If we start from the premise that representations/semantics in the brain emerge through neural computations and that representations implicitly correspond to types then several remarks are in order:
1. Independently of how the brain models physical systems it is worth noting that units of measure correspond to types. // https://varkor.github.io/blog/2018/07/30/types-for-units-of-measure.html
2. While Turing machines are useful for modelling the computational complexity of neural computations, they say nothing about types. // https://cs.stackexchange.com/questions/13158/to-what-does-typing-correspond-in-a-turing-machine?stw=2
3. I have found good work on typed recurrent neural networks( http://proceedings.mlr.press/v48/balduzzi16.pdf) but nothing on the application of type theory to neural information processing i.e. nothing within the scope of computational neuroscience.
Where am I going with this?
If we recognise that the brain can only construct representations of reality and that semantic information falls outside the scope of information theory…we may use type theory to develop better models for the computational structure of spike trains.
If we recognise that the brain can only construct representations of reality and that semantic information falls outside the scope of information theory…we may use type theory to develop better models for the computational structure of spike trains.
@BAPearlmutter, @NoahGuzman14, @aha_momentum, @benlansdell might there be a relevant research paper that I have missed?
I think this might help us figure out how complex semantics emerge from compositional models for neural computation.
I think this might help us figure out how complex semantics emerge from compositional models for neural computation.
A few more remarks:
(1) There is a close relationship between type theory and category theory: https://ncatlab.org/nlab/show/relation+between+type+theory+and+category+theory
(2) Category theory has been used to model natural language semantics: https://www.cs.ox.ac.uk/files/5468/sadrzadeh_kartsaklis.pdf
(3) I am taking the symbol-grounding problem into account.
(1) There is a close relationship between type theory and category theory: https://ncatlab.org/nlab/show/relation+between+type+theory+and+category+theory
(2) Category theory has been used to model natural language semantics: https://www.cs.ox.ac.uk/files/5468/sadrzadeh_kartsaklis.pdf
(3) I am taking the symbol-grounding problem into account.
This discussion may be of interest to @KordingLab
How did this line of reasoning occur to me? Neural data science is about querying neural representations, and I asked myself what exactly do neural representations correspond to.
Science progresses when we build our methods on concepts that are not intellectual diversions.
Science progresses when we build our methods on concepts that are not intellectual diversions.