A neurocentric approach to Bayesian inference

Source

  • one-page paper arguing that Friston's free-energy view may not be complete. Some points made are

    1. the inversion operated assumes a generative model

    2. the use of surprise is defined using a frequentist approach not informational

      • one idea : from the frequentist measure one one can derive a conditional probability (a Xhi-2 distribution) of the probability. Not very far to the idea of Sahani & Dayan of a double probabilistic distribution

    3. explore surprise or avoid it: Fiorillo makes here a confusion of time scales. On the long term (learning) one tends to avoid surprise, on the short term (coding) this implies one jumps one surprise.

    4. points to his PLoS one paper: Fiorillo, C. D. Towards a general theory of neural computation based on prediction by single neurons. PLoS ONE 3, e3298 (2008)

  • once again, people love bipolarity: frequentists against probabilists, top-down vs. bottom-up, neurocentric vs global. neurons, areas, brains, groups of brains just don't care and evolve. it is our description that can be multiple. does a single one ("unified theory") exists iun today's language? at least I am convinced that (over generations) neurons adapt to behavior not the inverse, thus that if one has to seek for an information measure, it is certainly not in a ion's channel dynamic only.

  • the answer of Friston goes into that direction / correctly defines surprise / nice figure showing how one can learn "to be a Lorenz attractor" (certainly assuming a generative model of dynamics)

  • the open question is rather "how is the free-energy principle encoded in the brain's architecture and dynamics?"

reference

  • Christopher D. Fiorillo. A neurocentric approach to Bayesian inference, URL . Nature Reviews Neuroscience, 11(8):605, 2010

    A primary function of the brain is to infer the state of the world in order to determine which motor behaviours will best promote adaptive fitness. Bayesian probability theory formally describes how rational inferences ought to be made, and it has been used with great success in recent years to explain a range of perceptual and sensorimotor phenomena1, 2, 3, 4, 5. .

  • Karl Friston. Is the free-energy principle neurocentric?, URL . Nature Reviews Neuroscience, 11(8):605, 2010

    Recently, a free-energy formulation of brain function was reviewed in relation to several other neurobiological theories (The free-energy principle: a unified brain theory? Nature Rev. Neurosci. 11, 127–138 (2010) .