Limited Offer Get 25% off — use code BESTW25
No AI No Plagiarism On-Time Delivery Free Revisions
Claim Now
Prices start at $5 per page · plagiarism-free papers · free revisions for 14 days 24/7 support: support@bestwriters.com
Uncategorized

the integrate-and-fire neuron model

  • Posted
  • Authoradmin
  • Length2,052 words
  • Reading time9 min
Write My Assignment FREE

Coursework 3

This coursework has two parts: part A on the integrate-and-fire neuron model (worth 25% of the coursework grade) and part B on spike-timing-dependentplasticity (worth 75% of the coursework grade).

Write a brief report, around five or six or seven pages, including the figures and the comments specified. No need to be verbose, just say what you are asked to say. We will take a dim view of super-narrow margins or tiny fonts. Submit it in the pdf format together with the Python or Julia or MATLAB code by the deadline. Submit the report as pdf; make sure to set the size of the graphs in your plotting program, not by shrinking the graph to fit in the report, this makes the figure legends and axis numbers tiny and is super annoying. You don’t have to plot in Python or Julia or MATLAB, it is fine to output a data file and plot using gnuplot or whatever. All plots should have clear x- and y-axis labels, and units. Four marks will be docked for every plot that has missing or unclear labels.

To help you decide which results to include in your report, I will encode meaning in my text colour and formatting. If a sentence is coloured magenta, I expect to see a written response. For sentences formatted with italics I expect to see a plot.

Part A: Integrate-and-fire neurons (25% of coursework grade)

Question 1 [10 marks]

Simulate an integrate and fire model with the following parameters for 1 s: τm = 10ms, EL = Vrest = −70 mV, Vth = −40 mV, Rm = 10 MΩ, Ie = 3.1 nA.

Here EL is the leak potential, Vrest is the reset voltage, Vth is the threshold, Rm is the membrane resistance, that is one over the conductance, and τm is the membrane time constant. Use Euler’s method with timestep δt = 0.25 ms to solve the differential equations numerically. Make sure to use the differential equations, do not just plot the analytical solution for V(t) given in the lectures/notes – this will be useless for later questions anyway.

Plot the voltage as a function of time. For simplicity assume that the neuron does not have a refractory period after producing a spike. You do not need to plot spikes – once membrane potential exceeds threshold, simply set the membrane potential to Vrest. To keep track of the units I recommend converting everything to S.I. units inside your simulation code – you don’t have to do this but it often makes things easier.

Question 2 [15 marks]

Simulate two neurons which have synaptic connections between each other, that is the first neuron projects to the second, and the second neuron projects to the first. Both model neurons should have the same parameters: τm = 20 ms, EL = −70 mV, Vrest = −80 mV, Vth = −54 mV, RmIe = 18 mV, and their synapses should also have the same parameters: Rmg¯s = 0.15, ∆s = 0.5, τs = 10 ms; don’t get confused by being given Rmg¯s rather than g¯s on its own, to get τm rather than the capacitance on the left hand side of the integrate and fire equation everything is multiplied by Rm. For simplicity take the synaptic conductance

gs = ¯gss (1)

to satisfy

(2)

with a spike arriving causing s to increase instantaneously by an amount ∆s. This is equivalent to the simplest synapse model in lecture 16. Simulate two cases: a) assuming that the synapses are excitatory with Es = 0 mV, and b) assuming that the synapses are inhibitory with Es = −80 mV. For each simulation set the initial membrane potentials of the neurons V to different values chosen randomly from between Vrest and Vth and simulate 1 s of activity. For each case plot the voltages of the two neurons on the same graph (with different colours). Comment on what is happening in each case. Offer an explanation for what you observe.

[15 marks]

This question uses parameters from Q1 above.

  1. Compute analytically the minimum current Ie required for the neuron with the above parameters to produce an action potential.
  2. Simulate the neuron for 1 s for the input current with amplitude Ie which is 0.1 [nA] lower than the minimum current computed above, and plot the voltage as a functions of time.
  3. Simulate the neuron for 1s for currents ranging from 2 [nA] to 5 [nA] in steps of 0.1 [nA]. For each amplitude of current count the number of spikes produced, that is the firing rate. Plot the firing rate as the function of the input current.

Part B: STDP (75% of coursework grade)

Now make a new simulation of a single leaky-integrate-and-fire neuron (you can reuse appropriate bits of your code from Part A). Set the resting potential

Vrest = EL = −65 mV, spike threshold Vth = −50 mV, reset voltage Vreset = Vrest = −65 mV, a passive membrane leak conductance that gives an input resistance of Rm = 100 MΩ, and τm = 10 ms. Note that these values are different to those used above in Part A. Don’t include a background input current, Ie = 0. Add N = 40 incoming synapses (just model the synaptic responses to spike trains, do not explicitly simulate the somatic voltages of these presynaptic neurons). Model all synapses as a conductance-based, following a single-exponential timecourse with decay time constant τs = 2 ms, the initial peak conductances (strengths) all g¯i = 4 nanoSiemens for all synapses, Es = 0, and ∆s = 0.5. Let si(t = 0) = 0 for all 40 synapses. You should have 40 different si variables and 40 different g¯i variables in total, since you have 40 synapses. It’s probably easier to store them all in arrays and update them together, rather than code each one by hand. Use the Euler method to solve the system of ODEs (40 s variables plus 1 voltage variable), with a simulation timestep δt = 0.25 milliseconds.

Simulate each input spike train as an independent homogeneous Poisson process with the same average firing rate hri. A homogeneous Poisson process implies that average firing rate hri is fixed in time, but that the spikes themselves arrive randomly. Initially set hri = 15 Hz. There are multiple ways to simulate Poisson spike trains. In principle you could use the function from the previous coursework 2 to pre-compute poisson spike times, but for the current exercise it will be easier to do it online during the simulation.

