Synaptic scaling balances learning in a spiking model of neocortex (Rowan & Neymotin 2013)

 Download zip file   Auto-launch 
Help downloading and running models
Accession:147141
Learning in the brain requires complementary mechanisms: potentiation and activity-dependent homeostatic scaling. We introduce synaptic scaling to a biologically-realistic spiking model of neocortex which can learn changes in oscillatory rhythms using STDP, and show that scaling is necessary to balance both positive and negative changes in input from potentiation and atrophy. We discuss some of the issues that arise when considering synaptic scaling in such a model, and show that scaling regulates activity whilst allowing learning to remain unaltered.
Reference:
1 . Rowan MS,Neymotin SA (2013) Synaptic Scaling Balances Learning in a Spiking Model of Neocortex Adaptive and Natural Computing Algorithms, Tomassini M, Antonioni A, Daolio F, Buesser P, ed. pp.20
Model Information (Click on a link to find other models with that property)
Model Type: Realistic Network;
Brain Region(s)/Organism: Neocortex;
Cell Type(s): Neocortex V1 pyramidal corticothalamic L6 cell; Neocortex V1 pyramidal intratelencephalic L2-5 cell; Neocortex V1 interneuron basket PV cell; Neocortex fast spiking (FS) interneuron; Neocortex spiny stellate cell; Neocortex spiking regular (RS) neuron; Neocortex spiking low threshold (LTS) neuron; Abstract integrate-and-fire adaptive exponential (AdEx) neuron;
Channel(s):
Gap Junctions:
Receptor(s): GabaA; AMPA; NMDA;
Gene(s):
Transmitter(s): Gaba; Glutamate;
Simulation Environment: NEURON; Python;
Model Concept(s): Synaptic Plasticity; Long-term Synaptic Plasticity; Learning; STDP; Homeostasis;
Implementer(s): Lytton, William [billl at neurosim.downstate.edu]; Neymotin, Sam [samn at neurosim.downstate.edu]; Rowan, Mark [m.s.rowan at cs.bham.ac.uk];
Search NeuronDB for information about:  Neocortex V1 pyramidal corticothalamic L6 cell; Neocortex V1 pyramidal intratelencephalic L2-5 cell; Neocortex V1 interneuron basket PV cell; GabaA; AMPA; NMDA; Gaba; Glutamate;
/
stdpscalingpaper
mod
infot.mod
intf6.mod
intfsw.mod *
misc.mod *
myfft.mod *
nstim.mod *
place.mod *
sampen.mod *
staley.mod *
stats.mod *
tsa.mod *
updown.mod *
vecst.mod *
bpf.h *
misc.h *
mkmod *
parameters.multi *
                            
: $Id: updown.mod,v 1.16 2009/02/16 22:56:52 billl Exp $

NEURON {
  SUFFIX nothing
  : BVBASE is bit vector base number (typically 0 or -1)
  GLOBAL UPDOWN_INSTALLED, SHM_UPDOWN, NOV_UPDOWN, DEBUG_UPDOWN
}

PARAMETER {
  UPDOWN_INSTALLED=0
  DEBUG_UPDOWN=0
  SHM_UPDOWN=4   : used in updown() for measuring sharpness
  NOV_UPDOWN=1   : used in updown() to eliminate overlap of spikes
  CREEP_UPDOWN=0 : used in updown() to allow left/right "creep" to local minima
}

VERBATIM
#include <stdlib.h>
#include <math.h>
#include <limits.h> // contains LONG_MAX 
#include <sys/time.h> 
extern double* hoc_pgetarg();
extern double hoc_call_func(Symbol*, int narg);
extern FILE* hoc_obj_file_arg(int narg);
extern Object** hoc_objgetarg();
extern void vector_resize();
extern int vector_instance_px();
extern void* vector_arg();
extern double* vector_vec();
extern double hoc_epsilon;
extern double chkarg();
extern void set_seed();
extern int ivoc_list_count(Object*);
extern Object* ivoc_list_item(Object*, int);
extern int hoc_is_double_arg(int narg);
extern char* hoc_object_name(Object*);
char ** hoc_pgargstr();
int list_vector_px();
int list_vector_px2();
int list_vector_px3();
double *list_vector_resize();
int ismono1();
static void hxe() { hoc_execerror("",0); }
static void hxf(void *ptr) { free(ptr); hoc_execerror("",0); }
ENDVERBATIM

