ty 8 7 Ww. % is " .) . . tearning Objectives ) to be able to: Hebb’s theory for the . Describe the neural network PPFOACh, out ena Of leaming, weaknesses, lining its strengths and 4 Explain what is meant by neural plasticity n and sketch it i 5. Discuss the benefits of combining behaviour: male bas. neural approaches to learning, 2! Computer simulations and SEE rai Cp Hevement of evolution — anguably, the mast important achieve ain produce an accurate model Othe gross phy: chapters 5 and 18), it must also ‘code the regularities nas the Proxim ‘here inn, atures of the external world (gee that make thé world predictabl NeTesponses are no longer appropriate, However, learning and evolution should viewed on a single continu er ove! Sy “Carning is often, placed on the ‘nurture’ side of the a scent and is *uenly contrasted with biology (genetics and physiology). This dic eee le sii is not helpful and is, in certain respects, misleading as to es E eon ‘va jprilled nature and nurture. In some areas of psychology there has always 4 tink between leatni " Sihnooee chapter 1). This chapter highlights the ways in : RRGAGTEISGCENSHENCISES. damental pa ‘or example, the visual system » fc oo, visual “ator sision, bu the development chacisinnaceac aNTE oli 3). ‘The 5 eval then involves activity in ; lag ww neura ‘esses instanti Pa hi presents what is known about ho neural proc ntiate lang this chapter presents wha " " a cesses: neural plasticity. Learning and Genes ral c timated that there aré1}000 million neu ue a a of sand. But here ae only a small number of genes tus * and accept the upper estimate of 50,000 — even if there were hundred Of thou genes, the argument that follows is still valid). Even if every one Paes 506 were involved in wiring the brain (which is far from being true), then each gene ai This Outcon, is simply inconceivable. Thus, for hard-wiring, there seems to be a gene solution to this problem determine the general wiring of the systerm ~ this is a fi seenvand maturation. A second solution has ing us, genes rOduceasspectrum i, spabilitydearning) an nate potential set up during development In relating biological processes to learning shortage. 'S appropriate to ask to what exter ints on (a)jthetsteeturevand(b) the context relation to the structure Oflearningpitseemsethatlearni (Pumberofconstraints, For example, s {0 without any consequence (ie., they are not predictive of a salient event) are mot ficult to condition than novel stimuli (this ig © phenomenon know lia is particular constraint on learnin, coluvoe ary terms ‘ersane mie enon TeoUReeO py aa a fore teimks sense not to pay at In relation to the content! ‘aman olomnmcmne saa en. example, it is easier to conditi rather than oth d as Isto pictures of snakes and spiders than 010 dangerous stimuli, Suelras electric Sockets (this phenomenon is known - paredness), More speculativeh im “rs THay"ha ing St (e.g, jal dan cerns uli: inve! Qin NB (eg, ie * Sconcemngrcheatingrand-deception, a av} 1 eae 8» Opposite-sex pary - ese possible ® ty innek nd, see Pinker, 1997, 2002! : is clearly ber ASK YOURSELF urther ex: 7 — mee ‘mples of the biological constraints ‘ In evolutionary terms are "8 See Alcock, 2991) Hect; " 001). "Th there any disadvantages tq __D@SiS of suc learning?SSE etapa _ years, there has been great interest in ufifyingipeyeHOISgISINGPeSMpUCT en ning vgthmeurophysiological"approach ‘sort eI s, Ig keeping with the 0 perwee? the conceptual nervous system and the central nervous system (Hebb, 1955: 7 em , ter cove gn), this chat a ing learn, & re 1S acne * (b) artificialmeuralmetworkseand(c) ni “0 if : pavioural Approaches to Learning se! ony of psychology ha deed, whole areas of psychology have been ‘o this topic, and claims have been made that the study of behaviour and learn- te only legitimate approach to psychologyssradical'behaviourismibeing the most is declaration of this position (Watson, 1914; Skinner, 1938). nib om ofthis section is (0 summarize the major behavioural approaches to learning gro describe their principles and laws. These principles/laws are important in sub- an ent chapters when experimental findings relating to specific behavioural /learning = gms (¢&» latent inhibition in relation to schizophrenia) are discussed. pefinition of learning aruration ora transient state of the organism). learningrisinferred from measures. of performanceny usually in some for} iqur. However, learning is not always ‘ignifestiwBEH@Wiour. For example, consider the phenomeno arming. In this ype of learning, \s. For example, ifa rat exposed romauieiiaaiadiilelioatibior® iiiecaiaiaaten es ho any evidence, of. Jearningeuntilsit:issmotivated (e.g., by food); then its swimming spee ismuch faster than aonaive'rat’sypproviding clear evidenc reconuitioning is another example of a form of i see below). Other forms of lear learning that not nece: hat reflect nt, bu . the definition of learni i a) char “Bhand.(b) to the process of recall hy Wes of learnin, ig The hi the OP OF the psychology of learning is characterized by distinctions between a 35 conte umber Of processes (e.g. cla instrumental learning). A major debate on whet Sey diferent Teaming-processe : aCa teliierh ts Strength of response Figure 7.1. Habituation curve. Habituati stimulus with repeated and non-reinforc curve differences of experimental participants. can be shown from behay nctions In the study of these forms of | learni tion that i ip study of three simple forms of learning: : a ~ the shape of actual curves depends on the nature of the stimulus and indivig 2 1 L Trials on is observed 2 the weakening of ed trials. The figure shows an ideatseg PO Jual ¢ brain; or iariti ig. This is an im ural experiments that diffe eg portant issue. because learningesystemssintherbrain (iat, ) a7 ing, behavioural research has adopted the asim, much as a comply - This assumption has motivates (€.g., habituation); (b), a }» Pavlovian conditioning); and (c (©, instrumental larg ~ 1 Habituation Tl res xample, if an 80 dB(A), psychology experiment and some in the brain) were measured, then _Sentation of the-tone; but, after a veri , 50-ms tone were because it does mt me ics (ie., reinforcement). 