One simple method is to draw a random number on the unit interval for each synapse at every timestep, then if that number is less than hriδt, assume a spike has occurred at that synapse. This method is valid as long as.

Question 1 [15 marks]

Initially set all the synapses at the same strength, equal to the maximal conductance (4 nS). Simulating your model at this stage should result in the neuron firing irregularly at a rate of roughly 20 Hz. Plot the neuron’s voltage for one second of simulation time.

Question 2 [20 marks]

Now add STDP to the model using the following basic rule. First, let’s denote a presynaptic spike time as tpre, postsynaptic spike time as tpost, and the difference in pre-post spike times as ∆t = tpost −tpre (so that pre-beforepost timings are positive, while post-before-pre timings are negative). Now the rule for changing the synaptic conductance g¯syn → g¯syn + f(∆t), with

) if ∆t > 0 ) if ∆t ≤ 0

Initially set the four STDP model parameters as: A+ = 0.2 nS , A− = 0.25 nS, τ+ = 20 ms, τ− = 20 ms. The synaptic strength updates should follow the ‘nearest neighbour’ principle: only include the most recent pre and post spikes in your weight update calculations. Finally, impose hard limits on the synaptic strengths during the entire simulations: if an update would make a synaptic weight negative, set it to zero; if an update would make the synaptic weight greater than 4 nS, cap it at exactly 4 nS instead.

One convenient trick that simplifies programming is to note that potentiation happens at all synapses whenever a post-synaptic spike occurs, whereas depression happens at the synapses that receive a pre-synaptic spike.

Include a flag in your code that lets you switch the simulation mode from having STDP ‘on’ or ‘off’. STDP ‘on’ mode means that every spike (pre and post) triggers changes in the maximum conductances of the activated synapses according to the above rule, while STDP ‘off’ mode means that synaptic strengths are fixed and remain stuck at the same values throughout the simulation.

Now set the STDP flag to ‘on’, initialise all the synaptic strengths to 4 nS and set the input firing rates hri = 15 Hz. Run the simulation for 300 s of biological time. What qualitative shape does the synaptic strength distribution converge towards at the end of the simulation time? (You might need to run the simulation a few times to see the pattern). Plot a histogram of the steady-state synaptic weights after one run of the simulation. Also plot the average firing rate of the postsynaptic neuron as a function of time across the entire 300 second simulation (taking 10-second time bins). Now repeat the same simulation with STDP turned ‘off’, but with all synaptic strengths set equal to the mean value of the synaptic strengths you measured in the STDP ‘on’ simulations. Report the steady-state firing rate (as averaged over the last 30 seconds of the simulation) for both the STDP ‘on’ and ‘off’ simulation modes. To get a reasonable estimate for the firing rate, you may need to average over multiple realisations of the simulations. [Plotting the strength of each synapse as a function of time during the simulation can help you figure out what is going on.]

Question 3 [20 marks]

Vary the input firing rates from 10 Hz up to 20 Hz, both for a set of simulations with STDP switched on and for a set of simulations with STDP switched off. For both cases initialise the synaptic strengths at their maximum value, 4 nS. How does the steady-state output firing rate depend on the input firing rates in both cases? Plot the mean output firing rate as a function of the input firing rates for both cases. Plot the steady-state synaptic strength distribution for hri = 10 Hz and hri = 20 Hz for the ‘STDP on’ case. Give an explanation of what is happening and why you think it makes sense.

Question 4 [20 marks]

Up to now the input spike trains have been independent, and hence uncorrelated with each other. Now we will add correlations. Do this by making the firing rates of the input spike trains vary as a sinusoidal function of time:

hri(t) = hri0 + B sin(2πft)

where hri0 is the average firing rate. Now since the input firing rates are co-varying in time, the input spike trains will be temporally correlated. The degree of correlation depends on the parameter B, while the temporal extent of correlations depends on f. Fix f = 10 Hz, and hri0 = 20 Hz. Vary the degree of correlations by varying B from 0 to 20 Hz in steps of 5 Hz, so: B = [0,5,10,15,20] Hz. How does the degree of correlation affect the steady-state synaptic weights? Plot the mean and standard deviation of the steady-state synaptic strengths as a function of B. Plot example histograms of the steady-state synaptic strengths for B = 0 and B = 20 Hz. Give an explanation for what you think is happening.

[20 marks]

Now set B = 0 and r0 = 15 Hz again, and simulate for 300 s with STDP

‘on’. Measure the temporal cross-correlogram ρi(∆t) between all presynaptic neurons and the postsynaptic neuron (where the subscript i indexes the presynaptic neuron), and take the average:

, separately for the spikes from the first 20 seconds of the simulation and again for the spikes from the last 100 seconds of the simulation. Both these cross-correlograms should cover the interval -50 to +50 ms in steps of 1 ms.

Plot the cross-correlogram for both cases (start and end of the simulation) on the same plot. State what the curves look like, and what is different between the two curves, noting in particular what happens near ∆t = 0. How does this relate to the stabilisation of firing rates and synaptic strengths that occurs during the simulation? Give your explanation for what is happening, referring back to the STDP curve f(∆t). [hint: note that A− > A+]

The post the integrate-and-fire neuron model appeared first on My Assignment Online.

Plagiarism Free Assignment Help

Expert Help With This Assignment — On Your Terms

✓ Native UK, USA & Australia writers ✓ Deadline from 3 hours ✓ 100% Plagiarism-Free — Turnitin included ✓ Unlimited free revisions ✓ Free to submit — compare quotes

admin

Academic writer and editor at BestWriters.

Get your paper off your plate tonight

Post your instructions for free. Prices start at $5 per page.

Place an order