14:52:27 <RRSAgent> RRSAgent has joined #webmachinelearning
14:52:31 <RRSAgent> logging to https://www.w3.org/2025/10/09-webmachinelearning-irc
14:52:31 <Zakim> RRSAgent, make logs Public
14:52:32 <Zakim> please title this meeting ("meeting: ..."), anssik
14:52:32 <anssik> Meeting: WebML WG Teleconference – 9 October 2025
14:52:41 <anssik> Chair: Anssi
14:52:47 <anssik> Agenda: https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-10-09-wg-agenda.md
14:53:08 <anssik> Scribe: Anssi
14:53:13 <anssik> scribeNick: anssik
14:53:23 <anssik> Present+ Anssi_Kostiainen
14:59:23 <anssik> Present+ Fabio_Bernardon
14:59:46 <anssik> Present+ Markus_Tavenrath
15:00:21 <anssik> Present+ Tarek_Ziade
15:01:16 <anssik> Present+ Zoltan_Kis
15:01:47 <DwayneR> DwayneR has joined #webmachinelearning
15:02:23 <anssik> Present+ Dwayne_Robinson
15:02:34 <anssik> Present+ Ningxin_Hu
15:02:41 <anssik> Present+ Rafael_Cintron
15:03:45 <anssik> Present+ Reilly_Grant
15:04:39 <ningxin> ningxin has joined #webmachinelearning
15:04:39 <anssik> Present+ Thomas_Steiner
15:04:41 <anssik> RRSAgent, draft minutes
15:04:42 <RRSAgent> I have made the request to generate https://www.w3.org/2025/10/09-webmachinelearning-minutes.html anssik
15:05:08 <Fabio> Fabio has joined #webmachinelearning
15:05:14 <mtavenrath> mtavenrath has joined #webmachinelearning
15:05:45 <anssik> Anssi: we'll start by welcoming our latest new participant:
15:05:48 <mtavenrath> mtavenrath has joined #webmachinelearning
15:05:48 <Fabio> Fabio has joined #webmachinelearning
15:05:48 <ningxin> ningxin has joined #webmachinelearning
15:05:48 <DwayneR> DwayneR has joined #webmachinelearning
15:05:48 <zolkis> zolkis has joined #webmachinelearning
15:05:48 <reillyg> reillyg has joined #webmachinelearning
15:05:48 <tarek> tarek has joined #webmachinelearning
15:05:57 <anssik> ... please welcome to the WebML WG:
15:06:05 <anssik> ... Mark Foltz from Google
15:06:42 <anssik> ... Umar Iqbal from Washington University as an Invited Expert
15:07:04 <tarek> tarek has joined #webmachinelearning
15:07:04 <mtavenrath> mtavenrath has joined #webmachinelearning
15:07:04 <Fabio> Fabio has joined #webmachinelearning
15:07:04 <ningxin> ningxin has joined #webmachinelearning
15:07:04 <DwayneR> DwayneR has joined #webmachinelearning
15:07:04 <zolkis> zolkis has joined #webmachinelearning
15:07:04 <reillyg> reillyg has joined #webmachinelearning
15:07:19 <anssik> ... Aram Zucker-Scharff, Davis Shaver, and Stephen Erickson from The Washington Post
15:08:01 <anssik> ... welcome to all new participants, I look forward to working with you!
15:08:11 <anssik> Topic: Incubations
15:08:31 <anssik> Anssi: a debrief on the recent WebML Community Group developments
15:08:38 <anssik> -> WebML CG Teleconference – 2 October 2025 https://github.com/webmachinelearning/meetings/blob/main/telcons/2025-10-02-cg-agenda.md
15:08:54 <anssik> gb, this is webmachinelearning/webmcp
15:08:54 <gb> anssik, OK.
15:09:03 <anssik> Anssi: we had another WebMCP API brainstorming session and made important resolutions:
15:09:11 <anssik> ... - resolved to make the tools be part of the discovery mechanism
15:09:24 <anssik> ... - resolved to look into higher-level hooks to connect WebMCP with external agents for listing tools
15:09:39 <anssik> ... - resolved that tool execution should be able to start/stop yielding to the user throughout its lifecycle, in context of elicitation
15:09:49 <anssik> ... - resolved navigator.modelContext is the "root" object name
15:09:56 <anssik> q?
