2023-05-11-wg-agenda.md

WebML WG Teleconference – 11 May 2023 - 14:00-15:00 UTC

See the timezone table ...

San Francisco (U.S.A. - California)	Thu, 11 May 2022	07:00	UTC-7 hours
Boston (U.S.A. - Massachusetts)	Thu, 11 May 2022	10:00	UTC-4 hours
London (United Kingdom - England)	Thu, 11 May 2022	15:00	UTC+1 hours
Berlin (Germany)	Thu, 11 May 2022	16:00	UTC+2 hours
Helsinki (Finland)	Thu, 11 May 2022	17:00	UTC+3 hours
Shanghai (China)	Thu, 11 May 2022	22:00	UTC+8 hours
Tokyo (Japan)	Thu, 11 May 2022	23:00	UTC+9 hours
Corresponding UTC (GMT)	Thu, 11 May 2022	14:00 UTC

Other locations: https://www.timeanddate.com/worldclock/fixedtime.html?iso=20230511T14

Logistics

• Chair: Anssi
• Scribe: ? (see howto, 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/2023/05/11-webmachinelearning-minutes.html

Agenda

ℹ️ WebIDL and Infra standard conventions

• The constant() method steps

  • meta issue: webmachinelearning/webnn#210 (comment)
  • PR: webmachinelearning/webnn#365

• Sync async algorithms

  • issue: webmachinelearning/webnn#316
  • PR: webmachinelearning/webnn#329

• The builder.input() method steps

  • meta issue: webmachinelearning/webnn#210 (comment)
  • PR: webmachinelearning/webnn#364

• Review the rest of the standard conventions PRs (all ready for review):

  • meta issue: webmachinelearning/webnn#210 (comment)
  • open PRs: https://github.com/webmachinelearning/webnn/pulls

ℹ️ Enhancements, editorials, questions

• float16 support (discuss prototype findings to inform v2 feature work)

  • issue: webmachinelearning/webnn#373
  • prototype PR: Honry/onnxruntime#1
  • TC39 proposal: http://tc39.es/proposal-float16array/

• Subclass MLGraph based on the context that creates it

  • issue: webmachinelearning/webnn#344

• Error handling of MLNamedArrayBufferViews transfer algorithm

  • issue: webmachinelearning/webnn#351

• Support depth_multiplier > 1 for a depthwise conv2d op

  • issue: webmachinelearning/webnn#353

• Clarify interpolation algorithm for resample2d

  • issue: webmachinelearning/webnn#358
  • related issue: webmachinelearning/webnn#270

ℹ️ Support for transformers

Continue discuss transformers and related requirements and gaps. Test our improved contribution guidelines with these new ops: explore key use cases, sample models, cross-framework support, cross-platform implementability.

Status: The WG decided to start explore support for transformers in WebNN.

Next step: Identify use cases. Contributions welcome via issue 375.

• issue: webmachinelearning/webnn#375
• Guidelines for adding new ops: https://github.com/webmachinelearning/webnn/blob/main/CONTRIBUTING.md#proposing-and-adding-a-new-operation
• Minutes from the kick off: https://www.w3.org/2023/04/27-webmachinelearning-minutes.html#t11