1 / 104

Configural learning Learning about holistic stimulus representations

Configural learning Learning about holistic stimulus representations. no food. food. Structural discriminations George Ward-Robinson & Pearce, 2001. food. no food. Structural discriminations George Ward-Robinson & Pearce, 2001. Can this be solved in terms of simple associations?

colman
Télécharger la présentation

Configural learning Learning about holistic stimulus representations

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Configural learning Learning about holistic stimulus representations

  2. no food food

  3. Structural discriminations George Ward-Robinson & Pearce, 2001 food no food

  4. Structural discriminations George Ward-Robinson & Pearce, 2001

  5. Can this be solved in terms of simple associations? Can it be solved with conditional learning? food no food

  6. If green: red-left + red-right - If blue: red-left - red-right + If green: blue-right + blue-left - food no food

  7. If green: red-left + red-right - If blue: red-left - red-right + If green: blue-right + blue-left - relies on use of compound cues - red-left etc food no food

  8. so why not use fact these stimuli are all unique? red-left&green-right+ red-right&green-left - food no food

  9. Some types of learning associative theory cannot explain. Last week we saw how conditional learning can explain some of these Today we consider an alternative approach - configural learning Can associative theory adapt by changing the way in which the stimulus is represented?

  10. So far have assumed that a compound stimulus is equivalent to the sum of its parts: A --> food B--> food A --> cr B --> cr AB --> CR Predict SUMMATION

  11. Feature negative discrimination A --> food AB --> no food CR cr

  12. VA=  (  - V ) Learning stops when (=V ) A --> food AB --> no food VA = 1 VA + VB = 0

  13. VA= (  - V ) Learning stops when ( = V ) A --> food AB --> no food VA = 1 VA + VB = 0 A becomes excitatory: V = +1 B becomes inhibitory: V = -1 thus A alone predicts food, whereas A+B is neutral

  14. Feature positive discrimination A --> no food AB --> food cr CR

  15. VA=  (  - V ) Learning stops when (  = V ) A --> no food AB --> food VA = 0 VA + VB = 1

  16. VA=  (  - V ) Learning stops when ( = V ) A --> no food AB --> food VA = 0 VA + VB = 1 B becomes excitatory: V = +1 A eventually becomes neutral: V = 0 Thus A alone predicts nothing, but when B is present food is expected

  17. Performance on feature negative and feature positive discriminations can be explained by the Rescorla-Wagner equation If you condition to asymptote, it predicts perfect performance But how about.......

  18. Positive patterning discrimination: A --> no food B --> no food AB --> food cr cr CR

  19. VA=  (  - V ) Learning stops when (=V ) A --> no food B --> no food AB --> food VA = 0 VB = 0 VA + VB = 1

  20. A --> no food B --> no food AB --> food VA = 0 VB = 0 VA + VB = 1 This one is insoluble - you can never reach asymptote: what is gained on AB trials is lost on A and B trials

  21. A --> no food B --> no food AB --> food But associative theory can explain accurate performance Both A and B acquire associative strength on compound trials, and lose some on element trials Animals respond more on AB trials (when two signals for food are present) than on A or B trials (when there is only one) But it doesn't predict perfect performance

  22. Negative patterning discrimination A --> food B --> food AB --> no food CRCR cr

  23. VA =  (  - V ) Learning stops when (= V ) A --> food B --> food AB --> no food VA = 1 VB = 1 VA + VB = 0

  24. Simple associative theory can never predict accurate performance here A --> food B --> food AB --> no food If A and B have enough associative strength to elicit responding, then the compound of A and B must elicit more responding, not less -- violates summation principle So can animals learn nonlinear discriminations of this type?

