Perceptual Learning

Perceptual Learning

1. Discrimination Learning

   • Problem: How can you tell if an animal can discriminate between two stimuli? Solution #1: It’s possible that the two stimuli may elicit different unconditioned responses. But, what if the stimuli are behaviourally neutral and do not elicit unconditioned responses, or may elicit the same unconditioned response. Lights make an animal orient, noise makes an animal head jerk Slightly different. Solution #2 The two stimuli could be paired with different outcomes (e.g. food and no food). If animals show a conditioned response to the rewarded stimulus, but not to the nonrewarded stimulus then they can discriminate between the stimuli.
     • Easy (from difference modalities)
     • Hard
     • Very hard
   • What is the effect of preexposure on discrimination learning?
     • The effect of preexposure on conditioning: latent inhibition . We know this! Pre exposure is bad, slows it down.
     • Should be bad because retards learning. BUT gives them time to learn how to differentiate them and learn important qualities.
       • Compare preexposure and no preexposure
         • Gibson and Walk 1956 - same Gibson and the visual cliff.
           • Did this experiment in rats. Placed in cage, shapes inside. Animal could run around. Then had to learn discrimination in second stage. In tunnel, could run toward triangle or circle. Only approaching triangle will lead to food. Control animals just lived in cage wit no shapes. If latent inhibition, preexposure will make them slower. But if preexposure helps them discriminate, then control will be slower. The pre exposure helped, they were faster! Is latent inhibition rubbish?

2. Perceptual learning

   • Preexposure to stimuli aids their discrimination rather than hinders it. Preexposure leads to a process of differentiation (Gibson & Gibson, 1955). The opportunity for stimulus comparison allows their unique features to be identified. Not associative- suggests a non associative prpcess in learning
     • Over 100 years ago, William James noted: That “practice makes perfect” is notorious in the field of motor accomplishments. But motor accomplishments depend in part on sensory discrimination. In the purely sensorial field we have the well-known virtuosity displayed by the professional buyers and testers of various kinds of goods. (James, 1890) And he proceeded to give examples of the amazing feats of such experts. One man could tell whether a glass of Madeira had been drawn from the upper or lower half of the original bottle. Another could run newly ground flour through his fingers, and tell you whether it was ground from wheat grown in Iowa or Tennessee. Over the years, other commentators have provided numerous other examples.
       • Experts can sex day-old chicks, by brief inspection of their external genitalia, with an error rate of less than 1%, although they are hard pressed to state explicitly the criteria they use for making this discrimination, and novices, explicitly told what relevant, distinguishing features to look for, can do no better than about 70% correct. The photomicrograph of the stained section of brain and the X-ray revealing a lesioned disc that are readily interpreted by the expert neuroscientist or radiographer, mean nothing to the untrained observer. Mackintosh & Bennett 2009. People get better at doing something with eperience. People good at discriminating foods. People get good at discriminating things that are hard to discriminate and cant tell you how.

3. Perceptual and Associative learning

   • Is perceptual learning beyond explanation from associative learning theories? Theories that predict latent inhibition would appear to fail to explain perceptual learning. But, perceptual learning is not concerned with how rapidly stimuli condition, but how rapidly the ability to discriminate emerges

4. Stimulus representations

   • Elemental representations: stimuli are represented by a set of elements. Make some naive assumptions about how stimuli are represented. Common idea- could activate particular neurons. Two stimuli will share some common elements, but each stimulus will have unique elements. We can describe stimulus A and stimulus B as ax and bx (a is the elements that are unique to A and b is the elements that are unique to B and x is common elements that A and B share). X small if dissimilar and large if similar.
     • The notion of common elements explains why conditioning generalises between stimuli. Conditioning generalises more readily to similar stimuli (that share more common elements) than to dissimilar stimuli (that share fewer common elements). IF you learn about one stimulus it might generalise to other stimuli to the extent that they share common elements.
       • Generalisation of conditioned responding to novel stimuli --> (Guttman & Kalish, 1956)
         • Been trained to 580nm. In test, they change wavelength. Those close to trained one - still quite strong responding. Far away is weak. As stimuli get less similar, responding is decreased/ Pigeon key pecking to coloured light.

5. Latent inhibition of common elements

   • Latent inhibition will occur as a result of preexposure. However, the common elements receive twice as much preexposure than the unique elements of two stimuli (ax, bx). Are the common elements doubly latently inhibited? Preexpose then slow to learn. But might just be preexposing unique elements
     • If there are ten preexposure trials of A (ax) and ten preexposure trials of B (bx) then a will be presented ten times, b ten times and x twenty times. If the common elements receive more latent inhibition than the unique elements then the discrimination between the two stimuli will proceed more rapidly. Learning will generalise less readily between the stimuli.
   • Mackintosh, Kaye and Bennet 1991
     • Preexposure to the common element (lemon) was sufficient for perceptual learning to occur (i.e. rats discriminated between the two flavour compounds and conditioned aversion did not generalise to the sucrose-lemon compound)
   • If perceptual learning occurs because common elements receive greater latent inhibition then perceptual learning will occur only when stimuli are similar (and share common elements). If stimuli are dissimilar (and do not share common elements) then preexposure should have no effect on discrimination learning
     • Trobalon, Sansa, Chamizo & Mackintosh, 1991
       • A and B have lots in common. Only preexposure effect when hard!