:* src.updown(thresh,dlist,nqslist)
:  dest.updown(src)  -- default thresh=0; returns indices
: look for multiple threshold crossings to define peaks
: creates multiple parallel vectors for an NQS db
: counts peaks pointing upward -- should be all pos
: see eg decnqs.hoc:fudup() for usage
VERBATIM

  //** declarations
#define UDSL 500
#define UDNQ 11
// nq=new NQS("LOC","PEAK","WIDTH","BASE","HEIGHT","START","SLICES","SHARP","INDEX","FILE","NESTED")
#define LOC     nq[0] // loc of peak of spike
#define PEAK  	nq[1] // value at peak (absolute height)
#define WIDTH  	nq[2] // rt flank - lt flanks (? isn't it rt flank - LOC ?)
#define BASE  	nq[3] // height at base
#define HEIGHT  nq[4] // peak - base
#define START  	nq[5] // left flank of spike?
#define SLICES  nq[6] // how many slices found this spike
#define SHARP  	nq[7] // 2nd deriv at peak
#define INDEX  	nq[8] // consecutive numbering of spikes
//        	nq[9] // will use to fill in trace's file name at hoc level
#define NESTED  nq[10] // how many bumps are nested within this one
  //** procedure updown()
static double updown (void* vv) {
  int i, k, m, n, nqsz, nsrc, jj[UDSL], f[UDSL], lc, dsz[UDSL], nqmax, thsz, lc2, done, dbn;
  double *src, *tvec, *th, *dest[UDSL], *nq[UDNQ], *tmp, *dbx, lt, thdist;
  Object *ob, *ob2;
  void *vvd[UDSL], *vvth, *vnq[UDNQ];
  //** read in vectors and verify sizes, etc
  nsrc = vector_instance_px(vv, &src); // trace to analyze
  thsz = vector_arg_px(1, &th);        // vector of thresholds to check
  ob =  *hoc_objgetarg(2);             // storage for values for each threshold
  ob2 = *hoc_objgetarg(3);             // list of NQS vectors for returning values
  tmp = (double *)ecalloc(nsrc, sizeof(double));  // tmp is size of trace
  lc =  ivoc_list_count(ob);
  lc2 = ivoc_list_count(ob2);
  if (lc>UDSL) {printf("updown ERRF mismatch: max slice list:%d %d\n",UDSL,lc); hxf(tmp);}
  if (lc2!=UDNQ){printf("updown ERRB mismatch: NQS sz is %d (%d in list)\n",UDNQ,lc2);hxf(tmp);}
  if (nsrc<lc) {printf("updown ERRC mismatch: %d %d\n",lc,nsrc); hxf(tmp);} // ??
  if (lc!=thsz) {printf("updown ERRA mismatch: %d %d\n",lc,thsz); hxf(tmp);}
  if (!ismono1(th,thsz,-1)) {printf("updown ERRD: not mono dec %g %d\n",th[0],thsz); hxf(tmp);}
  // thdist=(th[thsz-2]-th[thsz-1])/2; // NOT BEING USED: the smallest spike we will accept
  for (k=0;k <lc;k++)  dsz[k] =list_vector_px3(ob , k, &dest[k], &vvd[k]);
  for (k=0;k<lc2;k++) {
    i=list_vector_px3(ob2, k, &nq[k],   &vnq[k]);
    if (k==0) nqmax=i; else if (i!=nqmax) { // all NQ vecs same size
      printf("updown ERRE mismatch: %d %d %d\n",k,i,nqmax); hxf(tmp); }
  }
  //** store crossing points and midpoints in dest[k]
  // dest vectors dest[k] will store crossing points and midpoints at each th[k] slice location
  // as triplets: up/max/down
  for (k=0; k<lc; k++) {   // iterate thru thresholds
    jj[k]=f[k]=0; // jj[k] is ind into dest[k]; f[k] is flag for threshold  crossings
    for (i=0;i<nsrc && src[i]>th[k];i++) {} // start somewhere below this thresh th[k]
    for (; i<nsrc; i++) { // iterate through trace
      if (src[i]>th[k]) { 
        if (f[k]==0) { // ? passing thresh 
          if (jj[k]>=dsz[k]){printf("(%d,%d,%d) :: ",k,jj[k],dsz[k]);
            hoc_execerror("Dest vec too small in updown ", 0); }
          dest[k][jj[k]++] = (i-1) + (th[k]-src[i-1])/(src[i]-src[i-1]); // interpolate
          f[k]=1; 
          tmp[k]=-1e9; dest[k][jj[k]]=-1.; // flag in tmp says that a thresh found here
        }
        if (f[k]==1 && src[i]>tmp[k]) { // use tmp[] even more temporarily
          tmp[k]=src[i]; // pick out max
          dest[k][jj[k]] = (double)i; // location of this peak
        }
      } else {          // below thresh 
        if (f[k]==1) {  // just passed going down 
          jj[k]++;      // triplet will be indices of cross-up/peak/cross-down
          dest[k][jj[k]++] = (i-1) + (src[i-1]-th[k])/(src[i-1]-src[i]);
          f[k]=0; 