7 in Presented to a human voluntect n a ne Tesponse (¢.g,, skin conductance or evoked pot be obseriedon fi'F* . eco \ fev. more presentations; this Nidan ear. Under these circumstances, the brain has ait msequence), and therefore needs no further setion (figure 7.1) In fact, it is not necessary to form of learning. We are habituatis were not the case, every sound of a car passing, every cl would elicit a large reaction, We jean in themselves, so we do not continu Ing sia 80 to a psychology laboratory to observe ths” ais to stimuli on a moment-by-moment E ont ambient env" the rr sing of the door, every ringing of signif, n rapidly that such sce. The world you just imagined m, fom 9Y NOt be to, f chapt 0 Far fr (aPC ae this POM, COnsider the behan” OM the world exper ist ro wo to t dey ipsondard conditioning terms, this stimulus is an un response, P 1¢ habitu- : 5 vans falls under other headings in nse ity and important qualifier r has acquired biological significance though conditioning), tf Another important point is that, serve the stimulus is.aversive (Le, norious, such ax anelectric shock), the habituate; indee e abituation and sensi from ‘loves: animals'to human Beings. As ures have been extens- ie studied to reveal th uli to e (i.e., acting attention) Habituation, has often Case, ae ychological » although it is basic, r, "S(@g., expectations), For example, 2 hfe iamiadaunaettaeiy tes/hours), theAN} (eee Associative learning dan enormous research literature and has 5 ning processes. In this section, Pay an ng) is reviewed. Vian | " yerate' Associative learning has gen derlying lear number of theories of the un! ditioning (also known as classical cond ‘Associative learning consists in che observationitnatsPewO'SiMUlTaREp ea Fer erhape the fundamental = processor Carmen ome theo X it underlies all forms eign, jve learnin anc ing the regularities of the work: tl ser (Shanks, gs form of learning can also be applied ings &y, ciative component Pavlovian conditioning ociative learning was initiated by the (1849-1936). He chanced upon the phenomen, components, by accident, through his wer 7 ie The rigorous experimental physiologist i associative learning, and many of its c r which he won fhe! Nobel Prize'in904 (Fey, m7) physiolo; Pavlovian conditioning consists ofthe following basic propertigSDOR Stag at (as indeed do most of us; this is ; us, UCS) ~ these are ‘unconditio is required, Pavlov was foreunate to have been studying digestion because of thee sé What he observed was as follows: sometimes his dy ce po! Figure 7.2. Ivan Pavlov with one of hi Library) of his experimental dogs. (Photo © Novosti / Scien(0 the sand it might be imagined something ae This Pavlov Was intrigued 8 nuisang u cor ;ste™ igued by this stra strange reg € for Pavlov’s sty y ation aed ns. What Pavlov obs, ea Spent the rest of his a poses tte 1 UCS) BOATS same ico this ee of Pallings ofthe CS-Uesir ° sim SSS eon ‘stenmedtherconitional simuas i r te er 0 Bo svn eae it ; : , the meat). (‘Conditioned’ Was incorrecth k; we ) Pavlov’s move from the sul themed oem ). A real fF ae y 1979). A real sense of the scientif é = ic flavour of this work can be found in jrwas originally thought that what happened during the sing W282 form of s , i vy sly ig the course of Pevlovian con- uc aier conditioning th i cy oq proved difficult to sustain. Response utipldation thesUCR. However, this the i nclaealapaialpiiiancsiine Saree, at . ). Today, the plemented with an internal process: $-0-R. The ‘(represents central states of the brain/mind. ler us consider an example to illustrate the problem with the response substitution theory. If we wereexposed to vanjunpleasant|UCS\(eigipelectric shock), 6tFUCRWouldy _Rpgipiupainenharenioctie, if we conditioned this UCSshoekitor=a\CS (eg, light), what should we expect the li is presented? Well, before i) wou ‘heconditioning phase, the to- be-conditioned CS (i an sametesponse, which we but, after Condit would 3 crucial question iis the CR the same asthe UCR? The des answer Ry These passive avoidance listen and learn how to avoid the aversive UCS ‘oct signalled by the CS (figure 7.3). hough p dimentar avlov's learning procedure seems rudimentary, 88 can hardly he eee Asan n example of this a during the 1960-1980 its importance in psycho- ee When y and challenging the 8 beh the century, learning to the pre- "ties wy But from the 1980s er importance. "Sime downgraded in relevance and imp chey form bag ME, associat “C argument for neural network models of | behaviour and cognition. In addition,210" Foundations us | ucR bb” UCS ____— (behavioural excitement) | (a) Before conditioning (shock) Mild orienting ont and and habituation - ig (b) During conditioning cs ' ucs ‘ (shook) (light) oS 4 (c) After conditioning on CS ____-> (behavioural suppression/ : os immobility) Figure 7.3 Example of associative learning, (2) Before conditioning the unconditioneg .é., an unconditioned re, stimulus (UCS; electric shock) elicits e NES respon UCR); the to-be-conditioned stimulus (to-be-CR; light) elicits"only"a'mildvorienting response. (b) During conditioning, presentatl a yw, after training, when the does : this is an examy many of the best lines of evidence for the importance of cognitive factors and centr states of emotion (e.g, fear and anxiety) come from learning paradigms inspired by ie work of Pavlov. Much of neuroscience research uses basic Pavlovian procedures — ifPala were to visit a neuroscience laboratory today he would feel very much at ease with te type of behavioural experiments found there. Sensory preconditioning In this procedure, tworstimuili (e.g, light and a 1. These sti need not haye,any-biological-significance, jonse. The bast observation is that if tworneutral:stimulisaresrepeatedly:presentedstogether then becomejassociated (i.e., linked). How is it known that they become associated? Let assume that the lightiand 'ton€ Were paired :in:thes'preconditioning™phase. Now a high is ina Po igpanditionippuasadase. It is established that this lh! | . Now what happens sented (remember this has ot beet conditioned to the UCS)? It woul Js the power 10 ect CR, even though it h: 7 It is easy to see how sensory preconditionii es For example, we may have had a led o as being bitten). The sen:OLN te LG (0) Palred prosontation(bolor conditioning) Ught + Ton No significant : responso (b) During conditioning F cs} ues (ight) (shook) (o) After conditioning = cr igh) evra suppression (ligt immobility) @ CruclltammsptetSte ong ext oR cs i > (behavioural suppression’ s (tone) immobity) ~ 14 fxample of sensory preconditioning (a) A li ) During a conditioning ae , which (c) after a sufficient number of eeorng wi let (CR; in this instance Sonora suppresionimmobilty. But now (d) when the’ tones presented - which fmm been associated with tl ic = it elicits the CR, re asocated. gatilization of our fear. What sensory preconditioning eonfirmsis'tharassoGatie learn eith ed vat the presence of re temporal cient for learning to occur (figure 7.4). Yo understand the applications of Pavlovian conditioning, especially to psy- Copethological conditions, such as fear and anxiety, itis necessary to know some more the action and exinion oF CRS, i “Bion of conditioned response ay .. he above discussion, it was rather loosely stated that Pavlovian conditioning occurs ‘Secs ieprexentea aranureamRIEAS'THE'UCS. In fc, the ordering ofthe ere are three types of CS/UCS temporal ordering, (a) Font. tails the C$ di t effective: thisientailsyocs (tone) (shook) ucs (a) cs (tone) ucs (shoo#) (o) CS (tone) cs (shock) (©) Figure 7.5. Types of CS-UCS pairing (ucs) and conditioned stimul (b) backward