@sigfpe What do you think of the following grand scheme:
- Define Haskell-style typeclasses for mathematical structures - rings, fields, vector spaces; instance them on float, etc.
- Now instance them on an alegebraic datatype representing symbolic calculations.
-
-
You should look into the more general idea of metaprogramming in Haskell, I think you should find some ways to execute your idea there.
-
I think it's interesting to try this stuff without metaprogramming. Eg. you can go a long way with automatic differentiation without having to do any metaprogramming even though naively it seems calculus requires something meta.
- 1 more reply
New conversation -
-
-
Not a language guy but OSL https://github.com/imageworks/OpenShadingLanguage … does auto differentiation symbolically. Doesn’t need quads or grids for finite differences. Used for texture filtering. Syntactically very C like and built on LLVM.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
There are deeplearning frameworks that do things very similar to what you describe https://github.com/denizyuret/Knet.jl/blob/master/README.md#AutoGrad-1 …
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I have no clue what you just tweeted, but it was beautiful man.... beautiful.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
fwiw there's a small literature around backprop, neural nets in GA / Clifford algebras --- mostly from a while before all the recent excitement
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.