940 likes | 1.1k Views
Conditional learning: Switching associations. Are there any types of learning that associative theory cannot explain?. "Red Light". "Red Light". "Red Light". sometimes what the CS is associated with depends on the context... e.g. lexical ambiguity task : ambiguous word e.g. bank
E N D
Conditional learning: Switching associations
Are there any types of learning that associative theory cannot explain?
sometimes what the CS is associated with depends on the context... e.g. lexical ambiguity task: ambiguous word e.g. bank subject must define based on semantic context - country walk? going shopping?
e.g. Continuous performance test (CPT) A x A y B y B y B x A y B y A y A x B x B y
e.g. Continuous performance test (CPT) A x A y B y B y B x A y B y A y A x B x B y must respond to x if preceded by A (not B) i.e. A: x+, y- B: x-
e.g. Continuous performance test (CPT) A x A y B y B y B x A y B y A y A x B x B y must respond to x if preceded by A (not B) must respond to y if preceded by B (not A) i.e. A: x+, y- B: x-, y+
e.g. Stroop! red yellow green
Patients with schizophrenia have been reported to have difficulty on such tasks: Lexical ambiguity e.g., Cohen et al, 1988 Continuous performance test e.g., Cornblatt et al, 1989 Stroop e.g., Wysocki & Sweet, 1985
Patients with schizophrenia have been reported to have difficulty on such tasks: Lexical ambiguity e.g., Cohen et al, 1988 Continuous performance test e.g., Cornblatt et al, 1989 Stroop e.g., Wysocki & Sweet, 1985 ... and in rats performance on such tasks disrupted by amphetamine -- restored by D1/D2 receptor antagonists
e.g. Dunn et al., 2005Instrumental biconditional task: Tone: Left lever --> food Right lever --> no food Click: Left lever --> no food Right lever --> food control 0.5 1.0 1.5 dose of amphetamine
So how are these tasks performed? They cannot be the result of simple associations • Tone: Left lever --> food Right lever --> no food • Click: Left lever --> no food Right lever --> food • Tone and click paired with food • Left and Right paired with food • One interpretation is that they are a type of conditional learning
Conditional cues are interesting because they are not just Pavlovian CSs... • - occasion setters do not extinguish • get better occasion setting when occasion setter is not a good predictor of the US • Simultaneous: light+tone-food tone- • Serial: light……tone-food tone- • so what are they? beyond the association...
Outline of lecture: Consider how conditional cues work Consider how conditional cues form Look at what conditional cues can do -- example of therapeutic implications
Illustrate with experiment with pigeons (Bonardi 1996) Click: red-->food red -->nothing Flash: green -->food green -->nothing Birds trained with two occasion setters - 10-sec presentations of diffuse cues - followed by 5-sec presentation of a keylight food Click red red
Question 1: How do they work? red food ? Click
Rescorla’s modulation theory (Rescorla, 1985) red food Click Whenever a CS is presented, it must activate the US representation to get a conditioned response. If the click is a positive occasion setter, it lowers its activation threshold -- making it easier for the CS to activate.
Holland’s and-gate theory (1983) food red Click The click acts as an and-gate, allowing activation to flow from the CS to the US, and so elicit a conditioned response.
Click: red-->food red -->nothing Flash: green -->food green -->nothing These theories make different predictions about transfer - will the clicker elevate responding to the green keylight?
These theories make different predictions about transfer - will the clicker elevate responding to the green keylight? green red food Click Holland says no Rescorla says yes
e.g. (Bonardi 1996) Click: red-->food red -->nothing Flash: green -->food green -->nothing Test: Same: Click: Red Flash : Green Diff: Click: Green Flash : Red
Click does elevate responding to green keylight ......so is Rescorla right?
but this transfer could be generalisation - red/green confusion .... so transfer not conclusive evidence for Rescorla
Plus Rescorla predicts click should be equally good with red and green - and it isn't... So can Holland's theory do better?
Holland predicts incomplete transfer - so is he right? --or is this generalisation decrement? Click: red-->food red -->nothing Flash: green -->food green -->nothing Test: Same: Click: Red Flash : Green Diff: Click: Green!!! Flash : Red!!! Novel stimulus combination disrupts responding on Diff trials????
(Bonardi 1996) Tests this possibility: Group OS - Click and Flash are occasion setters Click: red-->food red -->nothing Flash: green -->food green -->nothing Test: Same: Click: Red Flash : Green Diff: Click: GreenFlash: Red Expect more responding on same than different
(Bonardi 1996) Tests this possibility: Group pseudo OS - Click and Flash are NOT occasion setters Click: red-->food red -->food Flash: green -->food green -->food Test: Same: Click: Red Flash : Green Diff: Click: Green Flash : Red If previous result generalisation decrement, predict same here If previous result occasion setting, do not...
