crane_slides - Population coding in somatosensory cortex...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Population coding in somatosensory cortex Rasmus S Petersen*†, Stefano Panzeri‡ and Mathew E Diamond* Presenter: Crane Huang 05/20/2010 Introduction • A fundamental challenge in neurobiology is to discover the essential differences in the neural representations of two perceptually discriminable stimuli. • Aim: To compare candidate cortical population codes by identifying features of the neural response that might underlie stimulus discrimination. Introduction • Is information encoded by a widespread network or by a restricted subset of neurons? • Are stimuli encoded by spike count or does the ms precision convey additional information? • Does each spike code independently, or do correlated spike patterns convey additional information? Outline • • • Spatial organization of neural coding The role of spike timing The role of spike correlations Spatial organization of neural coding 442 S e ns ory syst e m s F i g ur e 1 • Is the activity essential to specifying stimulus site distributed across many columns (a widespread network)? (b) (c) (d) Stimulus Barrel cortex Essential surround Redundant surround 3 1 3.6mm Spikes per trial • Is the activity within the stimulated whiskerʼs homotopic cortical column sufficient to specify the stimulus site (restricted subset of neurons)? (a) 0 S p a t i a l or g a ni s f o r s t imu l u s l o c w h i s k e rs C 1 – C s t imu l a t i o n o f w s p i k e s s imu l t a n 1 0 0 mi c r o e l e c l ay e rs o f b arr e l a t e a c h e l e c tro mi c r o e l e c t r o d e h a s b e e n av e ra in t e rp o l a t i o n b e r e sul t s illus tra t w his k e r d e f l e c t l imi t e d t o a s i n vi e w s o f w i d e s F irs t , t h e w i d e s h omo t o p i c b a r e s s e n t i a l ro l e in (c) n o t e t h a t t h is suf fi c i e nt t o o b s e rv e r mus t In c o n t r a s t , o f f e s s e n t i a l: in p a c e n tr e e l e c tro d s t imu l u s ; a l l p a c o l umn s i s r e d p a n e l ( b ). Current Opinion in Neurobiology ANN can reproduce arbitrary mappings between input and output and hence, in principle, can accurately represent the ideal decoder. In practice, due in part to the need to avoid ‘overfitting’, an ANN will perform worse than the ideal decoder by an unknown amount [8]. that activity beyond the homo informative and supports the multicolumnar coding. Howev component of the cortical repres population d’ of the entire cortic to that of only on-centre colum Spatial organization of neural coding ANN method • ANN: more accurate in decoding from ensembles of ~30 simultaneously recorded neurons than single neurons. • Seem to favor fig1(c), that the population code depends on a large part of barrel cortex. • But...might be redundant. ly, and if the SD across both stimuli is , then: cortical population response patterns. To determine mong the neural responses to different whiskers, we m d my tection measure x (Green and Swets, 1966). d quan-opulation coding in somatos ensory cortex P e t ers e n, P anz eri and D iamond 4 4 3 P d .on responses to them. For a (1) ble two events are, based ponse variable, d i s simply the absolute difference sponses to the two events divided by the average SD. en used toigare 2 stimuli presented to sensory neurons the events study how well single two different F ur e levant stimulus features (Tolhurst et al., 1983; an pendent response variable is the spike count at Essick If the mean tudying stimof sresponseinforinstimuli x and y are S values ulu lo c ation c od g us g pling of thevariable tvibrissalserepresentation withto 10 10 electrode array. A, Disposition of the w (a) (c) po tion ’. (a) In is available, it , necessary a ely, and if the SD dacross simple t ca , ne responsepulacortical heboth sstimuli is is then: Spatial organization of neural coding Population d’ d ’ m e a s r e s d is c w e en a p r o led with letters uthemriminability bettrials.aiIff the responses to" = µB – µA for rat r30 