conditioning; an ‘Two factors are im| presented without lus (CS) takes oF .d (c) simultaneous con’ portant in CS/UCS pairing: (a) 10 20 rime (seconds) Simultaneous pairing L—_——_ —— iring of the unconditioned stimulus +. The temporal pai forms: (a) forward conditioning; ne of three ditioning. s important about ¢ known as non=veinporeel! (th i, When a C58 is seen dung cSH), .. g, where enetyes.f patel foot «iies6ther Unpalfed"(lenoredsCS-) (an example of differential fear condition’ 7 is given in chapter 15). The CR may be measured in a number of ways: ; ys ome re) 7.6). Generaltafion. This proc Generaltafion ess _ outst : associative learning. For example, if a or cy, fe - however, a CR (thou; - also be observed. The ugh weaker) to tones either side of this frequency woul ‘aneriees os a process can be extended sui | anid nor juserbe p="Learning and:Neural Plasticity 213 S s Bo a & Reinforced trials Non-reinforced trials Aquisition Extinction Figure 7.6 Idealized acquisition and extinction curve: iri ‘S) wath af Unonditioned stimulus (UCS) reine \d the strengtl itic (CR) increases durins len the CS is prest orced), then a = 2 = & —— eee 800 1000 1200 Tone (Hz) 7 Generalization curve. Generalizationrefers'to"a conditioned respor Figure ‘yl oaiaiaaeig aeiansaoadimansiangstatnestinior example, if the CS is a 1,000 Hz tone, then a weaker CR will be observed to tones (e.g., 800 and 1,200 H,) falling either side of this value. hus, it is not always clear what ‘was the original UCS that set intrain the chai ions in beings, a good example of this p i ee cannot be directlyyconsumed;andsitsis:only of value as long as it _ acquisition i.e., UCS) ~ in times of monetary collapse, it s. Yet, in our daily life, it is o! because its yaluestomusmresidesmnovonlysin its h hi ower tO i 5 i themsel This ‘Sherorder matrix of C and Pesonal ps ch 'sdevalued’s dogs, 3) The theoretical basis of: Latent inhibition i £ latent inhibition: the inhibiti seen in the phenomenon ol inhibition is aes le of nother examP arently random n' ation, but to an a= ubow, 1989: = CS/U! process chapter 8). Biological preparedness It is the case that come simulans or EHP CeRITOREN TSH ONS e.g. spies snakes vs. flowers/mushrooms, see chapter 15). For example, laboratory-reared th monkeys seeing a snake for the first time do not show fear, but their wil Show great fear. What accounts for this difference’ diti n they see a conspe’ ~ this fear TT minors sk if the monkey se face or 3 fees c and, oh trick with mirrors, is led to believe that the conspe' terrified by seeing flowers, then does this learning to x react to flowers? The answer is that th ei fend toa permanent comes nab and criss racial boo me terrified look on a conspecific’s face ~ whicl monkey); other stimuli (¢.g., flowers and mushrooms) simply do no Fo. today you were ill, you wi r example, if you tried a jon. There® induces a fear reactio? ae clue t, It is thus comrab eatning and. Neural Plast ould shy away from this food in future (perhaps for ever). What is important about spe seaf00\ the fact that youshadkalsoeatem*bread, poratoss, ire the day before does enter your mind as dane fodtls have med fttion amin blocking we ‘the phenomenon of blocki s i the temporalyco-occlanrencesof-stimuli (i.e., continui blocking two groups are required. (a) The ¢ more to associative learning than 1968). To Kami on ditioning WithyaycompoundlCs (e.g., light and tone), and after training either s a alone isisufficient to @licit the @R9(b) The experimental group is first conditioned to one of these stimuli (e.g., light), and only then go dure as the control ht + tone) stimulus. Blocking observed in the experimental group: onc to the ¢, the lichsendisieniogdalgc's the . This effect has attracted considerable theoretical attention because it seems to say something important about the process underlying associative conditioning. Specifically, it disconfirms the continuit oral association is not si and instead suggests that the important his blocking effect is interesting because it shows the limitation: . pethaps pointing,to biologicaliconstraints. In Nature, being able vo presi easels ieelyeregethesIMEAaRATaBCge; but blocking a mn "ASK YOURSELF when stimulus This effect may suggest why | Why is classical condition- it is difficult to change e acquired acer- | ing considered to be such tain beli en wh [a fundamental learning contradictory evidence. process? Instrumental learning franca ES TE ec ERNE aoNowretm by ‘rei is approach to learning was initiated’ by "Thorndike (1898, 1911) and Pioneer oT 35, 1953, 1957, 1966, 1971, 1984). Skinner attempted to develop a technology our as Wellas:provinga means to change behaviour in society (c.g., token economies used in prisons: prisoners’ behaviour shaped by linking responses to specific consequences). Skinner's idea of a utopian society based upon instrumental"learning=principlessis described in his 1948 Skinner's psychology was ambitious and wide anging, purporting to account for individual and collective behaviour, and providing a ‘echnology to ‘socially engineer’ society. . Now, unlike Pavlovian learning, in which th nt trumental learning the pres« cc “Pendent) upom-responisés=forexample, a food reward being dependent on a particular tiReward system Positive/ || Effect on itive | negative environment ‘ersiem I“ reinforcement Mere a a Prt PH fut behay® Motor |__, ' ‘Stimulus Cognitive |»! — | Behaviour inthe |-— system ; environment ec —— Praag y ett! Extinction’ | Effect on hein em I<] punishment environment Punishment System Figure 7.8 The instrumental learning process. Appetitiv non-punishment), mediated |, increase t contingent upon the muli; and aversive stimul and frustrative nen'reward), mediated by the behaviour « response. When the animal is allowed freely to re an operant. It is emitted freely by the animal onment: the e1 fee i S (€g., reward and punishment). Like Pavlovia response isistrengthened»with.increasing m not accompanied by reinforcement _ it then, This feedback is not aViOUrS ma $10 a desired response are reinf d, a pr ness ofthis behavioural technoag eos AM cual iw <0 Plotsspaceeralt, and is seen in the Tang oem nha. a rely q -€., punishing at \. decre: mse I and it ‘operates’ on the envi F salient in conditioning, the instrument and if behaviow Undergoes a process of extinction Is ‘instrumental’ in the sense that behavi Prec iencbavST TTS MONT core ems poe oot nat teinforcemeatiprovides a form.of feedback fromthe enitont#= mott en that ples eel because it has ‘herpowwersrainftuence, subse ' ® imals e hat lead ani » and eli unfavour Timental contingen, h cies uct e— —————ee forcement eit! gnforvunately, ‘reinfon roaches to learning. These differences need to be clarified in order to avoid various PP! ‘onfusion. In the strict form of Skinnerian behaviourism, the reinforcement is e . Served for envit es, cont fease ur, The process of ‘reinforcement’ the probabilit i be