WebML WG Teleconference – 7 May 2026 - 15:00-16:00 UTC

Latest version: https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md


WebML WG Teleconference – 7 May 2026 - 15:00-16:00 UTC
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md#webml-wg-teleconference--7-may-2026---1500-1600-utc>
See the timezone table ...
San Francisco   Thu, 7 May 2026 08:00
Boston  Thu, 7 May 2026 11:00
London  Thu, 7 May 2026 16:00
Berlin  Thu, 7 May 2026 17:00
Helsinki        Thu, 7 May 2026 18:00
Shanghai        Thu, 7 May 2026 23:00
Tokyo   Fri, 8 May 2026 00:00
UTC     Thu, 7 May 2026 15:00 UTC

Other locations: https://www.timeanddate.com/worldclock/fixedtime.html?iso=20260507T15


🤝 Logistics
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-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: internal-webmachinelearning<https://lists.w3.org/Archives/Member/internal-webmachinelearning/2023Jun/0000.html>
  *   Minutes: https://www.w3.org/2026/05/07-webmachinelearning-minutes.html


🧠 Web Neural Network API
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md#-web-neural-network-api>

Tip

Specification<https://www.w3.org/TR/webnn/> | Explainer<https://github.com/webmachinelearning/webnn/blob/main/explainer.md> | Implementation Report<https://wpt.fyi/results/webnn> | Implementation Status<https://webmachinelearning.github.io/webnn-status/>

Proposed new low-precision floating-point data types
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md#proposed-new-low-precision-floating-point-data-types>

Discuss the proposed new data types (bfloat16, fp8, nvfp4), use cases, cross-platform implementability & future-proof-ness.

  *   ⨀ webmachinelearning/webnn#930<https://github.com/webmachinelearning/webnn/issues/930>

Formulaic mitigation for integer overflows
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md#formulaic-mitigation-for-integer-overflows>

Review the latest proposal, agree on the next steps.

  *   ⨀ webmachinelearning/webnn#928<https://github.com/webmachinelearning/webnn/issues/928>

Upper bounds on concat inputs and split outputs
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md#upper-bounds-on-concat-inputs-and-split-outputs>

Discuss and agree on future-proof upper bounds informed by real-world models.

  *   ⨀ webmachinelearning/webnn#931<https://github.com/webmachinelearning/webnn/issues/931>

Bounded dynamic dimension
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md#bounded-dynamic-dimension>

Review any new feedback on the proposed new ops: mod, shape, range, dynamic[Reshape | Expand | Slice | Pad | Split | Resample2d]. Discuss the next steps, dispatch-time shape validation.

  *   ⨀ webmachinelearning/webnn#883 (comment)<https://github.com/webmachinelearning/webnn/issues/883#issuecomment-4302848951>

Effective MLComputePolicy exposure
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md#effective-mlcomputepolicy-exposure>

Discuss the effective policy exposure after compilation. Assumption: supersedes the earlier device-centric graph.devices proposal.

  *   Related: ⛙ webmachinelearning/webnn#923<https://github.com/webmachinelearning/webnn/pull/923>

MLComputePolicy enhancements: "low-latency" and "precision"
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md#mlcomputepolicy-enhancements-low-latency-and-precision>

Discuss the proposed enhancements: "low-latency" for audio processing use case; "precision" to avoid chopping off low bits.

  *   ⨀ N/A

PR queue cleanup
<https://github.com/webmachinelearning/meetings/blob/main/telcons/2026-05-07-wg-agenda.md#pr-queue-cleanup>

Approve and merge PRs ahead the meeting, discuss any blockers only.

  *   ⛙ webmachinelearning/webnn#910<https://github.com/webmachinelearning/webnn/pull/910>