Pearce-Hall

Pearce-Hall

1. Last week... CS intensity

   • R-W model: ΔV = αβ(λ-ΣV). Within the Rescorla-Wagner model the learning rate parameter for a given CS (α) is fixed and is simply determined by the intensity of that stimulus. The value of this parameter determines the rate at which an association involving a given stimulus will form
     • Mackintosh theory - α for a given CS approaches 1 if it is a better predictor of the US than all of the other CSs available. If it is no better a predictor, or a worse predictor, than the other CSs present, then its α approaches 0, and therefore no learning will take place.
       • The success of the model is demonstrated by its ability to explain effects such as downshift unblocking (Dickinson, Hall & Mackintosh, 1976) and blocking of unblocking (Mackintosh and Turner, 1971). It provides an alternative explanation of cue competition effects (e.g. blocking, overshadowing) It can explain effects such as latent inhibition, which was beyond the scope of the Rescorla-Wagner model

2. Attention in human causal learning--> Le Pelly & McLaren 2003

   • Allergy prediction task A,B,C,D,V,W,X,Y = foods O1-O4 = allergic reactions (itch, nausea, sweating, nausea)
     • A. B, C, D are informative in stage 1, A always to 1 etc. These are better stimulus, good predictors, attention up. Stage 2 - new combinations of food. Each food uniquely paired with outcome. Equally provide correct information. Test - should be 0.3, 0.4, 0.3, 0.4. Are they better at predicting first two because of stage 1? Learnt far more about AC and BD than other. Attention in Stage 1 affected learning in Stage 2. Change with experience.

3. Problems for Mackintosh - Hall-Pearce negative transfer effect

   • According to Mackintosh for Group Experimental in Stage 1 the tone will have acquired a high level of attention (because it was a good predictor of shock). Consequently in Stage 2 attention to the tone will be high for Group Experimental, but not for the control group (because they have never encountered the tone). Therefore, Mackintosh predicts that Group Experimental will learn faster than the control group.
     • Group Experimental learnt the tone-->SHOCK!! association slower than the control group. This result is the opposite to the result predicted by Mackintosh. If attention to the tone didn’t increase then what did happen? Did attention to the tone decrease?

4. Pearce-Hall: A new account of CS processing

   • Must be some serious failings of Mackintosh model. Learning affects attention, which can increase or decrease learning. Got nothing to do with US processing, everything to do with attention to CS. Assume learning affects attention, but different viewpoint to Mackintosh.
     • Pearce and Hall’s model makes very different assumptions about what governs the attention an animal pays to a stimulus. Claims there are two types of attention or processing - controlled and automatic (see also, Shiffrin & Schneider, 1977). Automatic processing is engaged when a task is well learnt and controlled processing is assumed to be engaged during learning. Our interest is in what determines controlled processing.

5. What are the determinants of controlled processing

   • To the extent that the animal is unsure about the consequences of a stimulus then that stimulus will receive controlled processing. Once the stimulus accurately predicts its consequences, controlled processing will cease and automatic processing takes over.
     • The amount of controlled processing (attention) a stimulus received (its alpha value) on a given trial (trial n) depends on how well it predicted what happened on the previous occasion on which it was predicted (trial n-1). If the discrepancy between what the stimulus predicted and what happened on trial n-1 was large, then attention on trial n will be greater than if this discrepancy was small on trial n-1. The intuition of the Pearce-Hall model is that (controlled) attention should be directed towards stimuli whose significance is uncertain - stimuli that need to be learned about.

6. Application of Pearce-Hall to habituation and latent inhibition

   • According to Pearce-Hall habituation and latent inhibition reflect a decline in attention. When a stimulus is first presented the animal will not know what its consequences will be and will orient towards it. With repeated exposure the animal will come to learn that nothing surprising happens after the light and will not orient (habituation). Subsequently, learning about the preexposed light will be slow (latent inhibition) because at the outset of conditioning, attention to the light will be low.

