Learning Pt4 PBS5

  1. What is generalisation?
    The transfer of past learning to novel events and problems.
  2. Learning individual associations can be made more __ if __ __ will allow responding to __ from previously learned assocaitions to new but __ stimuli. However, __ __ __ predicts that degree of __ depends on __ __ __ of other __ __ lying along the same physical continuum as the earlier stimulus.
    • efficient 
    • stimulus similarity
    • generalise
    • similar
    • associative learning theory
    • generalisation
    • interactive associative strength
    • conditioned stimuli (CSs)
  3. Explain peak shift. What study?
    • A phenomenon, observed in stimulus generalization, which takes place after discrimination training comprising two stimulants across a typical dimension (eg. light wavelength). The peak of the response gradient is moved in a direction away from the stimulant that was not reinforced (S-) to a place past the value of the stimulant correlated with reinforcement (S+).
    • Because one stimulus is reinforced (S+) and one isn't (S-)
  4. What is the theory behind peak shift?
    • Spence's theory of peak shifts (is this the same as associative learning theory?)
    • Because of combination of excitatory and inhibitory gradients established around S+ and S- respectively. 
    • Net association is sum of these gradients
  5. Give example of animal study testing peak shifts.
    • Hanson (1959)
    • pigeons
    • Trained to peck at 550nm light (food reward)
    • Some of these pigeons (S- group) were given additional discrimination trials with similar light (555nm), that was NOT reinforced.
    • Result: peak responding of two groups differed slightly. S- group peak responding shifted away from the nonreinforced stimulus (555nm), thus no longer having peak shift at 550nm but at slightly shorter wavelengths.
    • StimGen.GIF
  6. However, is this peak shift effect in pigeons seen in humans too?
    • No - humans often use cognitive rules (unlike pirgeon's basic associative system)
    • Accuracy continues to increase the further one moves along the dimension of stimus - response accuracy at highest at the most extreme values of the test range rather than peaking at stimulus values close to training stimuli (S)
    • Discriminatory behaviour in animals tend to be controlled by elementary physical stimulus properties rather than abstract conceptual relationships between stimuli
    • Extreme stimulus values elicit less responding (in pigeons) because they are less similar to S+ in terms of their absolute physical properties, but for humans, they are easier to classify in terms of their abstract/symbolic relationship to the training stimuli.
    • eg. 2 visual stimuli (2 shades of grey) might be similar to very dark or very bright stimulus, but abstract relationship between the stimuli (one being brighter than the other) can provide better conceptual basis for classifying stimuli.
    • Classification by human participants appears to be much more flexible and can engage reasoning and rule abstraction
  7. However, there are studies which show, in some cases, humans do show this peak shift effect. Give an example study.
    • Wills & Mackintosh (1998)
    • Trained and tested participants on stimuli from an artificial dimension rather than natural (like luminance) - where it was not possible to formulate a simple rule like one can with luminance
    • Method: Participants presented with stimulus - containing 12 meaningless small icons and must respond by pressing one button or another. They are told whether they have got it right or wrong. 
    • A Near+ stimulus will share 2 icons form the S+ stimullus (at different frequencies) but will also include one novel icon. Far+ will share 2 icons with Near+ and 1 with S+, again, shifted.
    • Result: Unlike with physical dimensions, participants' accuracy was worse for more distant stimuli, and much better for near stimuli, showing a peak effect like pigeons.
    • Explanation: One cannot look at one set of icons and say it is S+. Therefore, difficult to identify stimuli and work out its position on dimension - thus unable to devise a cognitive rule. 
    • Had to rely on associative system, whereby it was possible to generalise to stimuli which showed some similarity to S+ - but difficult to generalise for those which didn't. 
Card Set
Learning Pt4 PBS5
Lec4 - Associative and Rule-based learning