2011-07-25 computational and theoretical neuroscience

2011-07-25 09:45:31

the poll

  • On april 1st, 2011, I sent the following message to the comp-neuro and connectionists lists: (see http://www.neuroinf.org/pipermail/comp-neuro/2011-April/002613.html
  • Dear list
    
    A recent paper in PLoS Computational Biology
    
    > The Roots of Bioinformatics in Theoretical Biology
    > Paulien Hogeweg
    > Volume 7(3) March 2011 http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1002021
    
    makes a point in the evolution of the meaning of the field of bioinformatics with the advent of data-driven modeling.
    
    The same seems to have appeared in computational neuroscience. The sense slowly drifted from the original papers (such as Science, Vol. 241, No. 4871, 1988, pp. 1299-1306. by T. J. Sejnowski, C. Koch, P. S. Churchland) which I believed is perfectly captured in the sentence: "The ultimate aim of computational neuroscience is to explain how electrical and chemical signals are used in the brain to represent and process information."  (this does not exclude using computers of course).
    
    It seems to be solely a semantical problem, but this may generate some confusion (realpolitik translation: "and this may hinder the efficiency of your grant proposal"). Recently an (anonymous) colleague told me they called their group "computational AND theoretical neuroscience" (just as these two fields where separated) out of the lack of consensus on the meaning of words and to not exclude anyone. Nowadays, even in the university, there is a continuum of fields combining biology,  mathematics or computer science and all computational neuroscientists reflect this as individuals. so what's the situation in 2011?
    
    I often asked to fellow colleagues this question, "what is computational neuroscience?" and often got one of these answers (I try to be unbiased - please correct me):
    
     [ ] it is a field of neuroscience involving the use of computers (von Neumann machines, Dell boxes, macbooks, ...) to simulate and analyze data obtained from experimental neuroscience and advance our knowledge from this dialogue. Theoretical neuroscience is different in the sense that it proposes mathematical models of how it works.
    
     [ ] it is the field of neuroscience studying how information is represented and processed in neural activity. This involves a dialogue with experimental neuroscience to analyze and propose experiments. It proposes models, that is theories for the relation between function and structure. Theoretical neuroscience is a subset of computational neuroscience that tries to express these models in standard mathematical language.
    
     [ ] it is some field of neuroscience and why would you care to give an exact definition? its frontiers are moving and it has many facets, theoretical neuroscience being just one example. it cares about being less ignorant on relation between function and structure in neuroscience.
    
    If you want to express your opinion, you can so in one click:
    https://spreadsheets.google.com/viewform?formkey=dDc5X2dJRS1zMHRiSndSNERWelBkQlE6MQ
    results :
    https://spreadsheets.google.com/lv?key=0AueMPskll6yrdDc5X2dJRS1zMHRiSndSNERWelBkQlE&hl=fr&f=0&rm=full#gid=0
    
    cheers,
    Laurent

results

  • I have received 36 responses with the following results (see https://spreadsheets.google.com/lv?key=0AueMPskll6yrdDc5X2dJRS1zMHRiSndSNERWelBkQlE&hl=fr&f=0&rm=full#gid=0 ):
    1. [ 5 ] it is a field of neuroscience involving the use of computers (von Neumann machines, Dell boxes, macbooks, ...) to simulate and analyze data obtained from experimental neuroscience and advance our knowledge from this dialogue. Theoretical neuroscience is different in the sense that it proposes mathematical models of how it works.
    2. [ 21 ] it is the field of neuroscience studying how information is represented and processed in neural activity. This involves a dialogue with experimental neuroscience to analyze and propose experiments. It proposes models, that is theories for the relation between function and structure. Theoretical neuroscience is a subset of computational neuroscience that tries to express these models in standard mathematical language.
      • one answer is amended by "is a sub-field of theoretical neuroscience that seeks to understand how information is represented and processed in the nervous system by implementing and testing theories in the form of computer simulations."
      • another answer comments "Not a definition, but a comment. Since Comp. Neurosci. (or whatever it should be called -- some people now call it Neurodynamics) is already such a small field and barely represented on the map, I think it is foolish to further subdivide it. That is why I like the first definition a bit more, such that both are collected under one roof."
    3. [ 6 ] it is some field of neuroscience and why would you care to give an exact definition? its frontiers are moving and it has many facets, theoretical neuroscience being just one example. it cares about being less ignorant on relation between function and structure in neuroscience.
    4. [ 4 ] other free-form answers were given:
      1. with a slight modification to the first definition
        I agree with definition (1) except the "Theoretical neuroscience is a subset of ..." as I might argue that "Computational neuroscience is a subset of theoretical neuroscience".
      2. with a new definition close to the aims of theoretical neuroscience
        it is field of neuroscience that use mathematical models to analyze the data obtained from experimental neuroscience. Therefore, it gives a logical result to it and it can be explained instead to be as a magic box
      3. to a strict "computational" view
        It is the subset of theoretical neuroscience that hypothesises that the brain is a computer. This relates to the first definition to the extent that 'computation' is identified with 'information processing'. Theoretical neuroscience is simply the development of models (in any form, including mathematics or computer simulations) of neural processes. It is possible for a process to be simulated or analyzed using a computer - see the second definition - without claiming the process itself is an example of computation. 'Computational neuroscience' usually implies this stronger claim, though it is now often used more loosely (definition three).
      4. or to the interesting view that these different approaches overlap but correspond to different approaches:
        In my view, theoretical neuroscience is the non-experimenting version of neuroscience, much like theoretical physics is the non-experimenting version of physics.
        
        I would argue that theoretical neuroscience and computational neuroscience are different in their approaches.
        
        Computational neuroscience has a strong focus on simulation. It is the "virtual" extension of electrophysiology. The modeling philosophies of GENESIS and Neuron clearly reflect this. So called "biologically realistic" simulations are the gold standard in computational neuroscience.
        
        Theoretical neuroscience, by contrast, has its focus on mathematical descriptions and properties of nervous structures. Theoretical neuroscience starts, when the experiments, real or simulated, are done. The excellent books of Henry Tuckwell illustrate this. Here, simulation is not the method of choice, but the last resort after all pencils are broken and all paper is used up ;-)
        
        At best, I would say that there is overlap between the two rather than that one comprises the other.
        
        And while we are at definitions. Why not add the re-incarnated term of "Neuroinformatics" to the contest?

analysis

Comments

Comments powered by Disqus