        }
      }
    }
  }
  //** truncate dest vectors to multiples of 3:
  for (k=0;k<lc;k++) vector_resize(vvd[k],(int)(floor((double)jj[k]/3.)*3.));
  for (i=0; i<nsrc; i++) tmp[i]=0.; // clear temp space
  //** go through all the slices to find identical peaks and save widths and locations
  // tmp[] uses triplets centered around a location corresponding to a max loc in the
  // original vector; the widest flanks for each are then on either side of this loc
  for (k=0;k<lc;k++) { // need to go from top to bottom to widen flanks
    for (i=1;i<jj[k];i+=3) { // through centers (peaks)
      m=(int)dest[k][i]; // hash: place center at location
      if (tmp[m-2]<0 || tmp[m-1]<0 || tmp[m+1]<0 || tmp[m+2]<0) continue; // ignore; too crowded
      tmp[m]--;  // count how many slices have found this peak (use negative)
      tmp[m-1]=dest[k][i-1]; tmp[m+1]=dest[k][i+1]; // flanks
    }
  }
  //** 1st (of 2) loops through tmp[] -- pick up flanks
  // step through tmp[] looking for negatives which indicate the slice count and pick up 
  // flanks from these
  // nq=new NQS("LOC","PEAK","WIDTH","BASE","HEIGHT","START","SLICES","SHARP","INDEX","FILE")
  for (i=0,k=0; i<nsrc; i++) if (tmp[i]<0.) { // tmp holds neg of count of slices
    if (k>=nqmax) { printf("updown ERRG OOR in NQ db: %d %d\n",k,nqmax); hxf(tmp); }
    LOC[k]=(double)i;  // approx location of the peak of the spike
    WIDTH[k]=tmp[i+1]; // location of right side -- temp storage
    START[k]=tmp[i-1]; // start of spike (left side)
    SLICES[k]=-tmp[i];  // # of slices
    k++;
  }
  nqsz=k;   // k ends up as size of NQS db
  if (DEBUG_UPDOWN && ifarg(4)) { dbn=vector_arg_px(4, &dbx); // DEBUG -- save tmp vector
    if (dbn<nsrc) printf("updown ERRH: Insufficient room in debug vec (%d<%d)\n",dbn,nsrc); 
    else for (i=0;i<nsrc;i++) dbx[i]=tmp[i]; 
  }
  //** adjust flanks to handle nested bumps
  // 3 ways to handle spike nested in a spike or elongated base:
  // NB always using same slice for both L and R flanks; NOV_UPDOWN flag: (no-overlap)
  //   0. nested spike(s) share flanks determined by shared base
  //   1. nested spike(s) have individual bases, 1st and last use flanks from base
  //   2. nested spike(s) have individual bases, base flanks listed separately w/out peak
  // here use 
  // search nq vecs to compare flanks to neighboring centers
  // if flanks overlap the centers on LT or RT side,
  // correct them by going back to original slice loc info (in dest[])
  //*** look at left side -- is this flank to left of center of another bump?
  if (NOV_UPDOWN) for (i=0;i<nqsz;i++) { // iterate through NQS db
    if ((i-1)>0 && START[i] < LOC[i-1]) { // flank is to left of prior center
      if (DEBUG_UPDOWN) printf("LT problem %d %g %g<%g\n",i,LOC[i],START[i],LOC[i-1]);
      for (m=lc-1,done=0;m>=0 && !done;m--) { // m:go from bottom (widest) to top
        for (n=1;n<jj[m] && !done;n+=3) {     // n:through centers
          // pick out lowest slice with this peak LOC whose flank is to RT of prior peak
          if (floor(dest[m][n])==LOC[i] && dest[m][n-1]>LOC[i-1]) {
            // ??[i]=START[i]; // temp storage for L end of this overlap
            // replace both left and right flanks at this level -- #1 above
            START[i]=dest[m][n-1]; WIDTH[i]=dest[m][n+1]; done=1; 
          }
        }
      }
    }
    //*** now look at RT side
    if ((i+1)<nqsz && WIDTH[i]>LOC[i+1]) {
      if (DEBUG_UPDOWN) printf("RT problem %d %g %g>%g\n",i,LOC[i],WIDTH[i],LOC[i+1]);
      for (m=lc-1,done=0;m>=0 && !done;m--) { // m: go from bottom to top
        for (n=1;n<jj[m] && !done;n+=3) {     // n: through centers
          // pick out lowest slice with this peak LOC whose flank is to LT of next peak
          if (floor(dest[m][n])==LOC[i] && dest[m][n+1]<LOC[i+1]) {
            // ??[i]=WIDTH[i]; // end of overlap
            START[i]=dest[m][n-1]; WIDTH[i]=dest[m][n+1]; done=1;
          }
        }
      }        
    }
  }

