Motor system model with reinforcement learning drives virtual arm (Dura-Bernal et al 2017)

 Download zip file   Auto-launch 
Help downloading and running models
Accession:194897
"We implemented a model of the motor system with the following components: dorsal premotor cortex (PMd), primary motor cortex (M1), spinal cord and musculoskeletal arm (Figure 1). PMd modulated M1 to select the target to reach, M1 excited the descending spinal cord neurons that drove the arm muscles, and received arm proprioceptive feedback (information about the arm position) via the ascending spinal cord neurons. The large-scale model of M1 consisted of 6,208 spiking Izhikevich model neurons [37] of four types: regular-firing and bursting pyramidal neurons, and fast-spiking and low-threshold-spiking interneurons. These were distributed across cortical layers 2/3, 5A, 5B and 6, with cell properties, proportions, locations, connectivity, weights and delays drawn primarily from mammalian experimental data [38], [39], and described in detail in previous work [29]. The network included 486,491 connections, with synapses modeling properties of four different receptors ..."
Reference:
1 . Dura-Bernal S, Neymotin SA, Kerr CC, Sivagnanam S, Majumdar A, Francis JT, Lytton WW (2017) Evolutionary algorithm optimization of biological learning parameters in a biomimetic neuroprosthesis. IBM Journal of Research and Development (Computational Neuroscience special issue) 61(2/3):6:1-6:14
Citations  Citation Browser
Model Information (Click on a link to find other models with that property)
Model Type: Realistic Network;
Brain Region(s)/Organism:
Cell Type(s): Abstract Izhikevich neuron;
Channel(s):
Gap Junctions:
Receptor(s): GabaA; GabaB; NMDA; AMPA;
Gene(s):
Transmitter(s): Glutamate; Gaba;
Simulation Environment: NEURON; Python;
Model Concept(s): Learning; Reinforcement Learning; Reward-modulated STDP; STDP; Motor control; Sensory processing;
Implementer(s): Dura-Bernal, Salvador [salvadordura at gmail.com]; Kerr, Cliff [cliffk at neurosim.downstate.edu];
Search NeuronDB for information about:  GabaA; GabaB; AMPA; NMDA; Gaba; Glutamate;
"""
STIMULI

This code defines properties of different type of stimuli
and defines functions to generate the stimuli

"""

from pylab import array, rand, exp, zeros, hstack
import shared as s # Import population and connection data

#PMd, ASC, DSC, ER2, IF2, IL2, ER5, EB5, IF5, IL5, ER6, IF6, IL6, AMPA, NMDA, GABAA, GABAB, opsin, Epops, Ipops, allpops = cpd.names2inds() # Define populations

# Class with natural touch stimulus
class touch:
    name = 'touch'
    receptor = s.AMPA  # Set which receptor to stimulate
    isi = 0.5 # Interstimulus interval in s
    var = 0.0 # Variation in ISI in s
    width = 0.005 # Stimulus width in s
    weight = 100 # Weight of stimuli
    sta = [] # Starting time in s -- usually defined below
    fin = [] # Finishing time in s
    shape = 'gaussian' # Shape of stimulus
    #pops = [TCR]
    fraction = 1 # No idea what this should be...
    rate = 200 # 500 is the highest rate that can be used without significantly slowing down the simulation
    noise = 0 # Variability in onset time
    loc = array([[0.4,0.6],[0.4,0.6]]) # Location of cells to be stimulated
    falloff = [array([0.5,0.5]), 1e9] # Weight fall-off characteristics of the stimulus: center and Gaussian width in um

# class with optogenetic stimulus
class opto:
    name = 'opto'
    receptor = s.opsin # Set which receptor to stimulate
    isi = 0.100 # Interstimulus interval in s
    var = 0.0 # Variation in ISI
    width = 0.2 # Stimulus width in s
    weight = 10 # Weight of stimuli
    sta = 3 # Starting time in s
    fin = 6 # Finishing time in s
    shape = 'square' # Shape of stimulus
    pops = [s.ER5, s.EB5, s.ER6]
    fraction = 0.4 # No idea what this should be...
    rate = 500 # 500 is the highest rate that can be used without significantly slowing down the simulation
    noise = 0
    loc = array([[0.0,1.0],[0.0,1.0]]) # Location of cells to be stimulated
    falloff = [array([0.5,0.5]), 2000] # Weight fall-off characteristics of the stimulus: center and Gaussian width in um


# Allow defaults to be overriden by name, e.g. instead of touch(), stimmod(touch,sta=1,fin=3)
def stimmod(stimclass, name=None, receptor=None, isi=None, var=None, width=None, weight=None, sta=None, fin=None, shape=None, pops=None, fraction=None, rate=None, noise=None, loc=None, falloff=None):
    stiminstance = stimclass()    
    stimproperties = ['name', 'receptor', 'isi', 'var', 'width', 'weight', 'sta', 'fin', 'shape', 'pops', 'fraction', 'rate', 'noise', 'loc', 'falloff'];
    for prop in stimproperties:
        thisproperty = eval(prop)
        if thisproperty!=None:
            exec('stiminstance.' + prop + ' = thisproperty')
    return stiminstance


## Define stimulus-making code
def makestim(isi=1, variation=0, width=0.05, weight=10, start=0, finish=1, stimshape='gaussian'):
    from pylab import r_, convolve, shape
    
    # Create event times
    timeres = 0.005 # Time resolution = 5 ms = 200 Hz
    pulselength = 10 # Length of pulse in units of width
    currenttime = 0
    timewindow = finish-start
    allpts = int(timewindow/timeres)
    output = []
    while currenttime<timewindow:
        if currenttime>=0 and currenttime<timewindow: output.append(currenttime)
        currenttime = currenttime+isi+variation*(rand()-0.5)
    
    # Create single pulse
    npts = min(pulselength*width/timeres,allpts) # Calculate the number of points to use
    x = (r_[0:npts]-npts/2+1)*timeres
    if stimshape=='gaussian': 
        pulse = exp(-(x/width*2-2)**2) # Offset by 2 standard deviations from start
        pulse = pulse/max(pulse)
    elif stimshape=='square': 
        pulse = zeros(shape(x))
        pulse[int(npts/2):int(npts/2)+int(width/timeres)] = 1 # Start exactly on time
    else:
        raise Exception('Stimulus shape "%s" not recognized' % stimshape)
    
   # Create full stimulus
    events = zeros((allpts))
    events[array(array(output)/timeres,dtype=int)] = 1
    fulloutput = convolve(events,pulse,mode='same')*weight # Calculate the convolved input signal, scaled by rate
    fulltime = (r_[0:allpts]*timeres+start)*1e3 # Create time vector and convert to ms
    fulltime = hstack((0,fulltime,fulltime[-1]+timeres*1e3)) # Create "bookends" so always starts and finishes at zero
    fulloutput = hstack((0,fulloutput,0)) # Set weight to zero at either end of the stimulus period
    events = hstack((0,events,0)) # Ditto
    stimvecs = [fulltime, fulloutput, events] # Combine vectors into a matrix
    
    return stimvecs