defined in terms of the ‘ and. “of the ex} The can : ‘operation’ refers to the ee ad the behavior ¢ application/removal of a stimulus as i removal increases the probability-and/or strength of 7 u ng stimuli’ have different meanings In the ion or behaviour. There are two forms of reinforcement. ze 1 Positive winforcehient = application ofiaistimulus'thatleads to»inereased:probability or strength, of behaviour (e.g., food delivered following-a"particular'fesponse); ot ased probability 2. Negative reinforement = removal of a stimulus that leads to incre or Strength of behaviour (e.g., removal of shock followingyperticulan-response) In addition, in Skinnerian terminology, there are two operations that have behavioural consequences: 3. Extinction = stil of n-delivery of food following a response w ously was. e); ; 4. Punishinene = 27 icine |: Camelia strength of behaviour (eg. delivery of a shock following a particular response). In Skinnerian psychology, there is also a contingency known, as." i is! 7 which is the non-deliverysofa'punishitient conti ¢ (below, this type of response is called ‘relieving non-punishment’). More generally, following Pavlov's definition, ‘reinforcement’ is used to refer to.any i i een stimulior between — thing Jiaeisaiasmmteoremresienecaparanannicibeniay” stimulus and response). In this more general sense, a reinforcer can lead to conditioned inlybitions reflecting a strengthening i or rength of behaviour — thou, str gh these terms play no part in the Skinnerian conception of reinforcement and behaviour, where such terms as ‘inhibition’ are eschewed in favour of stimuli and responses that can be directly observed and manipulated (see above). Likewise, thi i r neal tinction of Whatever behaviour or inhibition of behaviour, was current. This non- Skinnerian, broader definition of extinction refer: feany form of a fang r nishment’ refers to the application of anes ation 5 ates ok, and consistent with a wider behavioural and Reurops al-state terms are ¢™ Joyed, often liberally and jn cho, ath Throughout this bo’ literature, such intern: efficacious way. “R' info yd jed to account for types of reinforcement that are n This definition is ne accounted for in Skinnerian terms. Fo antenna mene (‘rewardi 3) reinforcingypropertiesi® significant psychological event. In. a Poi, ing behaviour, in many rerrects, eesembles-punished-behaviour (it does noe, "Si, ion); more precisely, omission feat Is faced with extinction will try to escape th, © sin the behaviour seen under extinctit 1 (e.g., anima ck other animals). Importantly, the central state that motivates this punish, stark connat® ty tion or atta‘ the causa like response may be assu Skinnerian reinforcement and behaviour conceptions, tains an intervening variable: r ot [leads to] punishment-like behaviour: 1 cen expectedly Ina shion, if an animal is trained to expect punishment, but thertn punishmennisnotpresented, it behaves-asifithad been presented with a positive rey, gail th remo of expected rer sn to} central state of {relieving non-punishment’ [leads co] rewarLl Moving away from behavioural data to examining the effect of drugs on behavior (a) drugs that impaie responses to punishing in drugs that impair responss further supports these associations: to rewarding stimulivalso tend toimpair*responses:to-relieving mon-punishing stim (see Gray, 1975, 1987). Thus, thedelivery Of punishment, as well as the termination omission of expected reward, may be thous] » things that the animal would work to decrease); ai yp mination or om) (i.e., things that the animal would work to increase). vioural effects wit It is difficult, but not impossible, to account for such real beha’ ous tear expectation” Of pleasant and unpléacafit BUECORTIES onthe animals ht (even the humble laboratory rat) ~ such expectation is also observed in purely assoc sized throug tion effects, as seen in sensory preconditioning (see above). As empha: out this book, central states are important (e.g., emotion), and these a jour activate s ‘explanatory fictions’: they are seen a: : of sue a! uuinenmensalstmnuliays can be sh 2 ; UV Reinforcement-schedules t (usually the withholding of rewar"400- 300 Performance level 0 1 Time (minutes) Figure 7.9 Examples of fixed interval (F!) behaviour. The scheduling of reinforcement has a major influence on the rate and pattern of behaviour and learning. On a fixed interval schedule (e.g., FI5 is reinforcement every 5 minutes), there is a characteristic scalloped shape, with performance greatest just before the expected delivery of reinforcement and least just after reinforcement (adapted from Ferster & Skinner, 1957). ale ma déSivable"behaviours. A process known as successive approximation is used to reinforce behaviours that increasingly approximate the final desired behaviour. Through a process of selective shapi Is (and us) can be trained eha- yiours ~ i an beings, this ed in different types of training programmes. The sequencing, has important e1 & Skinner, 1957). Reinforce _ interval schedule). In addition to ratio and interval schedules, reinforcement ivered afté# each response)or intermittent/partial Goodale inforcement can also be fixed (ie), predictable) ormvariable reinforcement after a set nlimber of responses (e.g,, FR20, after every i a variable ratio schedule of ggeenePeremenelasniy (eg. VR20 has a mean of 20, but on each trial the precise number of responses varies). A‘fixed ine ppahnlanilaceciniamamneianonliid response, is made after the passage of a fixed period of time (F120, every 20 seconds); a variable nterval after a period of time which <= uapealaiaraiahieldir i Different schedules of reinforcement produce characteris ing acquisition. For example, on a fixed-interval schedule, immediately following a reward, response tate declines and then starts to increase as the (figure 7.9). Acombination of reinforcement schedules can be used to maximizs rk rate in a given Position, Skinner is responsible for developing a m ae 'nipulating reinforcement schedules; an = a to experimental pigeons and rats, most "einforcement schedule or another (piecework, performance-related pay, everjeSome™ fo ™ of partial schedule.(on..which re inforcement is made every nth response), which 4.