  //make sure left and right sides of bump occur at local minima
  //shouldn't creeping be before NOV_UPDOWN=1 overlap check???
  //creeping can result only in equal borders btwn two bumps
  //on one side, so it should be ok here...
  if(CREEP_UPDOWN) for(i=0,k=0;i<nsrc;i++) if(tmp[i]<0.){

    //move left side to local minima
    int idx = (int)START[k];
    while(idx >= 1 && src[idx] >= src[idx-1]) idx--;
    START[k] = idx;

    //move right side to local minima
    idx = (int)WIDTH[k];
    while(idx < nsrc-1 && src[idx] >= src[idx+1]) idx++;
    WIDTH[k] = idx;

    k++;
  }

  //** 2nd loop through tmp[] used to fill in the rest of NQS
  // needed to split into 2 loops so that could check for overlaps and correct those
  // before filling in the rest of nq
  for (i=0,k=0; i<nsrc; i++) if (tmp[i]<0.) { // tmp holds neg of count of slices
    // calculate a base voltage lt as interpolated value on left side
    lt=src[(int)floor(START[k])]+(START[k]-floor(START[k]))*\
      (src[(int)floor(START[k]+1.)]-src[(int)floor(START[k])]);
    BASE[k]=lt;         // base voltage
    PEAK[k]=src[i];     // peak voltage
    WIDTH[k] = WIDTH[k] - START[k]; // width = RT_flank-LT_flank
    HEIGHT[k]=PEAK[k]-BASE[k]; // redund measure -- can eliminate
    // measure of sharpness diff of 1st derivs btwn peak and SHM_UPDOWN dist from peak
    // to get 2nd deriv would be normalized by 2*SHM_UPDOWN*tstep
    // ??could take an ave. or max first deriv for certain distance on either side
    SHARP[k]=(src[i]-src[i-(int)SHM_UPDOWN])-(src[i+(int)SHM_UPDOWN]-src[i]);
    INDEX[k]=(double)k;
    k++;
  }
  
