Transcript BRIAN Simulator

BRIAN Simulator
11/4/11
NEURON is cool, but…
• …it’s not suited particularly well for large
network simulations
• What if you want to look at properties of
thousands of neurons interacting with one
another?
• What about changing properties of synapses
through time?
BRIAN Simulator
• BRIAN allows for efficient simulations of large
neural networks
• Includes nice routines for setting up random
connectivity, recording spike times, changing
synaptic weights as a function of activity
• www.briansimulator.org
– Should be able to run from the unzipped brian
directory
– http://www.briansimulator.org/docs/installation.html
Make sure it works!
from brian import *
brian_sample_run()
Building blocks of BRIAN
• Model consists of:
– Equations
– NeuronGroup – a group of neurons which obeys
those equations
– Connection – a way to define connection matrices
between neurons
– SpikeMonitor (and others) – a way to measure
properties of the simulated network
My First BRIAN Model
• To start, let’s walk
through the example
code on
briansimulator.org
homepage
• Copy this text into IDLE
(or whichever .py editor
you’re using)
• Make sure it runs first!
from brian import *
eqs = '''
dv/dt = (ge+gi-(v+49*mV))/(20*ms) :
volt
dge/dt = -ge/(5*ms) : volt
dgi/dt = -gi/(10*ms) : volt
'''
P = NeuronGroup(4000, eqs, threshold=50*mV, reset=-60*mV)
P.v = -60*mV
Pe = P.subgroup(3200)
Pi = P.subgroup(800)
Ce = Connection(Pe, P, 'ge',
weight=1.62*mV, sparseness=0.02)
Ci = Connection(Pi, P, 'gi', weight=9*mV, sparseness=0.02)
M = SpikeMonitor(P)
run(1*second)
raster_plot(M)
show()
My First BRIAN Model
•
•
•
First – we set up the model
Notice the units! These are important
in BRIAN, they help ensure
everything you’re modeling makes
sense
Equations are written as strings,
these are executed by the differential
equation solver in BRIAN
– Unit names/abbreviations are
reserved keywords in BRIAN
•
Create our NeuronGroup using this
model
– Define number of neurons, model
used, threshold & reversal potentials
•
Questions:
– What is the reversal potential here?
– How does this model differ from HH?
from brian import *
eqs = '''
dv/dt = (ge+gi-(v+49*mV))/(20*ms) :
volt
dge/dt = -ge/(5*ms) : volt
dgi/dt = -gi/(10*ms) : volt
'''
P = NeuronGroup(4000, eqs, threshold=50*mV, reset=-60*mV)
P.v = -60*mV
Pe = P.subgroup(3200)
Pi = P.subgroup(800)
Ce = Connection(Pe, P, 'ge',
weight=1.62*mV, sparseness=0.02)
Ci = Connection(Pi, P, 'gi', weight=9*mV, sparseness=0.02)
M = SpikeMonitor(P)
run(1*second)
raster_plot(M)
show()
My First BRIAN Model
• Initialize the neurons to
starting potentials
• Connect them
– Here, constant weight,
random connectivity from
each subpopulation
(excitatory & inhibitory) to all
neurons
• Which state variable to
propagate to in “target”
neuron: ‘ge’ for excitatory
synapses, ‘gi’ for inhibitory
from brian import *
eqs = '''
dv/dt = (ge+gi-(v+49*mV))/(20*ms) :
volt
dge/dt = -ge/(5*ms) : volt
dgi/dt = -gi/(10*ms) : volt
'''
P = NeuronGroup(4000, eqs, threshold=50*mV, reset=-60*mV)
P.v = -60*mV
Pe = P.subgroup(3200)
Pi = P.subgroup(800)
Ce = Connection(Pe, P, 'ge',
weight=1.62*mV, sparseness=0.02)
Ci = Connection(Pi, P, 'gi', weight=9*mV, sparseness=0.02)
M = SpikeMonitor(P)
run(1*second)
raster_plot(M)
show()
My First BRIAN Model
• Hook up your
electrophysiology
equipment (here, measure
spike times)
– Can also record population
rate, ISI – search
documentation for Monitor
– Only this info is saved from
the simulation
• Run the simulation!
• Plot
– Other plotting tools
(hist_plot, Autocorrelograms,
pylab, etc)
from brian import *
eqs = '''
dv/dt = (ge+gi-(v+49*mV))/(20*ms) :
volt
dge/dt = -ge/(5*ms) : volt
dgi/dt = -gi/(10*ms) : volt
'''
P = NeuronGroup(4000, eqs, threshold=50*mV, reset=-60*mV)
P.v = -60*mV
Pe = P.subgroup(3200)
Pi = P.subgroup(800)
Ce = Connection(Pe, P, 'ge',
weight=1.62*mV, sparseness=0.02)
Ci = Connection(Pi, P, 'gi', weight=9*mV, sparseness=0.02)
M = SpikeMonitor(P)
run(1*second)
raster_plot(M)
show()
My First BRIAN Model
from brian import *
eqs = '''
dv/dt = (ge+gi-(v+49*mV))/(20*ms) : volt
dge/dt = -ge/(5*ms) : volt
dgi/dt = -gi/(10*ms) : volt
'''
P = NeuronGroup(4000, eqs, threshold=-50*mV, reset=60*mV)
P.v = -60*mV
Pe = P.subgroup(3200)
Pi = P.subgroup(800)
Ce = Connection(Pe, P, 'ge', weight=1.62*mV,
sparseness=0.02)
Ci = Connection(Pi, P, 'gi', weight=-9*mV,
sparseness=0.02)
M = SpikeMonitor(P)
R = PopulationRateMonitor(P,.01*second)
H = ISIHistogramMonitor(P,bins =
[0*ms,5*ms,10*ms,15*ms,20*ms,25*ms,30*ms,35*ms,40*m
s])
run(1*second)
raster_plot(M)
hist_plot(H)
plt.figure()
plt.plot(R.times,R.rate)
plt.xlabel(‘time (s)’)
plt.ylabel(‘firing rate (Hz)’)
show()
Exercise: Can you make HW 4’s
networks in BRIAN?
• Try with both HH-style and
integrate and fire models
• Use documentation at
briansimulator.org for help
(especially Connection and
Equations)
– Hint: search HodgkinHuxley
and Library Models
• Let’s use simplified
exponential synapses
– Time constant tau = 3 ms for
excitatory, 6 ms for inhibitory
Feedforward
Inhibition
Feedback
Inhibition
Exercise: Can you make HW 4’s
networks in BRIAN?
Feedforward
Inhibition
Feedback
Inhibition
Spike timing dependent plasticity
Final weight
time
Synapse number
Count
• Sets up network of
Poisson spiking
neurons, all projecting
to a single neuron
• Each synapse
implements an STDP
rule
• By end of simulation,
synapses become either
strong or weak
Firing rate (Hz)
stdp1.py
Final weight
Final weight
time
Synapse number
Count
• Only difference is that
we’ve changed how
pre-before-post and
post-before-pre
synapses are weighted
(and, in model code,
stdp2 uses
ExponentialSTDP, to
make things easier)
Firing rate (Hz)
stdp2.py
Final weight