€., schedules of reinforcement) mental. respgnses (Ferster ate (Le., ratio schedule) or time whole technology of changing behaviour by d, in case it is thought that such changes are human beings are motivated by one form220. Foundations Rate of responding Partial schedule Continuous schedujg poppy | piyit Extinction trials has a random (unpredictable) element, usualy produces the highest work rate this fg is capitalized on by gami 1 likely to id whe Feeived toby some element of control (however illusory in practice) ~ despite the identical oucom simply receiving a small reward every nth response, irrespective of choice made, isdexs tvating to gamblers, who often show superstitious behaviour (e.g, they think they bm worked out how to beat the machine), which is itself reinforced by intermittent schedes of reinforcement. (If you think that the principles of behavioutism are not relevant human cognition, then consider the prevalence of gambling in socie:y — and then tit about the similarities of, ‘gambling principles, based on reinforcement schedules, and thos operating in everyday life: e.g, occupational behaviour:) {In terms of extinction, the major findings are that Partial schedules lead to slower et tion: this is known as an tion effect (PREE; figure 7.10.4 similar effect is four ather thanvfixedischedules, an effect tt may be “neha The theoretical basis of te effects is still debated, It may be TRI the subject it u les _ pote sersive OnUMTEBHOEMENCingeinee ste na regular fixe honkoccurrenceis salient Another interpretation to .€, Pavlovian conditi during Partial schedule it finds fest as =" i (Cs) 7 ASK YOURSELF. a te, W hee ‘What are the advantages Teste ie ta of adopting a ‘schedules of aay ‘ a cOUtnedfrnstration aberenen bewvinng (sROndingandsetdeGtion (could do tis bY Sd ‘ping bebe # Rn ones he gro ~—_s0ing:behaviotiytmmme-Learning’and Neural Plasticity” 221 imilarity of Pavlovian and instrumental conditioning Si what is learned in Pavlovian and instrumental paradigms is.the.associaviony between case of Pavlovian conditioning; instrumentalsresponsevand» a i the case contingency ee formed. In both cases, ICS; the response in our everyday lives, we know, with a certain probability, that ifwe behave in way x then consequence y follows). In other words, events must go together (contiguity) and this relationship must have some predictive value (contingency) Whether these two approaches to learning fepresent different behavioural expressions of the same underlying learning process is still debated. However, there exists behavi- oural evidence that Pavlovian and instrumental learning rely upon different learning processes (Gray, 1975); and, discussed in a later section, there is evidence from neural plasticity research to support this ‘two-process model of learning’ (Mowrer, 1960), . However, in both forms of learnin; 4 of tected BRET CMEGECSTUCS or reapaaeppaREanL oa _ comé)(reinforcement)'can be clearly seen in the case of vicarious-or.imitationslearning Much of rs. In this type of learning, responsesyareqnowjsimplysemitted: we watch others»behavingyandebeingereinforced andjlearn from thisvexperience. This form of learning may be understood in terms of the a . This cognitive perspective contrasts with the radicalibehaviowristyperspective in that the | ASK YOURSELF latter insists that the key to léarnifig"is'the association of | Do dassical conditioning s if there is no obvious response, | and instrumental condition ing differ only in terms of then what is there to learn? It i t response itniotneeded for earning to occur (@gassensonyer | 20% Simul and responses arranged? Bee Other forms of learning ~~ Pavlovian and instrumental processes have dominated the psychology of learning However, other forms of learning exist. For example, ‘ng the relationshij (on. Look at the ‘raving in figure 7.11. What do you see? If you have seen this image before, then your “ssver comes easily; if not, then it takes a while for your brain to work out the object ithe image. You should see a Dalmatian dog in the snow (head in centre of image, 2Xdy to the right). Once you h: is it PS too has your brain). Thus, we eral learning’): it is not simply waiting there to be seen. + uleunanigeiavOlves, the acquisitionof skilled: behavioursiand-habits.In the last 20 *: 4 major distinction has arisen bi de “ning and memory, Dede fespliesnneandanny involve the acqui aahsm 41 Dalmatian dog. Perceptual learning involves learning the relationship betsy ian important process in perception. What do you see in this image? Once, oon itis difficult not to see it the next time the image is viewed Figure 7. sensory stimuli have seen the dog, t Memory so» Non-declarativerimplict Ti. ZINN Facts Episodes Procedural Priming Classical Non-associative skills conditioning learning figure 7.12 The division of memory into declarativelexplicit (conscious) and non-declaratv! implicit (non-conscious) systems. (Griomeageartitt tere eonseiousawargnes: itis poss ela arned. In contrast, procedural te En occur in o withoucbaSei6Us'aWaHEHESOF the learned materials! somes le without awareness of thejlearning event (figure 7.12). n pur Much of the research on Pavlovian and instrumental learning has bee" behavioural, without recourse ro neurophysiological concepts. Now, at about the time as the experimental study of learning was taking off in earnest, important AETE NE Me ia atts 2a) es were being Made in neurophysiology that influenced a diferent way of thinking pout learning, and psychological processes, Especially important in this regard was the SPeovery of the NeUFON as a discrete processing unit in the brain this is credited to amon Y Cajal, who with Camilo Golgi won a Nobel Prize in 1906. Once the neuron had been identified it was natural to think in terms of a network of neuronal inter connections. The most influential psychologist in proposing, Jinks between psychol K YOURSELF. 9): his inlucncevcamibeweeh i Hetil net- | Can you think of any | examples of learning that , qisaaiaamanieshey Before turning to neural networks Pee aie wank ‘ fit the learning prom and neural plasticity the next section considers the central con- | &2 NOt EEG Manning p cesses discussed above? bution of Hebb. { Hebb’s Theory Donald Hebb's (1949) a lDhicory called the neu- attention to the promising work of th (1943) ~ in fact, this;was:thesfirstmneuralmnetworky = by an interdisciplinary collaboration Of ayseniorspsychiatrist/neuro- physiologist(Me@ulloch) and a Yo pgralismaisamins) Hebb was to state presciently, “Undoubtedly there is great potential value in such work...” (p. xii). It was to be another 40 years for the full potential of unis work co be appreciated. Before the full realization of neural network ideas in the 1980s, neurophysiological research was inspired by Hebb’s «wo highly influential postulations concerning: (a) and (b)ISyMAPLICNN strengthening, (ell assembly An active neural circuit, or cel : Telbasgembly DA reverberatory circuit, He cated a - acesbetwveeneurons hs in order fe me a ane fr O . . eels ec aa wire togetheree Foundations ext ND . . va Synaptic strengthening hat ns rength_of connections betwee , stulate, states: that e : cond postula ee ofcomeiaion netween-shelk act, In Heth Hebb‘ increase: 1x proportion tothe degree 0! words eratory activity 2, fy, Te sumption aha! he a pse(this process has nowy confirmed by studies of loug-iemm potentiation: see the ‘Neural Plasticity’ section tn Hebb (1949) svent on to suggest a possible neturophysiological pro change: ‘the lost probable suggestion concerning the Wa} » thea contact . . . with the second