WebML WG Teleconference – 6 October 2022 - 14:00-15:00 UTC

Latest version: https://github.com/webmachinelearning/meetings/blob/main/telcons/2022-10-06-wg-agenda.md


WebML WG Teleconference – 6 October 2022 - 14:00-15:00 UTC
San Francisco (U.S.A. - California)     Thu, 6 October 2022     07:00   UTC-7 hours
Boston (U.S.A. - Massachusetts) Thu, 6 October 2022     10:00   UTC-4 hours
London (United Kingdom - England)       Thu, 6 October 2022     15:00   UTC+1 hours
Berlin (Germany)        Thu, 6 October 2022     16:00   UTC+2 hours
Helsinki (Finland)      Thu, 6 October 2022     17:00   UTC+3 hours
Shanghai (China)        Thu, 6 October 2022     22:00   UTC+8 hours
Tokyo (Japan)   Thu, 6 October 2022     23:00   UTC+9 hours
Corresponding UTC (GMT) Thu, 6 October 2022     14:00 UTC

Other locations: https://www.timeanddate.com/worldclock/fixedtime.html?iso=20221006T14


<https://github.com/webmachinelearning/meetings/blob/main/telcons/2022-10-06-wg-agenda.md#logistics>Logistics

  *   Chair: Anssi
  *   Scribe: ? (see howto<https://github.com/webmachinelearning/meetings/blob/main/scribe-howto.md>, volunteers welcome!)
  *   IRC: irc://irc.w3.org:6667/#webmachinelearning

  *   IRC web client: https://irc.w3.org/?channels=#webmachinelearning

  *   Joining instructions: https://lists.w3.org/Archives/Member/internal-webmachinelearning/2020Apr/0000.html

  *   Minutes: https://www.w3.org/2022/10/06-webmachinelearning-minutes.html


<https://github.com/webmachinelearning/meetings/blob/main/telcons/2022-10-06-wg-agenda.md#%E2%84%B9%EF%B8%8F-webnn-api-candidate-recommendation-open-issues>ℹ️ WebNN API Candidate Recommendation open issues
Review and discuss the current CR issues, work out a plan to address the issues prior to the expected CR publication.

  *   Current CR issues https://github.com/webmachinelearning/webnn/labels/cr


  *   Web platform tests

     *    Recommended tolerances (Dwayne): webmachinelearning/webnn#265 (comment)<https://github.com/webmachinelearning/webnn/issues/265#issuecomment-1256242643>
     *    Exp op tolerance data (Bruce): webmachinelearning/webnn#288<https://github.com/webmachinelearning/webnn/issues/288>
     *   ➡️ Status: Review recommended tolerances, discuss extensions to webnn-baseline reference impl to support WPT
  *   Add method steps and normative algorithms to operations

     *    issue: webmachinelearning/webnn#210<https://github.com/webmachinelearning/webnn/issues/210>
     *    issue: webmachinelearning/webnn#211<https://github.com/webmachinelearning/webnn/issues/211>
     *   ➡️ Status: Discuss the required algorithm updates, work out a plan with contributors.

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2022-10-06-wg-agenda.md#%E2%84%B9%EF%B8%8F-webml-wg-charter-2023-2025-early-heads-up>ℹ️ WebML WG Charter 2023-2025 early heads-up
WebML WG Charter 2023-2025 brainstorming discussion. Start solicit use cases and model requirements for WebNN "v2". WG rechartering process to formally kick off early 2023. Develop draft WG charter during Q4. New technical proposals to be incubated in the CG prior to the WG adoption.

  *   WG Charter: https://www.w3.org/2021/04/web-machine-learning-charter.html

     *   GH repo: https://github.com/w3c/machine-learning-charter

  *   CG Charter: https://webmachinelearning.github.io/charter/

     *   GH repo: https://github.com/webmachinelearning/charter

  *   WebNN v2 issues: https://github.com/webmachinelearning/webnn/labels/v2


<https://github.com/webmachinelearning/meetings/blob/main/telcons/2022-10-06-wg-agenda.md#%E2%84%B9%EF%B8%8F-prs-in-review>ℹ️ PRs in review

  *   Define the data type of the padding, strides and dilations

     *    issue: webmachinelearning/webnn#269<https://github.com/webmachinelearning/webnn/issues/269>
     *    PR: webmachinelearning/webnn#294<https://github.com/webmachinelearning/webnn/pull/294>
  *   Reshape doesn't support full range of operand dimension size

     *    issue: webmachinelearning/webnn#289<https://github.com/webmachinelearning/webnn/issues/289>
     *    PR: webmachinelearning/webnn#291<https://github.com/webmachinelearning/webnn/pull/291>

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2022-10-06-wg-agenda.md#%E2%84%B9%EF%B8%8F-webnn-webgpu-interop>ℹ️ WebNN-WebGPU interop
Review WebGPU interop mechanism and its normative WebGPU dependencies to assess whether WebGPU interop is a feasible CR target or a v2 feature.

WebNN API interface to record the ML workload onto a WebGPU-compatible GPUCommandBuffer:

  *   MLCommandEncoder https://www.w3.org/TR/webnn/#mlcommandencoder


Normative WebGPU API dependencies:

  *   GPUBuffer https://gpuweb.github.io/gpuweb/#buffer-interface

  *   GPUCommandBuffer https://gpuweb.github.io/gpuweb/#command-buffers

  *   GPUCommandBufferDescriptor https://gpuweb.github.io/gpuweb/#dictdef-gpucommandbufferdescriptor

  *   GPUDevice https://gpuweb.github.io/gpuweb/#gpu-device

  *   GPUQueue https://gpuweb.github.io/gpuweb/#gpuqueue

  *   GPUTexture https://gpuweb.github.io/gpuweb/#gputexture


See also:

  *    22 Sep 2022 discussion https://www.w3.org/2022/09/22-webmachinelearning-minutes.html#t06

  *    related issue: webmachinelearning/webnn#264<https://github.com/webmachinelearning/webnn/issues/264>

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2022-10-06-wg-agenda.md#%E2%84%B9%EF%B8%8F-proposed-new-features>ℹ️ Proposed new features

  *   Reshape doesn't support full range of operand dimension size

     *    issue: webmachinelearning/webnn#289<https://github.com/webmachinelearning/webnn/issues/289>
  *   Need for type casting?

     *    issue: webmachinelearning/webnn#284<https://github.com/webmachinelearning/webnn/issues/284>
  *   Softmax should only support input of floating-point types

     *    issue: webmachinelearning/webnn#283<https://github.com/webmachinelearning/webnn/issues/283>

Revisit proposed new features discussed earlier:

  *   Should MLBufferView + MLOperandDescriptor be strongly typed

     *    proposed CR issue: webmachinelearning/webnn#275<https://github.com/webmachinelearning/webnn/issues/275>
  *   Support coordinate transformation modes for Resample2d

     *    proposed CR issue: webmachinelearning/webnn#270<https://github.com/webmachinelearning/webnn/issues/270>
  *   l2Pool2d algorithm clarifications

     *    issue: webmachinelearning/webnn#278<https://github.com/webmachinelearning/webnn/issues/278>