SP
Srijan Paul
@injuly
Formalizing ML for Functional Languages
Submitted Jun 6, 2026
At first glance, the typical ML training loop might seem infeasible to formalize in pure languages.
Most operations in an optimized CNN/Transformer model are destructive, and the core primitive that GPUs offer (fused multiply-add; or a <- a*b+c) is by definition destructive as well.
We’ll find out how a standard ML training pipeline can be represented in pure languages—and learn the underpinnings for frameworks like Hasktorch in the process.
Comments
Hosted by
The homepage for the Indian Functional Programming community
{{ gettext('Login to leave a comment') }}
{{ gettext('Post a comment…') }}{{ errorMsg }}
{{ gettext('No comments posted yet') }}