7. PearceHall and Blocking

   • Pearce-Hall predicts that attention to a new stimulus will be high, but that subsequent attention to it is governed by whether the US is surprising (i.e. is the consequence of the CS unknown or uncertain?). If prior training with another CS renders the US unsurprising, then attention to the added CS will be low after one trial, and no new learning will occur. Just like Mackintosh’s theory, the Pearce-Hall theory predicts that blocking will not occur on trial 1 of the blocking phase. This prediction has received some support (Mackintosh, 1975, but it is not always found, Balaz, Kasprow & Miller, 1982).Stage 2 - not surprising - attention to light stays low. Attention to noise also goes down because what happens wasnt exciting.

8. Pearce Hall and overshadowing

   • Pearce-Hall predicts that attention to the light will decrease because the US is predicted by both the light and noise. Therefore, for the overshadowing group the US is less surprising. Consequently, attention reduces to the light.

9. Evaluating the model

   • A stimulus will be attended to when its consequences have been unpredictable in the past and when the consequences of a stimulus are predictable attention to it will decline. Predictions: Habituation and latent inhibition will be observed to stimuli that have had predictable consequences in the past. Habituation and latent inhibition will not be observed if the stimuli have not had predictable consequences.
   • Why does negative transfer occur?
     • In stage 1 attention reduces to the tone because it is reliably followed by Shock. Consequently, attention will be low for Group Experimental at the beginning of Stage 2, but not for the control group.
   • Novel predictions
     • Attention will remain high to stimuli that are followed by inconsistent outcomes Kaye & Pearce (1984) – the influence of uncertainty on measures of overt attention
       • Attention orienting should go down for bottom two, high for top. Partial - orienting will remain high.The orienting response (i.e. looking at the light) was maintained for a stimulus that was followed by an inconsistent outcome, but not for stimuli that were followed by consistent outcomes
     • Attention can be restored to CSs by the introduction of a surprising outcome. Novel and pretrain +surprise learn association much faster. Reduction in attention happens because something unsrprising happens
       • Griffiths, Johnson & Mitchell (2011) have recently shown this effect in humans using an allergy prediction task
         • Relationship of light to food exacly the same for both groups. But for shift, light is ambigous as to whether it predicts tone. Shift learnt better!
   • The Pearce-Hall model is successful in demonstrating that uncertainty affects attention (and therefore CS processing.) It makes the correct prediction that attention is reduced for stimuli that are followed by reliable consequences. It makes the correct prediction that attention is restored for a CS by the introduction of a surprising outcome

10. Pearce Hall VS Mackintosh

    • Both models provide an account of cue competition effects (e.g. blocking, overshadowing). Both models claim that learning depends on the amount of attention paid to stimuli. However, Pearce-Hall cannot explain the Mackintosh effects Mackintosh can’t explain the Pearce-Hall effects.
      • Two types of attention - Mackintosh – automatic attention. Attention influenced by predictiveness. Is this cue relevant for learning? Pearce-Hall – controlled attention. Attention influenced by uncertainty.

11. Le Pelley's Hybrid model

    • Macktintosh - attentional associability. Pearce-Hall - salience associability.
      • The Mackintosh attentional process decides which cues will receive learning – good predictors receive more attention, bad predictors receive less attention. Once stimuli have been selected for learning the Pearce-Hall attentional process decides how much learning each cue receives – cues with unreliable outcomes receive more attention than cues with reliable outcomes. Seek out relevant cues (Mackintosh), Reduce uncertainty (Pearce-Hall).
        • Le Pelley’s hybrid model provides a way of accounting for the data that uniquely support Mackintosh and those that uniquely support Pearce-Hall. Pearce and Mackintosh (2010) have also proposed their own hybrid model.
    • Two types of attention - (Haselgrove, Esber, Pearce & Jones 2010). When faced with multiple cues, the best predictor recieves the most attention. A is better predicotr, recieves attention. Then use Pearce-Hall

Images and files in the map