WebML WG Teleconference – 13 March 2025 - 15:00-16:00 UTC

Latest version: https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md


WebML WG Teleconference – 13 March 2025 - 15:00-16:00 UTC

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md#webml-wg-teleconference--13-march-2025---1500-1600-utc>

 Important


This meeting takes place one hour later than usual in the US due to Daylight Saving Time<https://www.timeanddate.com/time/dst/2025a.html>.


See the timezone table ...
San Francisco   Thu, 13 March 2025      08:00 adjusted for DST
Boston  Thu, 13 March 2025      11:00 adjusted for DST
London  Thu, 13 March 2025      15:00
Berlin  Thu, 13 March 2025      16:00
Helsinki        Thu, 13 March 2025      17:00
Shanghai        Thu, 13 March 2025      23:00
Tokyo   Fri, 14 March 2025      00:00
UTC     Thu, 13 March 2025      15:00 UTC


Other locations: https://www.timeanddate.com/worldclock/fixedtime.html?iso=20250313T15


Logistics
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md#logistics>

  *   Chair: Anssi
  *   Scribe: Anssi
  *   IRC: irc://irc.w3.org:6667/#webmachinelearning

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

  *   Zoom joining instructions: https://lists.w3.org/Archives/Member/internal-webmachinelearning/2023Jun/0000.html

  *   Minutes: https://www.w3.org/2025/03/13-webmachinelearning-minutes.html


Agenda
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md#agenda>

🧪 Incubations summary

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md#-incubations-summary>

A summary of the most recent WebML Community Group meeting will be shared to keep the Working Group abreast of incubations.

ℹ️ Operator set Wave 3 and spin-offs
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md#%E2%84%B9%EF%B8%8F-operator-set-wave-3-and-spin-offs>


Call for final PR review. Discuss any opens as needed:

  *   ⛙ webmachinelearning/webnn#805<https://github.com/webmachinelearning/webnn/pull/805>
  *   Closes: ⨀ #93<https://github.com/webmachinelearning/webnn/issues/93>, ⨀ #467<https://github.com/webmachinelearning/webnn/issues/467>, ⨀ #772<https://github.com/webmachinelearning/webnn/issues/772>, ⨀ #767<https://github.com/webmachinelearning/webnn/issues/767>, ⨀ #773<https://github.com/webmachinelearning/webnn/issues/773>, ⨀ #779<https://github.com/webmachinelearning/webnn/issues/779>


🆕 Discuss new improvement suggestions spun off from this PR:

  *   dequantizeLinear emulation improvement by expand augmentation:

     *   ⨀ webmachinelearning/webnn#779 (comment)<https://github.com/webmachinelearning/webnn/issues/779#issuecomment-2689214272>
  *   Gather multiaxis operator proposal to more generically express a gathering operation:

     *   ⨀ webmachinelearning/webnn#767 (comment)<https://github.com/webmachinelearning/webnn/issues/767#issuecomment-2688528116>
     *   ☰ Multigather.md<https://github.com/fdwr/MachineLearningOperators/blob/master/Multigather.md>

ℹ️ Query mechanism for supported devices
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md#%E2%84%B9%EF%B8%8F-query-mechanism-for-supported-devices>


Call for additional use cases. Discuss use case feedback received.

  *   ⨀ webmachinelearning/webnn#815<https://github.com/webmachinelearning/webnn/issues/815>

🏷️ feature request<https://github.com/webmachinelearning/webnn/labels/feature%20request> issues

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md#%EF%B8%8F-feature-request-issues>

  *   Caching mechanism for MLGraph

     *   ⨀ webmachinelearning/webnn#807<https://github.com/webmachinelearning/webnn/issues/807>
     *   ☰ Discuss known implementation issues with model cache within browser sandbox, solutions. Gather further feedback on implementation-defined vs. explicit caching API.
     *   References:
        *   Chromium multi-process arch<https://www.chromium.org/developers/design-documents/multi-process-architecture/>, sandboxing<https://chromium.googlesource.com/chromium/src/+/main/docs/design/sandbox.md>
        *   WebKit multi-process arch<https://docs.webkit.org/Deep%20Dive/Architecture/WebKit2.html>
  *

Remove pool2d MLRoundingType

     *   ⨀ webmachinelearning/webnn#324<https://github.com/webmachinelearning/webnn/issues/324>
     *   ⨀ closes also webmachinelearning/webnn#374<https://github.com/webmachinelearning/webnn/issues/374>
     *   ⛙ webmachinelearning/webnn#770<https://github.com/webmachinelearning/webnn/pull/770>
     *   ☰ Address remaining PR review comments.

🏷️ bug<https://github.com/webmachinelearning/webnn/labels/bug> issues

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md#%EF%B8%8F-bug-issues>

  *   Reject promises for in-progress operations on an MLTensor if it is destroyed
     *   ⨀ follow-up to webmachinelearning/webnn#477<https://github.com/webmachinelearning/webnn/issues/477>
     *   ⛙ webmachinelearning/webnn#799<https://github.com/webmachinelearning/webnn/pull/799>
     *   ☰ Review PR.

🏷️ interop<https://github.com/webmachinelearning/webnn/labels/interop> issues

<https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-03-13-wg-agenda.md#%EF%B8%8F-interop-issues>

  *   Behavior when there are NaNs in argmin/max inputs

     *   ⨀ webmachinelearning/webnn#811<https://github.com/webmachinelearning/webnn/issues/811>
     *   ☰ Discuss how backends handle NaNs in the input and agree on next steps.
  *

(de)quantization behaviors on Core ML

     *   ⨀ webmachinelearning/webnn#822<https://github.com/webmachinelearning/webnn/issues/822>
     *   ☰ Discuss quantize/dequantize Core ML implementation feedback