first formant of vowels (frequency below or above 500 Hz). Right: no such separation is observed for the second formant of vowels (frequency below and above 1.5 kHz).
Dynamics of neuronal-network NH-PDMP models fitted to nonhuman primate neuronal population recordings: (a) Center-out reaching movement task. Spiking activity from a population of single-units was recorded via a microelectrode array implanted in the primary motor (M1) cortex of a macaque monkey during in a center-out-reach task (top right). Bottom right, cursors position across trials. (b) Top: spike train raster of 20 out of 115 sorted M1 single-units during the task. Time zero corresponds to the go-cue onset. Bottom: simulation of the corresponding 20-neuron network NH-PDMP model fitted to the data. The model included an input corresponding to the target direction in each of the corresponding center-out reach trial (as shown in A). (c) Recorded neurons were ranked according to their tuning to target direction as captured by the norm of the input parameters ||βI||. (d) Parameters for cross-history history kernels (left) and stimulus contribution (right). (e) The eight target directions corresponded to eight fixed point in the 20-dimensional diffusion SDE state space, respectively. Their positions along the first two PCs of the 20-dimensional state space are shown here as black circle. The fixed points corresponding to 72 new simulated target directions are shown as well. The fixed points are organized along a one-dimensional closed-orbit manifold, reflecting the rotational symmetry present in the task.
(left) Center-out reaching movement task. Spiking activity from a population of single-units was recorded via a microelectrode array implanted in the primary motor (M1) cortex of a macaque monkey during in a center-out-reach task (top right). Bottom right, cursors position across trials. (right) Speech processing task. Left: speech waveforms, and phonemic content for five example words used in the semantic categorization task. Right: Microelectrode array recordings located in the anterior superior temporal gyrus (aSTG). Brain picture in panel A adapted from NIH BIOART Source.
Dynamical repertoire of neuronal-network NH-PDMP models, Winner-take-all: (a) Top: Network of two Izhikevich neurons for winner-take-all dynamics. Bottom: history kernels of the estimated NH-PDMP models. (b) Top: each Izhikevich neuron receives a distinct input (blue and red lines) resulting in different spiking trains (blue and red vertical lines). Bottom: simulation of five trials of the fitted two-neuron NH-PDMP models, showing winner-take-all dynamics, except for inputs with small differences. (c) Parameter space of the deterministic part of the diffusion SDEs as a function of inputs to each neuron. Colored diamonds indicate different parameter choices for the examples shown in (d). (d) State space spanned by the two variables z1 and z2 in the diffusion SDEs. Left: larger input to the red neuron, showing that the fixed point moves to high (low) values of the red (blue) neuron state variables. Middle: larger input to the blue neuron, showing that the fixed point moves respectively to high (low) values of the blue (red) neuron state variables. Right: equally large inputs to both neurons, showing bistability with two competing stable fixed points.
(left) Network of two Izhikevich neurons for winner-take-all dynamics. (right) Network of five Izhikevich neurons coupled in a feedforward configuration to support traveling-wave dynamics.
Dynamical repertoire of single-neuron NH-PDMP models: (a) Tonic spiking. Top left, example of a step function stimulus (blue). Middle left, corresponding Izhikevich neuron’s tonic spiking response (black). Bottom left, simulation of five trials of the fitted NH-PDMP model. Top right, history kernels of the estimated NH-PDMP model. Bottom right, parameter space of the deterministic part of the diffusion SDEs spanned by the input current and β1. Increasing the input I(t) for a fixed β1 leads to a transition along the horizontal arrow shown on the parameter space. Isoclines show the corresponding firing rate (in spikes/s) for the same neuron. The model transitions from a stable fixed point with very low rate to a stable fixed point with higher firing rate. (b) Phasic spiking. Same as in (a) for phasic spiking response. The model transitions from a stable fixed point to a stable focus regime with higher firing rate and high damping factor. (c) Tonic bursting. Same as in (a) for tonic bursting response. The model transitions from a stable fixed point to a stable focus with smaller damping factor and higher imaginary components, enabling the model to stay longer in the excited state. (d) Phasic bursting. Same as in (a) for phasic bursting response. The model transitions from a stable fixed point to a stable focus regime with higher firing rate and low damping factor. (e) Type I. Same as in (a) for type I response. The model transitions from a stable fixed point with very low firing rate to a stable fixed point with higher firing rate. The two horizontal arrows correspond to two input currents with different amplitudes. (f) Type II. Same as in A for type II response. The model transitions from a stable fixed point to a stable focus regime with higher firing rate and high damping factor.
Stability of single-neuron NH-PDMP models: (a) Lambert W{0,1}(x) function. Hatched corresponds to the region with two real branches. (b) Parameter space of the deterministic part of the diffusion SDEs for a history kernel with two exponential functions with parameters β1 and β2, respectively. Colored regions indicate different dynamical regimes of existence and stability of the first fixed point obtained by analyzing the deterministic part of diffusion SDEs (eq. (4)). Hatch indicate the existence of an unstable second fixed point. Colored diamonds indicate different parameter choices for the examples shown in (d). (c) Parameter space of the moment equations of the corresponding full diffusion SDE. Colored regions indicate different stability regimes. Black circles indicate divergence of the conditional intensity (and number of spikes) in the exact continuous-time simulations of the NH-PDMP models (Eq. 2). The length of the circle indicates the number of initial conditions leading to diverging simulations (Methods). (d) Each column shows an example for different choice of the β1 and β2 parameters (colored diamonds in the parameter space: (1) β1 =−1,β2 =−1;(2)β1 =2,β2 =−0.1;(3)β1 =5,β2 =−5;(4)β1 =14, β2 =−1. Firstrow: history kernels with two exponential functions defined by the above corresponding two parameters. Second row: 5 simulations of the corresponding NH-PDMP models (Eq. 2). Third row: state space spanned by the two variables z1 and z2 in the diffusion SDEs. Full circles indicate stable fixed points, colored lines are nullclines, and black arrowed lines show example trajectories.
Nonlinear-Hawkes piecewise deterministic Markov process (NH-PDMP) models for spiking neuronal networks: (a) Schematic of a neuronal network based on NH-PDMP models with inputs. (b) The diffusion approximation transforms the SDE representation of the NH-PDMP into diffusion SDEs. (c) The dynamics of the diffusion SDEs can be analyzed qualitatively and quantitatively using dynamical system theory tools, such as state space and bifurcation analyses.
Smothered mate with knight and queen
A picture of a leaf taken by smartphone and in macro mode. The fractal patterns are visible.
A bird in the sky. The photo is from an upper angle as there's a valley underneath it.