15:10:08 <anssik> Topic: F2F Agenda brainstorming
15:10:14 <anssik> gb, this is webmachinelearning/meetings
15:10:14 <gb> anssik, OK.
15:10:28 <anssik> Anssi: F2F Agenda issue #35
15:10:29 <gb> https://github.com/webmachinelearning/meetings/issues/35 -> Issue 35 WebML WG/CG F2F Agenda - TPAC 2025 (Kobe, Japan) (by anssiko)
15:10:42 <anssik> Anssi: I want to discuss, review and update draft WebML WG/CG F2F Agenda based on your feedback
15:10:59 <anssik> ... now that we're getting closer to the F2F it makes sense to lower the level of abstraction and look at specific issues of interest
15:11:04 <RafaelCintron> RafaelCintron has joined #webmachinelearning
15:11:05 <anssik> ... note on logistics:
15:11:10 <anssik> ... registration open until 3 November
15:11:52 <Ehsan> Ehsan has joined #webmachinelearning
15:12:22 <anssik> ... meeting dates are 10-11 November 2025 (start on 9/10 for remotes in Pacific timezone!)
15:12:33 <anssik> ... please export invites as .ics from:
15:12:38 <anssik> -> 10 November 2025 https://www.w3.org/events/meetings/f63193ec-259b-4ab8-ad65-a5a6e0adf556/
15:12:45 <anssik> -> 11 November 2025 https://www.w3.org/events/meetings/091a2581-034b-4afa-8ddc-91155bd4d710/
15:13:15 <anssik> Anssi: the first day 10 Nov dedicated for WG / WebNN, second day 11 Nov for WebMCP, Built-in AI APIs
15:13:27 <anssik> ... we can still do adjustments based on feedback
15:14:52 <anssik> Present+ Ehsan_Toreini
15:15:53 <anssik> -> https://github.com/w3c/tpac2025-breakouts/issues
15:16:39 <anssik> -> https://www.w3.org/2025/11/TPAC/#schedule
15:17:13 <anssik> Anssi: I observe good participation, both familiar names and new faces
15:17:20 <anssik> ... currently 42 in-person participants including observers, excluding remote participants
15:17:33 <anssik> ... to set the expectations for the F2F meeting:
15:17:46 <anssik> ... F2F is an opportunity to get to know people, including folks outside the group and the wider community
15:17:52 <anssik> ... humans usually work better together when they know each other
15:18:13 <anssik> ... we will not do low-level specification PR reviews on a big screen at the F2F, an async GH-driven work mode is better for that
15:18:33 <anssik> ... rather we try to make resolutions and seek consensus on important issues, chart the path forward, and eat Japanese food in a great company
15:19:16 <anssik> Anssi: F2F Agenda issue #35
15:19:17 <gb> https://github.com/webmachinelearning/meetings/issues/35 -> Issue 35 WebML WG/CG F2F Agenda - TPAC 2025 (Kobe, Japan) (by anssiko)
15:20:15 <anssik> Anssi: I put up a draft to solicit feedback via comments for both TBA topics and timing to avoid scheduling conflicts as best as we can
15:20:49 <anssik> ... we've made good progress in closing down open issues for the WebNN API, stabilizing the spec
15:20:58 <anssik> ... recently a lot of energy has been put on broadening the implementation experience
15:21:12 <anssik> ... this is busy time as we race to meet release branch milestones and pass quality gates
15:21:34 <anssik> ... after this major push we get the API in the hands of early adopters to help us iron out remaining kinks
15:21:51 <anssik> ... the timing of this coincides with our TPAC meeting, so appreciate your contributions during this busy time
15:22:32 <anssik> q?
15:22:39 <reillyg> q+
15:22:42 <anssik> ack reillyg
15:23:20 <mtavenrath> q+
15:23:29 <anssik> Reilly: I think implementation-wise are waiting on Windows ML backend, a big missing piece, getting very close
15:23:45 <anssik> ... another thing is interop of backends, take a good look at our wpt coverage and gaps
15:24:20 <anssik> Anssi: who would be best to lead the wpt discussion?
15:24:57 <anssik> Reilly: the group could do a triage pass over wpt result and that could help answer whether there are any implementation differences that warrant spec changes
15:25:25 <anssik> q?
15:25:38 <anssik> Ningxin: I wil check our team working on wpt tests
15:25:39 <anssik> q?