Group OS Group POS So it's not generalisation decrement... can you think of a reason why you might get more responding on same than on different trials? Wagner is a clue....
So occasion setter is specific to the particular CS... (and Rescorla himself has generated evidence against his theory e.g. Rescorla, 1991a; 1991b) - so probably Holland wins... But how about the US? is click more effective with CSs paired with the same US as red? food red Click
So occasion setter is specific to the particular CS... (and Rescorla himself has generated evidence against his theory e.g. Rescorla, 1991a; 1991b) - so probably Holland wins... But how about the US? is click more effective with CSs paired with the same US as red? different food green ? red food Click
Morell & Davidson, 2002 Light: tone -->suc Light- tone- Group Same US click-->suc or Group Diff US click -->oil Test: Light tone Same CS Same US Light click (group Same) Diff CS Same US Light click (group Diff) Diff CS Diff US
Same CS Diff CSDiff CS Same US Same US Diff US So if you change the US get dramatically reduced effect
So occasion setter is specific to the particular US as well... Conclude: evidence supports Holland - acts on association - do get transfer, based on CS-CS and US-US generalisation - if change anything in association, transfer attenuated different food green red food Click
Question 2: How do they form? If this is not an association, then where does it come from? red food Click
Where do associations come from? Associative learning explained by Rescorla-Wagner model (among others) Shows certain characteristics - e.g. blocking: A-->food AX --->food X? learning about X poor red food Click Does occasion setting show blocking?
Biographie de Al CAPONE : Ni McGurn, ni Capone ne pensèrent un seul moment que l'assassinat planifié de Bugs Moran serait un événement qui deviendrait notoire pour plusieurs dizaines d'années. Capone se prélassant en Floride, comment pouvait-on le rendre responsable du meurtre d'un contrebandier ? «Machine Gun» McGurn avait le plein contrôle de l'attaque. Il rassembla une équipe de première classe composée gens de l'extérieur : Fred «Killer» Burke en était le chef et était assisté par un tirreur du nom de James Ray. Deux autres membres étaient John scalise et Albert Anselmi, qui avaient été utilisés pour le meurtre de Frankie Yale. Joseph Lolordo faisait aussi partie du groupe, tout comme Harry et Phil Keywell, du Purple Gang de Détroit. Le plan de McGurn était créatif. Un contrebandier invita les membres du gang de Moran à le rencontrer dans un garage afin de leur offrir du whisky de qualité à un prix imbattable. La livraison allait être faite à 10h30, le 14 février. Les hommes de McGurn allaient les attendre vêtus d'uniformes de police et d'imperméables, donnant l'impression qu'un raid allait se dérouler.
Blocking of occasion setting (Bonardi 1991) Group Exp Click: noise-->food noise -->nothing Rats: 3-min presentation of a click, with embedded reinforced 5-s noise presentations. Outside click noise nonreinforced Click + + + + + noise
Blocking of occasion setting (Bonardi 1991) Group Exp Click: noise-->food noise -->nothing Group Con Click: noise / food noise -->nothing Click + + + + + noise
Blocking of occasion setting (Bonardi 1991) Group Exp Click: noise-->food noise -->nothing Group Con Click: noise / food noise -->nothing then all Click+Light: noise-->food noise -->nothing test allLight: noise ?? noise??
so blocking of occasion setting occurred - learning about L blocked only when Click an occasion setter Group E Group C
And just in case you weren't convinced... similar experiments have been performed to demonstrate the parallel effect with Sd's Click + + + + + response
Colwill & Rescorla, 1990 again looks at blocking, but relies on principle of unblocking mismatch between outcomes in two stages attenuates blocking e.g. Dickinson Hall and Mackintosh, 1976 A-->sh AX --->sh X? learning poor A-->sh AX --->sh-sh X? learning restored A-->sh-sh AX --->sh-sh X? learning poor A-->sh-sh AX --->sh X? learning restored
Colwill & Rescorla, 1990 noise (Sd) lever sucrose noise (Sd) chain food
Trained two further Sd's in compound with the noise: Stage 2: Same Different noise (Sd) lever sucrose noise (Sd) chain food noise (Sd) lever sucrose light (Sd) chain food noise (Sd) lever food flash (Sd) chain sucrose
will the Noise block the Flash or the Light ?? Stage 2: Same Different noise (Sd) lever sucrose noise (Sd) chain food noise (Sd) lever sucrose light (Sd) chain food noise (Sd) lever food flash (Sd) chain sucrose
animals learn the same thing about the noise in Stage 1 as they do about about the lightin Stage 2 Same ---> blocking but animals learn different things about the noise in Stage 1 and about the flashin Stage 2 Different--> unblocking therefore predict more control by the flash than by the light
Flash Light