[PROPOSED] Web Machine Learning Working Group Charter

[anssiko]

This is a work in progress draft charter for the Web Machine Learning Working Group. The charter builds upon the initial charter. Feel free to review the draft charter and raise issues.

The expected timeline is as follows:

• Q4 '22: Charter development
• Q1 '23: W3C Advisory Committee review
• Q2' 23: Charter approved

See also: Diff to the initial charter.

---

Preview | Diff

[anssiko]

I have opened issues to track feedback provided during the kick off call on 6 October 2022.

I will keep this draft PR updated to reflect the WG's consensus that is forming in the issue discussions.

Following this example, feel free to open more issues for feedback that has not yet been recorded. And please provide your comments in the existing open issues to help us prepare our next charter. Thank you for your contributions!

[anssiko]

I pushed an update that addresses the following charter issues:

• WebGPU interoperability #20
• WebRTC coordination #21
• Level of abstraction for neural net operations #22
• Set of ops supported must be more comprehensive #23
• OpenXLA coordination #28

See also: Diff to the initial charter.

[dontcallmedom]

• the Timeline section needs an update
• The expected completion date of WebNN needs an update
• have the prospects of the Model API integration in the charter evolved given the recent prototyping efforts?
• the charter should probably link explicitly to the ethics doc now that it exists

Otherwise, I need to check if the charter template has evolved, but the proposed changes themselves look good to me.

[anssiko]

@dontcallmedom thanks! I've pushed an update to address your review comments.

I also incorporated some updates from the evolved charter template, namely Adopted Draft and Exclusion Draft placeholders. I also observed the Timeline section is no longer included in some recent charters (notably GPU for the Web WG), so I dropped it.

have the prospects of the Model API integration in the charter evolved given the recent prototyping efforts?

The Model Loader API is still considered an incubation and its tentative specification status would allow it to advance should the conditions enumerated in the note be satisfied. I will loop in @jbingham to confirm the latest status of the prototype.

[wchao1115]

Since we already removed WebGL interop from the spec, should we also remove it here from the charter? Preferably noted as out of scope.

[anssiko]

Thanks for the suggestions. I pushed 33e2eb5 that should address your comments. Namely:

I removed the WebGL reference from this BLAS compatibility context and added the following text following this paragraph:

Interoperability between the WebNN and WebGL APIs is out of scope.

With this change, the only other reference to WebGL in addition to the above out of scope statement remains in the Motivation and Background section as follows:

Currently, machine learning inference in the browser uses the WebGL graphics API with limited or no access to platform capabilities beneficial for ML such as CPU parallelism, general-purpose GPU, or dedicated ML hardware accelerators.

I think this is appropriate for context setting and to give a historical perspective. With WebNN API becoming more broadly available, the need to use WebGL API for ML workloads is expected to diminish over time.

[wchao1115]

We excluded training capability from the original charter. Do we want to rediscuss that aspect of the spec for webnn v2?

[anssiko]

Thanks for raising this. I have opened #27 to solicit input on training capability. Please provide your feedback in that issue to feed into a broader WG discussion. We still have time to in-scope this if there is adequate interest.

[anssiko]

I met with @dontcallmedom and we merged the charter 2023-2025 PR. Dom is doing his final pass review before the charter starts its AC review.

[anssiko]

(W3M & horizontal review in process: w3c/strategy#367)