6. Is perceptual learning just latent inhibition of common elements?

   • Honey, Bateson and Horn (1994). Perceptual learning in the chick. Chicks were exposed to visual images. Chicks were placed in a cool chamber. Approaching the rewarded stimulus resulted in a stream of warm air (a hair dryer being turned on). Approaching the nonreward stimulus resulted in nothing.
     • Equal amounts of preexposure but they were interested in the way they got the preexposure. Mixed group had it in random order. Blocked had all exposure to A and then all to B. IF pre exposure allows them to discriminate, then blocked are less able to learn how they differ. Therefore, mixed exposure would be better. But the latent inhibition should be equal. So it is something more than latent inhibition of common elements. They showed that the blocked group was slower.
       • Both groups received preexposure, and therefore, the common elements should be equally latently inhibited for both groups. However, mixed exposure resulted in faster discrimination learning than blocked exposure. Perceptual learning is not just dependent on latent inhibition of common elements, but also on the schedule of exposure
     • Mixed vs Blocked exposure
       • Does mixed exposure allow stimulus comparison? Does mixed exposure allow the formation of associations that aid discrimination learning. What is it about the ability to compare? Something beyond associative learning? OR forming links to allow it to discriminate?
         • Within-compound associations
           • Maybe elements within a stimulus can form links with other elements? When stimuli are presented simultaneously they form an association with one another, e.g. sensory preconditioning When the tone was presented it retrieved the memory of the light. When the memory of light was paired with shock, the light made animals animals afraid (mediated conditioning). For the learning group, light causes a freezing effect as if theres been a tone. Must be forming links and associations
         • Within stimulus association
           • If stimuli are represented as a set of elements then those elements may form associations with one another when the stimulus is presented. For example, when a black square is presented the colour black may form an association with the shape of the square. When AX is presented A will form an association with X and vice versa. A becomes predictive of X and X becomes predictive of A. When BX is presented B will form an association with X and vice versa. B becomes predictive of X and X becomes predictive of B. Unique bits form associations with common bits.
             • So what happens with mixed exposure? Trial order: AX, BX, BX, AX, AX, BX, BX, AX… When AX is presented the two stimuli predict one another, but X also predicts B, but B is not present. So what happens? According to the R-W model when AX is presented the ability of X to predict B is reduced (there is negative prediction error). A does not predict B, but it also undergoes a reduction in associative strength with B. Therefore, its ability to predict B goes below zero; it becomes a conditioned inhibitor for B. HAppens the other way around
               • What are the consequences of the unique elements (A and B) being conditioned inhibitors of one another? Anything that is learnt about AX will not generalise to BX. Why is this? When AX is presented X can retrieve a memory B, such that B can also form an association with the outcome. BUT, A will inhibit B, such that the memory of B cannot be retrieved and, therefore B will not form an association with the outcome on AX trials. The inhibitory associations between A and B stop mediated conditioning taking place. Nothing will transfer between A and B as have mutual inhibiton
       • Blocked
         • Trial order: AX, AX, AX, AX, BX, BX, BX, BX. What happens on the last exposure to AX? When AX is presented the two stimuli predict one another. X has not yet been paired with B, therefore, it does not predict B. Consequently, A has no opportunity to become a conditioned inhibitor for B. Therefore, anything that is subsequently learnt about AX will generalise to BX, because of mediated conditioning.
           • Evidence for inhibitory associations --> Dwyer, Bennett & Mackintosh, 2001
             • Stage 2 - exposed to compound of sucrose and saline, to form an association. Gave them a diet in reduced salt so crave saline. Motivate them to want salt, then test how much sucrose they drink. If they think sucrose leads to saline. Blocked -> sucrose lemon, then saline lemon. Preexosure never together. Should make them appear more different. Mixed - intermixed trial by trial. If mixed preexposure gives more inhibitory links then should be hard to learn the sucrose - saline association later (because they are inhibitors of each other). Then wont drink a=much sucrose because dont have strong link of it with saline.
               • The mixed group drank less than the blocked group. This suggests that the blocked group associated sucrose with saline more than the mixed group. Therefore, they drank more sucrose because it ‘reminded’ them of salty, saline. Why did the mixed group fail to drink the sucrose at test? The formation of inhibitory associations during preexposure reduced the ability of sucrose and saline to form an association. Found mixed preexposure drank less than block group. Blocked learnt more about sucrose-saline association. Mixed reduced their learning

7. Perceptual learning

   • Perceptual learning has been claimed to reflect learning through stimulus comparison (a process that may not require associative learning). However, by applying knowledge gained from the study of associative learning it is possible to demonstrate that perceptual learning reflects the operation of standard associative learning mechanisms. Perceptual learning reflects, in part, latent inhibition of common elements AND Inhibitory links between unique features of stimuli. Associative learning isn’t simply a mechanism that is engaged when a neutral stimulus is paired with learning shows that the mechanisms of associative learning are engaged even in passive experience of stimuli.a motivationally significant stimulus. Perceptual

8. So what have we learnt about associative learning?

   • During the last nine weeks we’ve been looking at various aspects of associative learning. However, the main question has been ‘what are the conditions that allow associative learning to occur? By answering this question we’ve studied processes of surprise and various aspects of attention. Associative learning is surprisingly complex. Numerous factors determine whether an association is learnt. We’ve also found that these processes do a pretty good job of explaining things that we may not have thought required associative learning, such as habituation and perceptual learning.

Images and files in the map