rescorla-wagner model

22
Rescorla-Wagner Model US-processing model Can account for some Pavlovian Conditionin henomena: acquisition blocking unblocking with an upshift conditioned inhibition US-pre-exposure effect Cannot account for some Pavlovian Condition henomena: extinction (i.e., spontaneous recovery) unblocking with a downshift latent inhibition temporal factors (i.e., CS-US interval)

Upload: garrison-munoz

Post on 05-Jan-2016

45 views

Category:

Documents


0 download

DESCRIPTION

Rescorla-Wagner Model. US-processing model. Can account for some Pavlovian Conditioning phenomena:. acquisition. blocking. unblocking with an upshift. conditioned inhibition. US-pre-exposure effect. Cannot account for some Pavlovian Conditioning phenomena:. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Rescorla-Wagner Model

Rescorla-Wagner Model

US-processing model Can account for some Pavlovian Conditioning phenomena:

• acquisition• blocking• unblocking with an upshift• conditioned inhibition• US-pre-exposure effect

Cannot account for some Pavlovian Conditioning phenomena:

• extinction (i.e., spontaneous recovery)• unblocking with a downshift• latent inhibition• temporal factors (i.e., CS-US interval)

Page 2: Rescorla-Wagner Model

Pearce-Hall Model

attention model of conditioning

a CS-processing model

according to the model, it is highly adaptive to paypay attention to, or process, CSs that could becomevalid predictors of important outcomes (i.e., USs)

it is also adaptive not to pay attention to, or process, CSs when the important event is already predicted by something else

Page 3: Rescorla-Wagner Model

Pearce-Hall Model

also based on the concept of surprise when the subject is surprised, attention to, orprocessing of the CS occurs

as the US becomes predicted by a CS, and is less surprising, processing of the CS declines

The amount of processing, that is associability of a CS,changes on each trial depending on whether the US waspredicted (on the previous trial) If the US was predicted, then attention to the CS declines If the US was not predicted, then attention to the CS increases

Page 4: Rescorla-Wagner Model

Pearce-Hall Model

Recall from the RW Model, ΔVA = k(λ – VT)k = constant; salience or associability of the CS

With the PH Model, k changes across trials (CS processing model, not a US processing model)

Page 5: Rescorla-Wagner Model

Pearce-Hall Model

kAN = λN-1 – VA

N-1

kAN = associative strength or associability of CSA

on trial N

λN-1 = strength of the US on previous trial

VAN-1 = strength of CSA on previous trial (could

become VT if more than one CS)

Important point: k depends on what happened on the previous trial; on first exposure, novelty causes some attention

Page 6: Rescorla-Wagner Model

Pearce-Hall Model

kAN = λN-1 – VA

N-1

Early in training, when the strength of the CS is low (i.e., λ – V is high) see high k value and thus, moreattention to the CS

When the CS is strong in later trials (i.e., λ – V is small) attention to the CS is low

The important point is that attention to the CS changes across trials

Page 7: Rescorla-Wagner Model

Pearce-Hall Model

Attention to, or processing of, the CS can be measuredin terms of an OR (i.e., looking at a L)

This is different than the CR

Support for the PH Model comes from the finding that subjects orient towards novel stimuli and maintain their orientation, provided the stimulus is a poor predictor of the US

Page 8: Rescorla-Wagner Model

Kaye & Pearce compared the OR in 3 groups of rats

Group 1: L alone

Group 2: L condensed milk

Group 3: L milk/no milk (inconsistent/random)

Looked at OR to L

Attention (OR) was high on the first trial since the Lis novel

Page 9: Rescorla-Wagner Model

Group 3

Group 2

Group 1

OR to L

Blocks of Training Trials

Page 10: Rescorla-Wagner Model

Group 1: L alone

kAN = λN-1 – VA

N-1

k stays low (decrease attention)

Group 2: L milk

VA gets bigger over time which makes the total term smaller (this means small kand decrease in attention)

Group 3: L milk/no milk

Attention remains high since VA is low

Page 11: Rescorla-Wagner Model

When the CS is not a good predictor, rats maintained their attention to the cue

If the CS is a good predictor (of the US or no US), then attention decreases

Page 12: Rescorla-Wagner Model

Pearce-Hall Model and Blocking

like the RW Model, all CSs combine to predict the US

if one CS already predicts the US, then pay less attention to all CSs on that trial

when a new CS is added, should pay attention to itbecause it is novel

therefore, should see some conditioning to the new cueon the first trial based on the salience of the CS

Page 13: Rescorla-Wagner Model

Pearce-Hall Model and Blocking

only after first trial is over would the animal know that nothing new had happened

according to the model, should see blocking from trial2 and onwards

however, in most cases see blocking right from the start

Page 14: Rescorla-Wagner Model

Pearce-Hall Model and Unblocking

when subjects encounter a US that is not well predicted,or is surprising (either bigger or smaller), then subjectsshould pay attention to all CSs on that trial and getunblocking

kAN = λN-1 – VA

N-1

because the formula includes the absolute value ofλN-1 – VA

N-1 it doesn’t matter if the US is bigger or smaller

if the US changes we’ll see increase in attention and thus, learning

Page 15: Rescorla-Wagner Model

Pearce-Hall Model and latent inhibition

When the CS is given by itself, see decrease in attention to the CS over trials (λ = 0)

However, a problem with the model is that it cannot explain the context-specificity of LI

If CS pre-exposures are given in one context, and conditioning occurs in a second context, there is noretardation of learning

According to the model, k should be low regardless ofcontext

Page 16: Rescorla-Wagner Model

The Comparator Hypothesis

developed by Ralph Miller

this is a model of performance, not learning

according to Miller, all CSs have excitatory power;there is no separate inhibitory process

the strength of performance (or CR) depends on therelative strength of the various excitatory associations

a subject compares the excitatory strength of the explicit CS to the strength of other cues present in thesituation, such as apparatus cues

Page 17: Rescorla-Wagner Model

The Comparator Hypothesis

when the strength of a CS is relatively greater thanthe background cues, get a measurable CR

when the strength of a CS is weaker than the background cues, get weakened level of excitation (whatothers might call inhibition)

according to the theory, the competition between twoexcitatory reactions controls performance

Page 18: Rescorla-Wagner Model

The Comparator Hypothesis

during normal excitatory, get CS-US pairings – but the US is also paired with background cues and thesebackground cues are the comparator stimuli

because these background cues are also present duringthe ‘no-US’ condition, they are typically weaker than the explicit CS

so, under normal conditioning procedures, the CS hasstronger excitatory strength than the comparator cues

Page 19: Rescorla-Wagner Model

The Comparator Hypothesis

during inhibitory conditioning, the CS is weak relativeto the background cues

during inhibitory conditioning, have CS – no US pairings; but the background cues are paired with the USand the absence of the US

thus, the CS is weaker than the background cues and see little CR to the CS

Page 20: Rescorla-Wagner Model

• Prediction – After training one can manipulate the excitatory value of the context and this will affect the excitatory value of the CS

• E.g. – After conditioning, give repeated exposure to the context alone followed by exposure to CS

• One will see greater responding to CS

The Comparator Hypothesis

Page 21: Rescorla-Wagner Model

Temporal Factor Models

designed to explain the effects of time in conditioning

effects of time not considered in US-processing modelslike the RW model nor in CS-processing models like thePH model

CS-US interval is one important temporal variable

a more critical temporal variable appears to be the ratio of the ISI to ITI

Page 22: Rescorla-Wagner Model

Midterm ExamThursday, Feb. 17, 2005

-covers everything up to and including today’s lecture-in the case of a storm, the exam will take place during the very next class