cal! @ 8) > ~ ‘As noted in the following section, there is evidence for thy ASK YOURSELF specific suggestion. With Hebb’s insights in mind, let usar How does Hebb’s theory on to considering how- learning. relate to classical and n = instrumental conditioning? Artifical Neural Networks One useful way of thinking about the relationship betweei/léatniigiali the ae See mince, terms of. artificial neural networks implem in late Geman brain ae neither ean eons te eg the workings‘of ti@'¥alineuron (see chapter 3); they are an example of therconceptual enous sytene( Hebb, 1955); they are to use wate and Pitt’s phrase)neuro:logical iHOUEls? These computer i ; ul ig and (For an introduction to connectionism in co, tee nitive psych & Hump 1999; Harnish, 2002.) bas logy aia F 7 which alls tional compute mi Cifiefunctionsein cor syst the basis,ofisome learning rule (see below).TUM ELC le gods ae i) rk on neural networks has been motivated by the recognitio ain wo pantbaihe humanby enpaesinansentie i TERERENTy from thejconventional digital computer. ‘The brain cot Iso isa isthe ©] ; neurons and their interconnections). ‘has peychological functions are highly"coniplex athiematcaly, entailing non-linear trans formations which either are unknowmorcannot-be-casily-modelled. Neural networks ter modes is that degraded ther ike human brain damage, eens: to work, albeit less effectively,.when the network js ‘lesioned’. Also, rather like the r tem can-reorganize after damage: In other words, it.can relearn. Neural networks are now i ions, many of. whichshave nothing to do ind the worl for example, face recognitions predicting: stock-market mi dynamics. This discussion concerns only those networks relating to the biological brain Good introductory texts on neural networks include Carling (1992), Aleksander and Morton (1995), Rojas (1996) and Haykin (1999). Some of the earlie m psy- i ay be traced tolSigmund Exner, who, in 1894, published ‘Mental Phenomena,.which contained a monograph, Proj a number of drawings of neural networks (figure 7.13) Figu magne Sigmund Exner (1894), essing at se isness’ imeInput—output \ x functions N X20 WL We = | ee | ~ Wa ~ ] : = Inputs “Wa >{ Threshold | Transfer Xs, Ws 7 value value 7” x 86 A Wy . % Wa x, . : Neuron con Excitatory in igur Gi PUL our > Inhibitory inpwtioupes e 7.14 i (neuron,,,). A processin element com, siferent exctane ara neu inhibitory inputs (x, to x,), each of which has a dij influence; w, to w,) erent wont found: (a) the ° eight im (input value multi fun XW + 2Wy + X,W,)), and th his el Sa F the the weighteg Value go hold value; and (b) W Summed j eee 9., linear function). A neural netw 101 ork consists of net ement is the foundation of all Mput Works oF sich neural network Models. At about the same time, another Sigmund, this time Freud, also flirted with the pe. sibility of ° his promising his fantasy theo: ry (Stein & Ludik, 1998) Hower: (1895) was rapidly abandoned in network model w: favour » he first development of a formal eur 943). After a period of disillusionme ‘Os, the 1980s saw a Tesurgence of neural network models in the 1960s and 1971 and today such models are commonplace ing together knowledge networks i: i Although there are si networks, interes, Processing element (PE; neuron,,,,) (conceptual or artificial neuron) (figu Pim 7.14; the conceptual neuron wm has a weight (Le, the amOURE GF influence)(recall that the hillock). jyeneuroMoseharsastnansfemulue, which species hye msnanttyfininstesetnans ; the firing y should fre fofpaidbasimpuataenna tthe learning law nngainethessystemelhisslearning law may be based on a number of different learn hreshold of yearning rules \ number of learning rules are available works. These rules are mathematical algorithms used to determine the weights of inputs Hebb’s Rule ‘This is the oldest and best-known rule, introduced by Donald Hebb (1949) This rule states: if ameuronlfSeeNSiaiinputifrom anotherneurongand if both=arehigh!y active qndjhayesthe:same'signi(either excitatory or inhibitory). then,the weight bermicen the peypansslestrenatbened, Hopfield Rule. This rule is similar to Hebb’s Rule; it states: if the desired output and the inpupjarebothjactiveror bothyinactiveWlinicrementsthesconnectiongweight by therleamnntnns, rate, otherwise decrement the weight by the learning rate (i¢., a predetermined-constant).~ Delta Rule. This is another variation of Hebb’s Rule, and is widely used in neural networks. In this system, it n, 01 ights. The procedure kK processing neuron,,,, With the large rs. Input and output layers 7 Let us illustrate how a simple WOUFaDWETWOFK May WOFR wittilG CAANNpIE UE SACCET recognition. The problem faced by this simple netorksissisiivlightondarkyOne neuron... ca (light or dark; neuron, neuron,,.,). This input layer ig that the inpu the neurons would i d thus is ork. This processing may be vl us give each input a value of 1 for ‘light and —t for Sik: when the naive FE FE “here pn is one eon, _dark’y-thenethe=valueawould be 1 “value if0 would indi ween ‘dark’ and ‘light’ (ligure 7.15) T his simple neural network cont Neurons . i me importanl on ATE SC} sol a computer may . rf S '8e would be read as the onesponding binary code). Sebys ; ‘Dark’ Light Dae reuton neuron Figure 7.15 2x2 light/dark input layer. The 2 2 Pe era Sensitive t0 ing, < input value of ~1; input value of hat ‘nd to light-dark (dark has an input value of ~1; light oe intermediate values ‘Pom 1 to +1). When the input is light” the input value is a (, i when dark, -4 (4 x -1). Association (hidden) layer Figure 7.16 Input, association and output layers. Most neural networks contain (a) an ings layer; (b) a (© an association layer, itor fe underlying structure of Joutput pattern (this layer can be scrutinized by statistical analysis to determine i structure). wats (that are pre-specified) that, after appropriate tra! . a respond to the input. Often there are intermediateassociation:layers’ beoween 100" ons xjaumlaveneT complex systems, these ee infer complex relations and (! 