based on a photomicrograph o d riation amongbaAd(dorsal)ertospEes(ventral). B, Array position acrossof ik fired by stimuli s e on the numb ! r exposed byindependent, isthenthethe populationof the electrodes is visible (shown in black), and the rem then:craniotomy. ply 500 m length d i s A om th statisticallyne neuxro inm iys case, d’ sim , the n .he me n e co nts, of di e t (1) ! t thethed sumCofedbbytwheeesquaredpaitheu10 spike10response d s of tissue.ffrereniscPlacementadesvi ktindividual microelectrode for rat r28. A barrel map was drawn from a nit chanoff et no mal1993) andarused ion ofahetemplate. Then an electrode grid template was positioned to give th al., e t stand d as t 6): c ount. i al o bi t t was ascribedWuhet ngitvheenceoancdhitsotnmupurs bsaaplipyrooxfimhaetely that matched its principal whisker (see Materials and Methods the columnari location s co l een used lip;ikstudynose;pwellrostral ssensory neurons topauenon,,nd’ ihow lrvi, d usinglea ure of ll, lower G ssia s a sim e an seful me vibrissae. elevant stimulusnability. (b) H ow can d’ b e g enet lial.,to features (Tolhurst 1983; Essick cr 2 µA µB d diospuilmtiions? If de neurons are uncorreelraatesde, d , (2) p a th i arguments’ of mpuited fo eaand on individ al an d c o Green available, (1966), the generalization of among whiskers are conserved one responseis variable r isch neurSwetsuitly is dnecessary to t h e p o ul a t i o n simply at takesamongpthemd’acrossthe square Ifottheheresponses to ariation covariance isinto account of t trials. ro is: ships among the barrel-columns iminability at the ithm C mvariable ,(electrode). response d m m (3) en applied to sequentially recorded neurons, where covariance the response vectors across allGeisler and C is the m are mean covariance (Z ohary, 1992; electrodes, C is was available matrix ever, neocortical neurons are not, in general, denotes matrix statisariances between pairs of electrodes, and between pairs of electrodes awne and Richmond, 1993; Z ohary et this 1994; Lee of popuIn the pattern recognition literature al., measure given set of stimuli the response covariance can etimes known as the Mahalanobis distance (see Duda and dancy or synergy (Snippe, 1996; Oram et al., 1998; 99), in and covariances must2 will overestimate data. If there means which case Equation be estimated from (for restimate neurons do not respond discriminability. weakly, at which (for synergy) the true or respond very se covariance matrix can become subject to serious sampling ase this problem was avoided in large part by repeating each y times (279 –500 trials per whisker) and by restricting the ctrodes for which the neurons yielded a significant response. were 10 responding electrodes in the 100-electrode array, sensory (d) 6 (Woolsey and Van cortex All On-centre the present experiments the 40 Off-centre trodes of the array resulted in eac 3 to three electrodes; a cortical co was never excluded (Fig. 1C). In each of the three experimen 0 cortical columns 40were recorded 0 100 Post-stimulus time (ms) maps representing the spatial dis response as a f unction of stimulu the method by using rat r28 stimulated 279 times, and a PST Population d" sum of the square d single neuron d’ value s. statisticallyow ever, neuronal responthene uthely population d i s (b) H independent, s e s ar sual T 1 som wh t cx v iat e d, y s sh individual y o a ot of the sum eof athearsquaredown schemxatically response d s here . (c) Provid e d that the c ovarianc e struc ture 966): is similar for diff erent stimuli, this c an b e t ak en y into a c c ount by transforming the re spons e s to a T c oor st2 c r ls d uncodrinaattee dysdermhin wxhaimhptlheovabi)a,btheis iare (2) it , e r el . Fo e e f( s a chieve d by a 4 5 º rot ation. P anel (c) shows the i proje c tion of the re spons e distributions of (b) PetersonhRS,nDiamondaME,n 2000 the onto t e s e e w c oordin t e s, i which riminabilityndatdual d’s ith bresponses variable (electrode). i ivi the c an e c ombine d a in the u or lat e d c as e . (d) W