15:25:41 <anssik> ack mtavenrath
15:26:46 <RafaelCintron> q+
15:26:47 <anssik> Markus: regarding NVIDIA, our backend is enabled, operator tests are fine, a few have accuracy problems due to reduced format we use internally, DML provider and choosing the backend had some issues
15:27:48 <anssik> ack RafaelCintron
15:28:58 <anssik> Rafael: I propose we discuss the system setup separately
15:28:59 <anssik> q?
15:29:24 <anssik> Anssi: 10 November 2025 is the Working Group F2F with a WebNN API focus, here's the top-level view:
15:30:00 <ningxin> If it's a Chromium implementation issue, feel free to open an issue at https://issues.chromium.org/issues/new?component=1456206&template=0
15:30:32 <anssik> Anssi: - Orientation
15:30:32 <anssik> ... group's charter framing, triage pass over WebNN issues, as a group exercise
15:31:05 <reillyg> q+
15:31:08 <anssik> ack reillyg
15:31:54 <anssik> Reilly: I think we did a pass over issues in the beginning, running through issues live can be productive in real-time space
15:33:48 <anssik> ... maybe the editors can put together a report summarizing them
15:34:31 <anssik> ... - New features
15:34:31 <anssik> ... 2-4 issues, can include supporting presentations
15:36:19 <anssik> q?
15:36:57 <anssik> ... - Customer feedback and collaborations
15:36:57 <anssik> ... please bring any feedback from frameworks, end-users, ISVs
15:37:39 <anssik> q?
15:40:22 <anssik> ... - Interop and technical cross-group coordination
15:40:22 <anssik> ... interop is the cornerstone of the web platform, wpt topics and any coordination with other W3C groups goes here
15:40:45 <anssik> ... - Implementation plans and trials
15:40:45 <anssik> ... we discuss upcoming trials, learnings from browsers, backends and frameworks that implement and integrate with the WebNN API
15:40:56 <anssik> ... - Horizontals
15:40:56 <anssik> ... we get to know experts behind horizontal groups: ethics, sustainability, privacy, security, all areas where we've recently recruited more participants to join us
15:41:21 <anssik> ... - Dinner
15:41:21 <anssik> ... we eat Japanese food in a great company!
15:41:33 <anssik> Anssi: feedback welcome via GH comments, on these calls, via email
15:41:43 <anssik> q?
15:41:55 <anssik> Topic: New features and operator specific issues
15:42:12 <anssik> Subtopic: Drop support of 8-bit integers input for CumulativeSum
15:42:17 <anssik> Anssi: issue #892
15:42:17 <gb> Issue 892 not found
15:42:21 <anssik> ... Ningxin proposes to drop support of 8-bit integers input for CumulativeSum due to lack of backend support
15:42:30 <anssik> gb, this is webmachinelearning/webnn
15:42:30 <gb> anssik, OK.
15:42:35 <anssik> #892
15:42:36 <gb> https://github.com/webmachinelearning/webnn/issues/892 -> Issue 892 Drop support of 8-bit integers input for CumulativeSum (by huningxin) [operator specific]
15:42:46 <anssik> ... issue notes 8-bit integer input for cumulativeSum is not supported by any of the Chromium backends: Core ML, DirectML, ONNX, TFLite
15:42:49 <anssik> ... for symmetry, reduceSum also doesn't support 8-bit integer input
15:42:59 <anssik> ... I think we all agree to drop this, I see Phillis +1
15:43:06 <anssik> q?
15:43:19 <anssik> Reilly: SGTM
15:43:50 <anssik> Subtopic: Flexible input sizes
15:43:55 <anssik> Anssi: issue #883
15:43:56 <gb> https://github.com/webmachinelearning/webnn/issues/883 -> Issue 883 Support flexible input sizes (by huningxin) [feature request] [operator specific]
15:44:02 <anssik> ... I put this on the agenda as a reminder to check with Guenther for ORT Web and WebNN EP perspective for the feature
15:44:06 <anssik> ... I guess we're still awaiting Guenther's feedback?
15:44:48 <anssik> Rafael: I haven't heard feedback yet, he personally thinks this is important
15:45:39 <anssik> Anssi: do we block on Guenther or can we do some investigation ourselves in the interim to further this?
