We present two higher order combinators for tensor expressions inspired by relational algebra and show how these naturally lead to a nameless representation of expressions. Next we show how to encode affine constraints into these combinators to make our language more expressive. Finally, we present a polymorphic type system for shapes of tensors.
Bharathi Ramana Joshi
Venkatesh Choppella
Technology designed but not implemented.
Scientific computing, machine learning.
Array languages, Domain Specific Languages.