1 / 65

Computing with spikes

Computing with spikes. Romain Brette Ecole Normale Supérieure. White Nights of Computational Neuroscience 2012. Spiking neuron models. Spiking neuron models. Output = 1 spike train. Input = N spike trains. What transformation ?. Synaptic integration.

huslu
Download Presentation

Computing with spikes

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Computing with spikes Romain Brette Ecole Normale Supérieure White Nights of Computational Neuroscience 2012

  2. Spikingneuronmodels

  3. Spikingneuronmodels Output = 1 spike train Input = N spike trains What transformation ?

  4. Synapticintegration action potential or « spike » postsynapticpotential (PSP) spikethreshold temporal integration « Integrate and fire » model: spikes are producedwhen the membrane potentialexceeds a threshold

  5. The membrane • Lipidbilayer (= capacitance) with pores (channels = proteins) outside Na+ Cl- 2 nm inside K+ specific capacitance ≈ 1 μF/cm² total specific capacitance = specific capacitance * area

  6. The restingpotential • Atrest, the neuronispolarized: Vm≈ -70 mV • The membrane issemi-permeable • Mostlypermeable to K+ • A potentialdifferenceappearsand opposes diffusion Membrane potential Vm=Vin-Vout V diffusion K+

  7. Electrodiffusion • Membrane permeable to K+ only • Diffusion creates an electricalfield • Electricalfield opposes diffusion • Equilibriumpotential or « Nernst potential »: or reversal potential extra intra F = 96 000 C.mol-1 (Faraday constant) R = 8.314 J.K-1.mol-1 (gas constant) z = charge of ion

  8. The equivalentelectrical circuit I = capacitance leak or restingpotential Linear approximation of leakcurrent: I = gL(Vm-EL) leak conductance = 1/R membrane resistance EL-70 mV : the membrane is « polarized » (Vin < Vout)

  9. The membrane equation Iinj outside =1/R Iinj Vm inside membrane time constant (typically 3-100 ms)

  10. Synapticcurrents synapse synapticcurrent postsynapticneuron Is(t)

  11. Postsynapticpotentials (PSPs) Presynapticneuron (extracellularelectrode) Postsynapticneuron (intracellularelectrode) (cat motoneuron)

  12. Idealized synapse • Total charge • Opens for a short duration • Is(t)=Qδ(t) Dirac function EL Spike-based notation: w=RQ/τis the « synapticweight » at t=0

  13. Temporal and spatial integration • Response to a set of spikes {tij} ? • Linearity: i = synapse j = spikenumber Superposition principle

  14. Synapticintegration and spikethreshold action potential PSP spikethreshold « Integrate-and-fire »: If V = Vt (threshold) then: neuronspikes and V→Vr (reset)

  15. The integrate-and-fire model Differential formulation Integral formulation spikeat synapse i If V = Vt (threshold) then: neuronspikes and V→Vr (reset)

  16. Is this a sensible description of neurons?

  17. A phenomenologicalapproach Fittingspikingmodels to electrophysiologicalrecordings Injectedcurrent Recording Model Rossant et al. (Front. Neuroinform. 2010) (IF with adaptive thresholdfittedwith Brian + GPU)

  18. Results: regularspikingcell Winner of INCF competition: 76% (variation of adaptive threshold IF) Rossant et al. (2010). Automatic fitting of spiking neuron models to electrophysiological recordings (Frontiers in Neuroinformatics)

  19. Good news  Adaptive integrate-and-firemodels are excellent phenomenologicalmodels! (response to somatic injection)

  20. Advanced concepts • Synapticchannels are alsodescribed by electrodiffusion • Neurons are not isopotential gs(t) ionicchannel conductance open closed synaptic reversal potential presynapticspike The « cableequation »

  21. Simulatingspikingmodelswith Goodman, D. and R. Brette (2009). The Brian simulator. Front Neurosci doi:10.3389/neuro.01.026.2009. frombrianimport * 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,model=eqs, • threshold=’v>-50*mV’,reset=’v=-60*mV’) P.v=-60*mV+10*mV*rand(len(P)) 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() Ci P Pi Pe Ce briansimulator.org

  22. Computingwithspikes I:rate-basedtheories

  23. Statistics of spike trains • Spike train: • A sequence of spike times (tk) • A signal • Rate: • Number of spikes / time (= lim k/tk) • Average of S: tn+1 – tn = « interspikeinterval » (ISI) t1 t2 t3 (Up to a few hundred Hz)

  24. Rate-based descriptions Inputs with rates F1, F2, ..., Fn F1 Is(t) = total synapticcurrent F2 F3 F4 Is(t) F Rate F Can we express F as a function of F1, F2, ..., Fn?