15:46:25 <reillyg> q+
15:46:38 <anssik> ack reillyg
15:47:21 <anssik> Reilly: I think the question is how is this getting implemented by backends, what is the role of WebNN in this decision, the framework could build multiple graphs
15:47:56 <anssik> ... I suspect that has all sorts of performance bottlenecks, I want to understand what is the form that various backends would prefer, something to abstract over
15:48:58 <anssik> Dwayne: would need to familiarize myself with TFLite and Core ML, as of importance of this feature, interested in prototyping to see the possibility
15:49:40 <anssik> Reilly: I haven't looked at this yet in TFLite and Core ML, DML EP can execute models with dynamic shapes, Joshua/HF used WebGPU EP and it has some support?
15:49:46 <anssik> Dwayne: right
15:49:47 <anssik> q?
15:50:20 <anssik> Markus: it can be expensive for ORT to have multiple graphs
15:51:16 <anssik> Reilly: my intuition also, to require multiple graphs, pushing that deep into the stack the particular implementation can avoid recreating them, should figure out resource sharing, to push this down to component interact with hardware
15:51:27 <anssik> Markus: in TensorRT we have dynamic shapes and it is handled by our EP
15:51:33 <anssik> q?
15:51:59 <anssik> Rafael: how ORT talks with EPs is an implementation detail
15:52:15 <anssik> Markus: I recall some frameworks allow defining max size with flexible input sizes
15:52:16 <anssik> q?
15:53:00 <anssik> q?
15:53:28 <anssik> Topic: Core operator set
15:53:33 <anssik> Anssi: issue #573
15:53:33 <gb> https://github.com/webmachinelearning/webnn/issues/573 -> Issue 573 Core operator set (by philloooo) [question] [opset]
15:53:37 <anssik> -> Machine Learning Operator Mapping - All Raw Operators https://onedrive.live.com/edit?id=EE82F5C6F06C7371!345450&resid=EE82F5C6F06C7371!345450&ithint=file%2Cxlsx&authkey=!AK8f-RDTleqlLXE&wdo=2&cid=ee82f5c6f06c7371
15:53:47 <anssik> Anssi: we had a good discussion at our prior meeting we stand with this core operator set
15:53:51 <anssik> -> prior meeting minutes https://www.w3.org/2025/09/25-webmachinelearning-minutes.html#7aff
15:54:00 <anssik> Anssi: Fabio wanted to get back to the group after talking with the NVIDIA team
15:54:39 <anssik> Fabio: we're collecting all the ops that'd benefit from being in the set, one class is various attentions
15:54:51 <anssik> ... also gathers, MoE, TopK
15:55:08 <anssik> ... looking for other ops that'd benefit from not being composed
15:55:20 <anssik> q?
15:55:24 <reillyg> q+
15:55:27 <anssik> ack reillyg
15:56:20 <anssik> Reilly: I'm curious about MoE and attentions, my concern with these high-level ops that are tied to particular model architectures, while they give performance boost, not necessarily long-lived
15:56:56 <anssik> ... found out this by looking at e.g. LSTM but actual implementation details matter, and there were compatibility issues between implementations
15:57:01 <anssik> Fabio: I will look into this
15:57:42 <anssik> ... do we have any feedback?
15:57:48 <anssik> Topic: Privacy and Security
15:57:52 <anssik> Anssi: proposed changes to privacy considerations in PR #890
15:57:53 <gb> https://github.com/webmachinelearning/webnn/pull/890 -> Pull Request 890 Revise privacy considerations (by anssiko)
15:58:01 <anssik> ... this PR suggests more changes than the minimal one-liner proposed by Reilly here:
15:58:05 <anssik> -> "No information from the underlying platform is exposed directly." needs to be revised https://github.com/webmachinelearning/webnn/issues/886#issuecomment-3281558389
15:58:06 <gb> https://github.com/webmachinelearning/webnn/issues/886 -> Issue 886 Revise privacy considerations (by anssiko) [privacy-tracker]
15:58:17 <anssik> Anssi: if the group would prefer a minimal change, I will update the PR accordingly
15:58:39 <anssik> q?
15:59:18 <anssik> Anssi: finally, the security review was completed with positive feedback: "well-written in a narrative form"
15:59:23 <anssik> -> https://github.com/w3c/security-request/issues/85
15:59:27 <gb> https://github.com/w3c/security-request/issues/85 -> CLOSED Issue 85 Web Neural Network API 2025-03-20 > 2025-06-20 (by anssiko) [REVIEW REQUESTED] [pending] [CR]
15:59:38 <anssik> ... that means once the privacy revising issue #886 is addressed we've completed the latest wide review round!