  int iNumBumps = k;

  //count # of other bumps nested within each bump
  if(!NOV_UPDOWN){
    for(i=0; i<iNumBumps; i++){
      NESTED[i] = 0;
      int j = 0;
      for(;j<iNumBumps;j++){
        if(i!=j && LOC[j] >= START[i] && LOC[j] <= START[i]+WIDTH[i]){
          NESTED[i]+=1.0;
        }
      }
    }
  } else for(i=0;i<iNumBumps;i++) NESTED[i]=0.0;

  //** finish up
  for (i=0;i<lc2;i++) vector_resize(vnq[i], nqsz);
  if (k!=nqsz) { printf("updown ERRI INT ERR: %d %d\n",k,nqsz); hxf(tmp); }
  free(tmp);
  return jj[0];
}


ENDVERBATIM

:* PROCEDURE install_updown()
PROCEDURE install_updown () {
  if (UPDOWN_INSTALLED==1) {
    printf("$Id: updown.mod,v 1.16 2009/02/16 22:56:52 billl Exp $\n")
  } else {
  UPDOWN_INSTALLED=1
  VERBATIM {
  install_vector_method("updown", updown);
  }
  ENDVERBATIM
  }
}

Rowan MS,Neymotin SA (2013) Synaptic Scaling Balances Learning in a Spiking Model of Neocortex Adaptive and Natural Computing Algorithms, Tomassini M, Antonioni A, Daolio F, Buesser P, ed. pp.20

References and models cited by this paper

References and models that cite this paper

Binzegger T, Douglas RJ, Martin KA (2004) A quantitative map of the circuit of cat primary visual cortex. J Neurosci 24:8441-53 [PubMed]

Busche MA, Eichhoff G, Adelsberger H, Abramowski D, Wiederhold KH, Haass C, Staufenbiel M, Ko (2008) Clusters of hyperactive neurons near amyloid plaques in a mouse model of Alzheimer's disease. Science 321:1686-9

Carnevale NT, Hines ML (2006) The NEURON Book

Chandler B, Grossberg S (2012) Joining distributed pattern processing and homeostatic plasticity in recurrent on-center off-surround shunting networks: noise, saturation, short-term memory, synaptic scaling, and BDNF. Neural Netw 25:21-9

Dan Y, Poo MM (2004) Spike timing-dependent plasticity of neural circuits. Neuron 44:23-30 [PubMed]

Frohlich F, Bazhenov M, Sejnowski TJ (2008) Pathological effect of homeostatic synaptic scaling on network dynamics in diseases of the cortex. J Neurosci 28:1709-20 [PubMed]

Lamsa KP, Kullmann DM, Woodin MA (2010) Spike-timing dependent plasticity in inhibitory circuits. Front Synaptic Neurosci 2:8

Lefort S, Tomm C, Floyd Sarria JC, Petersen CC (2009) The excitatory neuronal network of the C2 barrel column in mouse primary somatosensory cortex. Neuron 61:301-16 [PubMed]

Lytton WW, Omurtag A, Neymotin SA, Hines ML (2008) Just in time connectivity for large spiking networks Neural Comput 20(11):2745-56 [Journal] [PubMed]

   JitCon: Just in time connectivity for large spiking networks (Lytton et al. 2008) [Model]

Lytton WW, Stewart M (2006) Rule-based firing for network simulations. Neurocomputing 69:1160-1164

