"Functional Collection Programming with Semi-Ring Dictionaries"

[dmm]

arxiv.org/abs/2103.06376

page 2: "Semi-ring dictionaries realize the well-known connection between relations and tensors" (from "In-Database Learning with Sparse Tensors" 2016-2018 paper)

Comments

[dmm]

Via https://www.meetup.com/sf-types-theorems-and-programming-languages/events/297724195/ (Jan 13)

[dmm]

"In-Database Learning with Sparse Tensors":

https://www.cs.ox.ac.uk/dan.olteanu/papers/nnos-pvldb17.pdf

or

https://ora.ox.ac.uk/objects/uuid:2d852e0d-889d-46fe-890e-b1ac5687c798/download_file?file_format=application%2Fpdf&safe_filename=main.pdf&type_of_work=Conference+item

(the second version got a new first author! wow)

[juan_gandhi]

Чудесненько обобщили. Но изложено варварским языком. Вообще, это должна быть не статья, а книга, в Спрингере. Не 38 страниц, а страниц триста.

[dmm]

So, what I need to figure out (given that this seems to be related to my favorite V-values which are a natural platform for differentiable programming in the style of functional immutable collections) is whether this is GPU-friendly, how good are their optimizations, is there an open-source implementation anywhere, and how tightly is this paper related to my structures.

In this sense, a conversation with GPT-4 does give a good quick preliminary impression:

https://chat.openai.com/share/513a5ee8-6b9c-41e1-afd8-dc4f81c90bb0

* Yes, this is closely related, and they do all kinds of cute things with these structures

* No, they don't seem to optimize for GPUs

* Yes, there is a variety of potentially very useful optimization techniques there

* No, it seems that there is no publicly available implementation

So it makes sense to try to attend that Jan 13 meetup, see what people have to say about this.