15:59:38 <gb> https://github.com/webmachinelearning/webnn/issues/886 -> Issue 886 Revise privacy considerations (by anssiko) [privacy-tracker]
15:59:44 <anssik> -> wide review tracker https://github.com/webmachinelearning/webnn/issues/239#issuecomment-2740740891
15:59:45 <gb> https://github.com/webmachinelearning/webnn/issues/239 -> Issue 239 Wide review tracker (by anssiko) [process]
15:59:48 <anssik> Topic: Query supported devices
15:59:54 <anssik> Subtopic: Before graph compilation
16:00:02 <anssik> Anssi: spec PR #895 and explainer PR #884
16:00:03 <gb> https://github.com/webmachinelearning/webnn/pull/884 -> Pull Request 884 Update explainer with new proposal for simple accelerator mapping (by zolkis)
16:00:03 <gb> https://github.com/webmachinelearning/webnn/pull/895 -> Pull Request 895 Add a simple accelerator selection mechanism. (by zolkis)
16:00:12 <anssik> ... thanks Zoltan for submitting these two PRs, ready for review now
16:00:17 <anssik> ... the spec PR suggests a simplified boolean-returning MLContext.accelerated and MLContext.cpuFallbackActive API
16:00:21 <anssik> ... proposed IDL change:
16:00:28 <anssik> ```
16:00:28 <anssik> interface MLContext {
16:00:28 <anssik>   undefined destroy();
16:00:28 <anssik> + readonly attribute boolean accelerated;
16:00:28 <anssik> + readonly attribute boolean cpuFallbackActive;
16:00:28 <anssik>   readonly attribute Promise<MLContextLostInfo> lost;
16:00:29 <anssik> };
16:00:29 <anssik> ```
16:00:39 <anssik> Anssi: this minimal API change is per our discussion
16:00:55 <anssik> ... I'd like to get review from implementers, and if no concerns merge this PR
16:01:02 <anssik> q?
16:01:36 <RafaelCintron> q+
16:01:39 <anssik> Zoltan: just mentioning I haven't identified steps that handle power options, could do that separately
16:01:40 <anssik> ack RafaelCintron
16:02:11 <anssik> Rafael: I have one questions about MLContext, what in practice is the use case when accelerated and cpuFallbackActive both are false
16:02:51 <anssik> Zoltan: currently the steps that I added do not allow this case
16:03:44 <anssik> Rafael: what if the backend accelerates some of the ops?
16:03:57 <anssik> ... when accelerated and cpuFallbackActive are both true?
16:04:24 <anssik> Zoltan: accelerated refers to massively parallel acceleration
16:04:51 <anssik> Rafael: I guess there could be a case, if there's a CPU backend doing SIMD it could be considered accelerated with cpuFallbackActive
16:05:07 <anssik> massively parallel acceleration expects GPU and NPU
16:05:39 <anssik> Rafael: why do we need two booleans, if when accelerated is true there's no cpu fallback?
16:05:56 <anssik> ... want to understand the use cases when both are true or both are false
16:06:17 <anssik> Zoltan: both are false is redundant, we should specify to avoid this combination
16:08:11 <anssik> Zoltan: good input, but also need to include power options in the picture, based on its setting we could select NPU or GPU
16:08:18 <anssik> ... comments via PR welcome
16:08:20 <anssik> q?
16:09:15 <anssik> RRSAgent, draft minutes
16:09:17 <RRSAgent> I have made the request to generate https://www.w3.org/2025/10/09-webmachinelearning-minutes.html anssik
16:09:44 <anssik> s/latest new participant/latest new participants
16:12:24 <anssik> s/be best/be the best
16:13:39 <anssik> s/over issues/over issues at previous TPAC
16:14:00 <anssik> s/- New features/Anssi: - New features
16:17:22 <anssik> s/interact/that interact
16:18:00 <anssik> s/we stand with this/with regard to
16:19:28 <anssik> s/but actual/where actual
16:19:45 <anssik> s/do we have/Anssi: do we have
16:20:55 <anssik> s/one questions/one question
16:22:15 <anssik> RRSAgent, draft minutes
16:22:16 <RRSAgent> I have made the request to generate https://www.w3.org/2025/10/09-webmachinelearning-minutes.html anssik
18:19:02 <Zakim> Zakim has left #webmachinelearning