16), Let us consider a more complex problem using light/dark neuron: Jewer (figure 7.17). The input array could comprise a 5 x 7 grids (in this case 26, one for each letter of the alphabet) would share this 35-pixel iMTaam Tel) (ire a ecaatag PZ) (a) Start of training Tr Input stimulus (b) After training "T neuron maximally stimulated Figure 7.17 Neural network T. This example shows a 5 x 7 output grid and a set of 26 input neurons (only one is shown) that correspond to the letters of the alphabet: these neurons represent a 35-pixel input space. At the start of training there is only a random corespondence between input and output (panel (a)); but, as training progresses, the system learns that the output of a specific input pattern of light and dark corresponds to the letter T (eanel(b). This is achieved by input weights changing until the neurons... conver ge on the Correct T set (ie., there is maximal activation of T neuron and inaction of all other neurons treurons, 3) 4 the beginning of training, there would be little correspondence'betWeeiliniput and ‘stput (panel a), but as training progresses, the sy; ‘omrespondi of ight Pr mn that pr Ss to A 'ever'T (panel b). To achieve this end, the weightings between neurons,,, could change ee Until they eventually converge on th correct set to allow the output to cor- 1 i , . oe i ‘o the input. At each random change of wei the al "model against the standard input would be assessed: i the random changes worsenanew random change is r i Jone, a” the T incormanly one (He The T Meuron ed, n, cura”, weighted inputs thar os My, vt ant rah d¢ neu ‘ ‘b, 1 mo ay "1 avin ant by being " ke the following form. (a) The desitey oy c i to the [etter re training WO" fe “that should learn this pattern, (¢) 1, ocifie cases gelecte in In this spect n, led : ween the pixel and ¢ 4 pattern ran (A co on ne of cach Tink oe Mach i A between iy ’ 1 jg added to the e weight the in, active pisel, TIS Js, 1 is subtract’ from 8 resentation of the pany 2a Tec es nin Pe piece SH Seon ” is and the coer we en ae tis the oulp’ oh imall “ ning. _— to this weighting One solution than trial an error. An important feature op & jin networks where only a portion of the data is, re highly efficient ‘ach the final solution. be used to tral make inferences 10 re method is mo! method is that it can able ~ that is, it can Psychological processes seful for understanding how neural networks work, but psychological purposes they are too simple. Psychological problems are complex - eres apparently simple ones — and require interactions, often considerably complex one, between a large number of neuronson Under these conditions, the system is allowej der to achieve a correspondence betwee to develop its own network of neurOMS., in oi ‘This procedure is an example of unsupervised trainin or inpaegudisnecficd gaa ther form of training is the network “The above examples are Us in, Learning rules are applied in ores ; et rected a stonlloyeasystem to sete Upo a table SolutioneThe precise way in which the systn achewes soups not clear, and for this reason farther analysis ofthe system a ~ be needed, With five association layers, the final network might look something i that shown in figure 7.18. " _, nm ble n conarcingpycholgicly viable network concerns the deget Ci as H.W meron ae to well specie (overt), den hy ee He j Ea ‘orm iwiesly on the trained set of inputs-ourputs; but they may be —- fa “ a ; to rae novel or degraded input, and thus do not show the is characte ' eae te a real brain, On the other hand, the a balance to be achieved between under. and Poorly onthe tained st. There #4 inder- and overiting the model Neural networks and associative learningInput Association: Output layer five layers layers Figure 7.18 Network: input, five association layers, output. This model could start with a number of inputs: for example, type of UCS, maghitude of UCS, type of CSs, prior associative strength of CSs, number peels (eg., hunger) of subject. These "data would represeiit the input layer. Next, the model could specify behavioural responses: for RR, learning and general- ization curve, extinction r en the system could be trained usi Then the network could form an intemal represeitation, Via the association layers, to provide a correspondence between input and output. The adequacy of the model could be tested by inputting new variable values that have known (behavioural) output values. Assuming that an adequate internal representation of the relation between inputs and outputs has been achieved, then two further things are possible. The first is to enter new values (e.g, imaginary experiments) and observe the outputs. This could lead to new knowledge. Second, the network could be interrogated and the structure of the association layer (via statistical analysis) ascertained in order to determine the — causal structure underlying associative conditioning, It is | a porte ‘ "Si | How are neural networ hoped that by such analysis of the network's interconnections ed real into the workings of the real igh maybe gained imo the workings Belton nervous system. . neural Plasticity Neural plasticity refers to the brain's.capacity to change.and.reonganize.in,responsesto | sources. Plastic op: tion ofPerec hat neuronal changes correspond to learning? In other wor the braj ery that makes learning and memory possible? Much of the work ai i. in med ary this question has adopted Hebb’s (1949) theon "i ‘see above), which after eT ee FR Ng tem memory by permanent eu chaMEE inthe ran. eh, | and neuronal alterations). ig ‘There is ample evidence that changes in the brain can be observed fol}, lation and learning. For example, rats reared in an enriched environment ie, 4, ™ ‘the home cage that are novelyandvallow exploration “andvactivity), compa; ve "oy exp dint Tight, quiet cagey;sha lowing si, red with. reared in a nt (i me ve a : ie cme ee ood 2 “i protein content (Rosenzweig & Bennett, 1996). Turner and Greenough (193). ™ ) zone of the synapse) Greenough, Juraska and Volkmar (1979) found Greenough (1982) divided the rats PU cal sum, * [gig gsinsmtentorerenenmeweeeh on! gansshemisolaazay hhe results showed ih, . changes in but is still involved in learning, LTD may be defined aptic membrane to a Synaptic input caused by stimulation of Presynaptic ter tons. One role for LT) TP. The brain has billions of neurons, and, ite. New neural circuits need to be formed, fulness need to be weakened. LTD plays a ‘minal but- although this number is large, it is not infin and old circuits that have outlive ed their us crcl role in this neural reorganization, Dudek and Bear (1992) stimulated neurons in hippocampal tissue with 900 pulses, deliv- ered at between 1 and 50 Hz (ie, cycles per s ju led to LTP, but frequencies below 10 LTD has been found to occur under a number of conditions. For example, at th raed As phe ee EP Nenbnanelsetherweaydeponeean a Polarized. As high frequency stimulation is moteslikelystovactivatesthe postsynaptic membrane (by temporal summation), this is more likely-to lead to depolarization and as if neurons in this reverberatory circuit are all‘singing fromthe same hymn sheet”. Mechanisms of action of LTP and LTD What causes the increase in ” There are a number of logical possibilities. They could be produced: (a) neurotransmitter; (b) Pessonapilly by iced number oe ea (0) by ii i synaptic membrane; (d) em. br by increased entailing both pre- and postsynaptic alterations. Experimental evidence is available to support all of these possibilities (Carlson, 2000), Research has focused on postsynaptic sensitivity. Experiments have demonstrated that : ith synaptic strengthening takes 7 zation d by ; |'stte 7.25). For example, if depolarize the hippocampus are stimulated by ¢ neurons is strengthened his is the fundamental i . rp