  25. Approach #1: the « perfectintegrator » • Neglect the leakcurrent: • More precise: replace Vm by

  26. The perfectintegrator • Normalization • Vt=1, Vr=0 et si V=Vt: V → Vr = change of variable for V: V* = (V-Vr)/(Vt-Vr) Same for I

  27. The perfectintegrator • Integration: • Firing rate: if otherwise Hz Brette, R. (2004). Dynamics of one-dimensional spiking neuron models. J Math Biol

  28. The perfectintegratorwithsynaptic inputs (superposition principle) timing of spikeat synapse k constant (from change of variables) Jk = postsynapticcurrent (= 0 for t<0)

  29. The perfectintegratorwithsynaptic inputs • Input firing rates F1, F2, ..., Fn. • Let the « synapticweight » of synapse k • Output firing rate is ([x]+ = max(x,0)) Proof: first prove where F is the rate of events (tj)

  30. Approach #2: meanfield • Meanfield approximation: • replace I(t) by itsmean • use the current-frequencyfunction: F=f(I) where

  31. Meanfield vs. perfectintegrator meanfield: neglect variations of I(t) perfectintegrator: neglect the leak

  32. Approach #3: Poisson inputs • What if ? (balancedregime) • Assumption: input spike trains are independent Poisson processeswith rates Fi F1 F2 Fn ... w w identical synapses w If Ti = Poisson with rate Fi, thenUFi = Poisson with rate Fi Conclusion: Fout = f(Fi)

  33. Summary: • Neglectleak: • Meanfield • Independent inputs (+Poisson +1 synapse type) integral of synapticcurrent (perfectintegrator) current-frequencyfunction (meanfield) transmission probability undeterminedfunction Variation: diffusion Otherwisetheremight not be a univocal input-output rate relationship

  34. Computingwithspikes II:asynchronoustheories

  35. The precision of spike timing The same variable currentisinjected 25 times. Spike timing is reproductible evenafter 1 s. IF model: Mainen & Sejnowski (1995) (cortical neuron in vitro, somatic injection)

  36. The neural « code » Time Rank Count • Code: • spike count • (rate code) • spike timing • (temporal code) • spikerank • (rankordercode) Thorpe et al (2001), Spike-based strategies for rapid processing. Neural networks.

  37. Decodingrankorder • How to distinguishbetween AB and BA? • Solution: excitation and inhibition A B Excitatory PSP Inhibitory PSP - - + +

  38. Prey localization by the sand scorpion Inhibition of opposite neuron → more spikesnear the source (polar representation of firing rates) Conversion rankorder code → rate code Stürtzl et al. (2000). Theory of arachnidpreylocalization. PRL

  39. Predictivecoding input output linearread-out Each output neuronspikes to reducesomeerrordefined on the read-out References: Boerlin & Denève (2011). Spike-Based Population Coding and Working Memory. PLoS Comp Biol S Denève (2008). Bayesian spiking neurons I: inference. Neural Computation S Denève (2008). Bayesian spiking neurons II: learning. Neural Computation

  40. Computingwithspikes III:synchrony-based computation

  41. Reliability of spike timing In spiking model: Z. Mainen, T. Sejnowski, Science (1995) Spike timing isreproducible in vitro for time-varying inputs Brette, R. and E. Guigon (2003). Reliability of spike timing is a general property of spiking model neurons. Neural Comput Brette (2012). Computing with neural synchrony. PLoS Comp Biol

  42. Selectivesynchronization Consequence: Similarneuronssynchronize to similar time-varying inputs ? What impact on postsynapticneurons?

  43. Coincidencedetection: principle ThresholdVt d Same rate F Delay d threshold threshold spike no spike Spike if

  44. Coincidencedetection in noisyneurons How about in realistic situations? Balancedregime = VmDpeaksbelowthreshold Rossant C, Leijon S, Magnusson AK, Brette R (2011). Sensitivity of noisyneurons to coincident inputs. J Neurosci

  45. Synchrony-based computation A B C no response D signalssimilaritiesbetween A and C

  46. Synchronyreceptivefields A B no response « Synchronyreceptivefield » = {S | NA(S) = NB(S)} Whatdoesitrepresent?

  47. Synchronyreceptivefields: examples Synchronywhen: S(t-dR-δR)=S(t-dL-δL) dR-dL = δR +δL Independent of source signal

  48. Synchronyreceptivefields: examples Synchronywhen: S(t-δA)=S(t-δB) PeriodicsoundwithperiodδA-δB Synchronyreceptivefields signal someregularity or « structure » in the sensorysignals

  49. Structure in stimuli A structured stimulus: M The transformation M introduces structure in sensorysignals S=M(X) X Sensory signal Object in environment Examples: Binaural hearing M source sound S M is location-specific (FL*S,FR*S) stimulus Pitch M M is pitch-specific sound in phase space

  50. Structure and synchrony dR-dL = δR +δL Brette (2012). Computing with neural synchrony. PLoS Comp Biol

More Related