Learning to smell reaps rewards

Single-neuron recording studies in vertebrates and invertebrates have shown that associative learning can alter the response properties of cells. Less well explored is how whole networks of neurons respond to learning situations. Faber and colleagues1xAssociative learning modifies neural representations of odors in the insect brain. Faber, T., Joerges, J., and Menzel, R. Nat. Neurosci. 1999; 2: 74–78Crossref | PubMed | Scopus (208)See all References1 present a study that uses optical imaging in a simple invertebrate preparation in order to observe neural representations for different stimuli, and to investigate whether these representations are altered by associative learning. In their experiment, honeybees were trained to discriminate a rewarded odor from an unrewarded one. A third odor was not presented during training and was tested before and after conditioning as a generalization control. Two main changes in neuronal activity as a result of training were observed. First, it was found that activity (as measured by functional calcium imaging) for the rewarded but not the unrewarded odor was increased. The control odor produced a slight increase in activity, likely to be a result of generalization. Changes in behavioral response probabilities with training corresponded to the changes found in neuronal activation patterns for odors: honeybees learned to respond differentially to rewarded but not unrewarded odors, with some degree of generalization to the control odor. The second, perhaps more interesting, observation was that the activity patterns for rewarded versus unrewarded odors became de-correlated with discrimination training, thereby making the stimulus representations less similar and presumably easier to discriminate. These data suggest potential neural-network mechanisms of both associative and perceptual learning, and provide much food for thought for those interested in developing computational models of these processes.