sume tine as depolarization ofthe (eceiving a tization a to repe; r. It should be noted at this “order for the neuronal changes that underlie LTP to occur. Point th, lendritic depolarization is not the same as depolarization of the axo ‘at this de ization is dep resi k; see chapter 3), of all dendritic impulses being summed at the axon hillock: see chapPeace Depolarize cell (0) Terminal button ‘Synaptic. vessel ~ Perforated synapse Dendtitic~ spine Before long-term Alter long-term potentiation potentiation Figure 7.25. Synaptic strengthening of sy napses (before/after LTP). Synaptic strengthening takes place when molecules of the i neurotransmitter bind with postsynaptic receptors lot! in the dendriticspine thatisiima statesof depolariza i crucial that depolarization 0 Figure 7.2 the (sending) neuron occurs at the same time 's. depolarization of the (receiving) neuron: at this is fundamental to LTP (a). The nature of ure alterations between synaptic oe connections is shown (b).. molecule Activatior Let . ag ofS" E “Snow turn to the molecular machinery that underlies the strengthening 0!) aptic connections (ie., the Hebb synapse) *eceptor an excit NMDA receptors , Produce LTP depends on certain changes to gluse tors, ne that concern: hi? | ign can be stimulated by the chemical NMDA pat at s (M 5 \-methyl-p-aspartate); other glureceptor Molecule of Qg noe glutamate o_ (ms Calcium channel blocked by magnesium (Mg?*) at rest Stimulation at nearby synapse evicts magnesium — from channel @ Glutamate molecule now binds with NMDA receptor —> A and opens ion channel fawe 7.26 NMDA and non-NMDA mechanisms. Glutamate isan excitatory neurStansmitter jon channels fo a a vat the NMDA receptors, the ion channels of ‘the NMDA Molecule are blocke ich do not pass through the channel. Activation of NMDA receptors requi “ceptors also exist (i.e., non-NMDA receptors). At non-NMDA receptors, glutamate is aciatory neun opening ion channels for sodium to enter the neuron id thus ao the dendrites. At the NMDA receptors, Jutamate Fides neither excitatory nor i fects. The Teason TOF this UnFeSpOnsiveness Sanat isthat d iumoth glutamate (or the dry n sition is that the only Way Moy % agit b rors requires “The activation of NMDA receP! sean dp of magnesium ions: The Teg, the removal ee ctivate A the NMDA ghee 1, depolarization of te, repeatedly, t » same end. This depolarization te ti and enables Pe now known to be sufficient © ata toa! % electros both sodium and calcium ion: : . es; and these 7 i z drites and increase the future . By, den Kalat, 1998). LTP thus requires jn Yn iveness of the acti The | and (b) . ; both te that the ion channel controlled by the NMDA ao a Dea : voltage-dependent channel. Although the precise ee mn shows that, during LTP, the postsynaptic neuron tt hi © (Possibly ni, i ‘0 release more of its neurotransmg, ey the presynaptic neuron t wulat he rey hen sing 1e NOH NMDA receptors 7h, . when stnveraralechangess accompany these ee ted by the finding that maintenance of {7p Structural change underlying LTP is sugges' does not depend on NMDA receptors. These structural changes renal receptors. Support for this NMDA model comes from studies showing that, when thse NMDA receptors’ channels are blocked by drugs, then the establishment of LTP is abo Brown et al., 1989). ished, but existing LTP is unaffected ( then signals marize, when glutamate stimt fimulate Structural synaptic changes ‘As the dendriti inal button elongate, Hosokawa et al,, 1995). Studies have shown that ; (pee for . Permanent LTP appears to involve a further process of po t is thus dependent o1 the entry of calcium ions. {These calcium ions scm ala est enzymes kno ese enzymes set in train a chel at ickening, results in high concentrations of proteins, which can result in postsynaptic thi Structural changes may involve a growth in dendritic spines and reorientation (© bem more precise contact with the presynaptic membrane (Hosokawa et al., 19%) LTP and learning Moaweake Synapse WACEHVE (e.g, in response to a CS), he no Nas 6 UCS), also located on the CS presynaptic neuron, cause: or neuron (0 | CR), then a dend olarizes the nevenoush ™ thus pe or en permitting the weak syapse (C5 ro yn with the motor system’s neurons: an associative linMorris (2002) provide a review, of neural plasti al plas o formation. Plasticity in relation to, learning and memory rp and memory drugs receptor, enhances lea What might be the In the future, sp als. | a beneficial applications of ic drugs to enhance learni Knowledge of th : y of the neural should be available to soften the Cognitive decline associated | Mechanisms of teaming with old age, and memory? a Unifying Learning Perspectives , An interdisciplinary consideration of the Perspectives afforded by behavioural, neural mee ory and neural plasticity research provides ee weirs Powerful way of thinking about and bnderstanding learning. For exannple, thar bee been considerable debate concern. ing the nature of Pavlovian conditioning does 4 depend upon stimulus substitution or response multiplication; and is this type of learning dependent on the existence of a well. defined unconditioned response? A neural network approach call lls attention to the internal representation, or mapping, of inputs and outputs, and the complex interconnections of Processing units (neurons), which ate related by weights. This perspective diverts our attention away from the behavioural focus on the stimulus and response ends of learn- ing, and draws attention to internal states of mediation, An appreciation of the neural activity involved in learning, entailing activity- dependent Presynaptic facilitation (ADPR), long-term potentiation (LTP) and long-term depression (ETD), draws our attention to the machinery thatis needed to instantiate learning in the brain. Thus, what may look like instrumen 3ion (not requiring any response; as seen in sensory preconditioning, diately evident in behaviour; alternatively, rnin, nd on pure associ- and may not be imme- what may seem like Pavlovian conditioning may well involve a reinforcement Component, in the instrumental sense of this term, In these two cases, consideration of ADPF (im volved in reinforcement-based learning) ancl LIP/LTD iis? sila distinguish between ifferent type ing) may allow us, at the neuronal level, truly to f learning, Such information is crucial for modelling ‘hese processes in artificial neural networks the future will surely see a still g “eural networks and neurophy: hike ofthe hum; ience. Although there formidable problems still to be solved, to be sure there are considerable seienttic "nefits to b ter integration of behavioural experiments, ology, all designed better to under an brain and its unrivalled power to benefit fiom exp and the workings the concept ervous system and be derived from this joint consideration of the conceptual nervous syste oh J Sentral nervous system in lea rning.