  25. Wagner (1971) and Rescorla (1972) suggested the unique stimulus account: A stimulus compound should be treated as the combination of its elements... + A B

  26. A stimulus compound should be treated as the combination of its elements... PLUS a further stimulus that is generated only when those elements are presented together: + A B ab

  27. A stimulus compound should be treated as the combination of its elements... PLUS a further stimulus that is generated only when those elements are presented together: + A B configural stimulus not very salient; so only learned about when absolutely "forced" ab

  28. Now the negative patterning discrimination looks like this: A --> food B --> food AB --> no food

  29. Now the negative patterning discrimination looks like this: A --> food B --> food AB ab --> no food

  30. Now the negative patterning discrimination looks like this: A --> food B --> food AB ab --> no food VA = 1 VB = 1 VA + VB+ Vab = 0

  31. A --> food B --> food AB ab --> no food VA = 1 VB = 1 VA + VB+ Vab = 0 B becomes excitatory: V = +1 A becomes excitatory: V = +1 ab becomes inhibitory: V = -2 ...and the discrimination is solved...

  32. Rescorla tested this interpretation with the following experiment: A + B + AB - AB + A ? B ? A + B + C - AB + A ? B ? Which group will respond more in the test?

  33. Stage 1 Stage 2 Test A + B + AB ab - AB ab + A ? B ?

  34. Stage 1 Stage 2 Test A + B + AB ab - AB ab + A ? B ? In Stage 1 A and B become excitatory and ab inhibitory; the combination of A, B and ab should therefore be neutral

  35. Stage 1 Stage 2 Test A + B + AB ab - AB ab +A ? B ? In Stage 1 A and B become excitatory and ab inhibitory; the combination of A, B and ab should therefore be neutral In Stage 2 the neutral AB ab is paired with food; the food is surprising, and A, B and ab all gain associative strength

  36. Stage 1 Stage 2 Test A + B + AB ab - AB ab + A ? B ? In Stage 1 A and B become excitatory and ab inhibitory; the combination of A, B and ab should therefore be neutral In Stage 2 the neutral AB ab is paired with food; the food is surprising, and A, B and ab all gain associative strength In the Test A and B now have more associative strength than they started with

  37. Stage 1 Stage 2 Test A + B + C - AB + A ? B ?

  38. Stage 1 Stage 2 Test A + B + C - AB + A ? B ? In Stage 1 A and B become excitatory

  39. Stage 1 Stage 2 Test A + B + C -AB + A ? B ? In Stage 1 A and B become excitatory In Stage 2 the excitatory A and B both predict food -- thus two foods are predicted, but only one happens; this produces inhibitory learning, and the strength of A and B drops...

  40. Stage 1 Stage 2 Test A + B + C - AB + A ? B ? In Stage 1 A and B become excitatory. In Stage 2 the excitatory A and B both predict food -- thus two foods are predicted, but only one happens; this produces inhibitory learning, and the strength of A and B drops... In the Test A and B now have less associative strength than they started with

  41. Responding to A and B

  42. So.. can Rescorla & Wagner explain everything? Not quite: consider the following discriminations: Discrimination 1: A+ AB- Discrimination 2: AC+ ABC- In the second case a common element C has been added on both reinforced and nonreinforced trials; this should make the discrimination harder...

  43. So.. can Rescorla & Wagner explain everything? Not quite: consider the following discriminations: Discrimination 1: A+ AB- Discrimination 2: AC+ABC- In the second case a common element C has been added on both reinforced and nonreinforced trials; this should make the discrimination harder...

  44. BUT Rescorla & Wagner's theory predicts that the AC+ ABC- discrimination will be learned most easily Because AC has more elements than A, it will acquire associative strength faster Discrimination 1: A+ AB- Discrimination 2: AC+ABC-

  45. BUT Rescorla & Wagner's theory predicts that the AC+ ABC- discrimination will be learned most easily Because AC has more elements than A, it will acquire associative strength faster Discrimination 1: A+ AB- Discrimination 2: AC+ABC-

  46. on first trial VA=  (  - V ) =( - 0 ) Vc= (  - V ) =  (- 0 ) So AC will have twice as much strength as A after trial 1 Faster EXCITATORY learning Discrimination 1: A+ AB- Discrimination 2: AC+ABC-

  47. And the more AC predicts food, the greater the surprise on ABC- trials, and so the faster B will become inhibitory Faster INHIBITORY learning Discrimination 1: A+ AB- Discrimination 2: AC+ABC-

  48. The faster the excitatory and inhibitory learning is acquired, the faster the discrimination is acquired oops!

More Related