McClelland JL, McNaughton BL, O'Reilly RC (1995) Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychol Rev 102:419-57 [PubMed]

Neymotin S, Kerr C, Francis J, Lytton W (2011) Training oscillatory dynamics with spike-timing-dependent plasticity in a computer model of neocortex Signal Processing in Medicine and Biology Symposium (SPMB), IEEE :1-6

Neymotin SA, Lee H, Park E, Fenton AA, Lytton WW (2011) Emergence of physiological oscillation frequencies in a computer model of neocortex. Front Comput Neurosci 5:19-75 [Journal] [PubMed]

   Emergence of physiological oscillation frequencies in neocortex simulations (Neymotin et al. 2011) [Model]

Prieto G, Parker R, Vernon_iii F (2009) A Fortran 90 library for multitaper spectrum analysis Computers Geosciences 35:1701-1710

Rowan M (2011) Effects of Compensation, Connectivity and Tau in a Computational Model of Alzheimer's Disease Proc. IJCNN, IEEE :543-550

Rowan M (2012) Information-selectivity of Beta-amyloid pathology in an associative memory model. Front Comput Neurosci 6:2-23 [PubMed]

Rutherford LC, Nelson SB, Turrigiano GG (1998) BDNF has opposite effects on the quantal amplitude of pyramidal neuron and interneuron excitatory synapses. Neuron 21:521-30 [PubMed]

Small DH (2008) Network dysfunction in Alzheimer's disease: does synaptic scaling drive disease progression? Trends Mol Med 14:103-8 [PubMed]

Trasande CA, Ramirez JM (2007) Activity deprivation leads to seizures in hippocampal slice cultures: is epilepsy the consequence of homeostatic plasticity? J Clin Neurophysiol 24:154-64 [PubMed]

Turrigiano G (2011) Too many cooks? Intrinsic and synaptic homeostatic mechanisms in cortical circuit refinement. Annu Rev Neurosci 34:89-103 [PubMed]

Turrigiano GG (2008) The self-tuning neuron: synaptic scaling of excitatory synapses. Cell 135:422-35 [PubMed]

van Rossum MC, Bi GQ, Turrigiano GG (2000) Stable Hebbian learning from spike timing-dependent plasticity. J Neurosci 20:8812-21 [PubMed]

Zhang LI, Tao HW, Holt CE, Harris WA, Poo M (1998) A critical window for cooperation and competition among developing retinotectal synapses. Nature 395:37-44 [PubMed]

Eguchi A, Neymotin SA and Stringer SM (2014) Color opponent receptive fields self-organize in a biophysical model of visual cortex via spike-timing dependent plasticity 8:16. doi: Front. Neural Circuits 8:16 [Journal]

   Simulated cortical color opponent receptive fields self-organize via STDP (Eguchi et al., 2014) [Model]

Neymotin SA, Chadderdon GL, Kerr CC, Francis JT, Lytton WW (2013) Reinforcement learning of 2-joint virtual arm reaching in a computer model of sensorimotor cortex Neural Computation 25(12):3263-93 [Journal] [PubMed]

   Sensorimotor cortex reinforcement learning of 2-joint virtual arm reaching (Neymotin et al. 2013) [Model]

Neymotin SA, McDougal RA, Sherif MA, Fall CP, Hines ML, Lytton WW (2015) Neuronal calcium wave propagation varies with changes in endoplasmic reticulum parameters: a computer model Neural Computation 27(4):898-924 [Journal] [PubMed]

   Neuronal dendrite calcium wave model (Neymotin et al, 2015) [Model]

Rowan MS, Neymotin SA, Lytton WW (2014) Electrostimulation to reduce synaptic scaling driven progression of Alzheimer's disease. Front Comput Neurosci 8:39 [Journal] [PubMed]

   Electrostimulation to reduce synaptic scaling driven progression of Alzheimers (Rowan et al. 2014) [Model]

(27 refs)