e recorded een appliednctoresequentially applie d the neurons, where p pulat n d me thod hematis e d in 1992; t covariance owasioavailablesc(Z ohary,parts (b) Geisler and and (c) to at a simult an usly c ord fr ever, neocorticaldneuronseoaredreinot,e dinom general, statis1 0 0 micro ele c trod e s implant e n rat Gawne and sRichmond,rt1993;contohary fet al., 1994; Lee omatos ensory c o ex. The Z ribution o n-c tre (homo ic) a off-c entre given set oofenstimulitopthendresponse covariance can non-homoto ic) neurons t stimulus it e ndancy or (synergyp(Snippe,o 1996; sOram et al., 1998; d is c t i s sh 999), in whichriimsinoaniornEquation roduwillwioverestimate (for case om [6•o].wn. Re p 2 c e d th p erm s i f Spike timing in population coding • • Spike count or spike timing? The precise firing of spikes (within inter-spike interval) is to reflect random processes, or is an additional, informative dimension of the response? 444 Spike timing in population coding S e ns ory syst e m s F i g ur e 3 1 01000000 1 01000000 0 00000000 1 00100000 2 01010000 0 00000000 1 00010000 0 Stimulus trial 01000000 1 0 Spike timing 01000010 00000000 40 Post-stimulus time (ms) • • R o l e o f s p i k e t i m i n g i n c o d i n g s t i mu l u s s i t e . ( a ) S p i k e s f i r e d b y a n e ur o n in b arr e l c olumn D 2 in r e s p ons e t o 1 0 d e fl e c t ions o f w his k e r D 2 . N e ur o n a l r e s p o n s e i s q u a n t i f i e d t o a l l o w s p i k e c o u n t a n d s p i k e t i m i n g c o d e s t o b e c omp ar e d (s e e t e xt f or d e t a ils). (b) Mutu a l inf orma t ion (b) 0.2 Mutual information (bits) Spike count 2 (a) Spike timing 0.1 0 Spike count 10 20 30 Post-stimulus time (ms) 40 c o nv e y e d b y b o t h c o d e s a b o u t s t i mu l u s s i t e w a s e s t i m a t e d u s i n g t h e S E M I m e tho d [ 1 4 ••] a n d is p lo t t e d a s a c umul a t iv e fun c t ion o f p o s ts t imulus t im e , av e ra g e d ov e r 1 0 6 sin g l e unit s in b arr e l c olumn D 2 . R e pro d u c e d w ith p e rmission from [ 1 5 ••]. B ars d e no t e s t a n d ard e rrors. ANN method: stimulus discriminability improves as bin size decreased from 40 ms to 6 ms---->spike timing may play an important role. Information theory: SEMI estimation confirm ANN method.or redundant spikes in an ensemble to exert synergistic (Figure 2d). Given that most neurons would be expected to fire at most a single spike within 12–16 ms of stimulus onset, the d’ time course suggests that these single, shortlatency spikes are an important component of the neural code. This was directly tested using the SEMI method, by estimating the degree of discriminability based on the whole spike train compared to that of the first spike (time) by itself [14••,15••]. 90% of the information transmitted by the whole spike train (both for single cells and cell pairs) could be accounted for by just the timing of the first post-stimulus spike in each cell. Moreover, although later effects [12,17–19]. A simpler type of cortical population code is one in which spikes encode stimuli independently of one another. The key question, for our purposes, is whether correlations provide the ideal decoder with additional information beyond that already available in the individual spikes. Synergistic coding of stimulus location has been tested using ANN and SEMI methods. A particularly useful feature of the SEMI is that it quantifies the contribution of Correlated spike patterns in population coding Po p ul ati on c o ding in s o m ato s e ns ory c ort ex P e t e rs e n, P a nz e ri a n d D i amo n d R o l e o f c orr e l a t e d s p i k e p a t t e rns in s t imulus l o c a t i o n c o d i n g . ( a ) A c r o s s - c e l l s p i k e p a t t e rn m e a n s t h a t t h e p r o b a b i l i t y o f u n i t 1 f iri n g a s p i k e a t t im e t 1 a n d u n i t 2 f iri n g a s p i k e a t t im e t 2 d i f f e rs fro m t h e l e v e l e x p e c t e d w e r e t h e s p i k e s t o o c c ur in d e p e n d e n t ly. A n illus tra t iv e j o