@article{Multidimensional:1616,
      recid = {1616},
      author = {Macellaio, Matthew},
      title = {Computational Benefits and Behavioral Implications of  Multidimensional Sensory Tuning},
      publisher = {University of Chicago},
      school = {Ph.D.},
      address = {2018-06},
      pages = {132},
      abstract = {Organisms’ behaviors and perceptions are shaped by the  manner in which the brain processes incoming sensory  information. Different sensory streams of the brain manage  and process disparate types of inputs from sensory  receptors in order to transform sensory information into  decision and action. To adequately achieve this goal,  neurons in many areas simultaneously encode multiple  features of the outside world. Jointly encoding various  features can, however, lead to ambiguity in the decoding  process: single-trial fluctuations in the spike rate of a  neuron can represent a change in any or all of the encoded  stimulus features. To explore the benefits and impacts of  joint coding, we probed the macaque visuomotor pathway. In  the first study, we used information theoretic methods to  quantify the benefit of joint coding of direction and speed  in single neurons in extrastriate area MT, in small  simulated populations, and in smooth pursuit eye movements.  We found that there is a consistent benefit to encoding  multiple stimulus features jointly rather than separately,  which we call stimulus synergy. Additionally, we determined  that the scale of stimulus synergy can be altered by the  tuning bandwidth and the number of stimulus features  encoded. The observation of stimulus synergy in pursuit eye  movements implies that the MT population is being read out  jointly, as a motion vector. In the second study, we  further investigated the interaction of direction and speed  in pursuit eye movements. We found that single-trial  direction and speed errors are correlated, bolstering our  claim that fluctuations in the MT population response are  jointly decoded to generate direction and speed estimates.  Finally, we confirm using simulated MT responses that such  correlations can only occur when decoding motion estimates  from MT as a motion vector, not as direction and speed  separately.},
      url = {http://knowledge.uchicago.edu/record/1616},
      doi = {https://doi.org/10.6082/uchicago.1616},
}