i n t P S T H f o r t w o n e ur o n s (l e f t ) i s c o m p a r e d t o t h e j o in t P S T H p r e d i c t e d from in d e p e n d e n t f irin g (ri g h t): t h e hi g h e r v a lu e s a l o n g t h e d i a g o n a l i n t h e l e f t p l o t r e v e a l s y n c hr o n i s e d c r o s s - c e l l p a t t e rn s . ( b ) A w i t h i n- c e l l s p i k e p a t t e rn m e a n s t h a t t h e p r o b a b i l i t y o f t h e u n i t f iri n g a s p i k e b o t h a t t im e t 1 a n d a t t im e t 2 d i f f e rs fro m t h e l e v e l e x p e c t e d w e r e t h e s p i k e s t o o c c ur in d e p e n d e n t ly. C om p aris o n o f t h e j o in t P S T H a n d i t s p r e d i c t or s h o w s t h a t t h e c e l l h a s s p i k e p a t t e rn s c h a r a c t e ri s t i c o f r e f r a c t o ri n e s s . ( c ) U s i n g t h e S E M I m e t h o d , t h e c o n t ri b u t i o n s o f i n d e p e n d e n t s p i k e s , c r o s s - c e l l s p i k e p a t t e rn s a n d w i t h i n- c e l l s p i k e p a t t e rn s w e r e e s t i m a t e d s e p a r a t e l y f o r 5 2 p a ir s o f c e l l s r e c o r d e d s imu l t a n e o u s ly f r o m b arr e l c o lumn D 2 [ 1 5 ••]. Th e c o n tri b u t i o n o f i n d iv i d u a l s p i k e s t o t h e t o t a l mu t u a l inf orma t i o n is c om p ar e d t o t h a t o f b o t h ty p e s o f s p i k e p a t t e rn c o ns i d e r e d t o g e t h e r. ( d ) C o n t ri b u t i o n s o f c r o s s - c e l l a n d w i t h i n- c e l l p a t t e rns ar e c om p ar e d d ir e c t ly. B ars d e n o t e s t a n d ard e rrors . Joint PSTH from true responses Joint PSTH from independent responses Cross-cell ime it Un t im t-st s o 1: p Unit 2: post-s tim time (b) Within-cell m -sti ost P e tim Post-st im time Question: whether correlation provide additional information beyond (c) (d) 0.2 0.4 that already available in theTotal individual spikes. 0.3 0.2 0.1 Individual spikes Spike patterns Information (bits) • (a) Information (bits) F i g ur e 4 445 0.1 Within-cell patterns 0.0 -0.1 Cross-cell patterns m e -sti tim t s im Po t-st s Correlated spike patterns in population coding ime ost-stim time (d) (d)0.2 0.2 (c) 0.4 0.4 0.3 0.3 0.2 0.2 0.1 0.1 0 0 Total Total Individual Individual spikes spikes Spike patterns Spike patterns 10 1020 2030 30 40 40 Post-stimulus (ms) Post-stimulus timetime (ms) Information (bits) Post-st P im t Information (bits) Po Information (bits) (c) Within-cell e ti m Information (bits) e o u s ly f r o m of i b ution of lp e s b o t h ty p e s -e r.e ll c owet hin- c e ll ti rs d e n o t e 0.1 0.1 Within-cell Within-cell patterns patterns 0.0 0.0 Cross-cell Cross-cell patterns patterns -0.1 -0.1 20 30 30 40 40 1010 20 Post-stimulus time (ms) Post-stimulus time (ms) Current Opinion in Neurobiology Current Opinion in Neurobiology that patterned activity distributed across the cortex might not that after trial-shuffling, ANN that patterned activity distributed across the cortex might not fter trial-shuffling, ANN be an essential element of the code, supporting the model hanged or slightly improved, be an essential element1d. What is the role of spike timing? of the code, supporting the model ed or slightly improved, illustrated in Figure stration [15••] that cross-cell illustrated in Figure 1d. What is the role of spike timing? on [15••] that cross-cell Precise spike timing (to at least 5 ms) conveys considerable Precise spike timing (to at least 5 ms) conveys considerable information beyond that available only in the spike count (Figure beyond that available post-stimulus spike is paro f p o pulati o n co ding information3b). The time of the first only in the spike count Conclusion Principles of population coding • • • Spatial organization of the neural code: Figure 1d Role of spike timing: Figure 3b Role of spike correlations: Figure 4 c,d ...
View Full Document

This note was uploaded on 09/15/2011 for the course COGS 1 taught by Professor Lewis during the Spring '08 term at UCSD.

Ask a homework question - tutors are online