13:56:41 RRSAgent has joined #webmachinelearning 13:56:41 logging to https://www.w3.org/2021/09/30-webmachinelearning-irc 13:56:44 RRSAgent, make logs Public 13:56:44 please title this meeting ("meeting: ..."), anssik 13:56:50 Meeting: WebML WG Teleconference – 30 Sep 2021 13:57:06 Chair: Anssi 13:57:13 Agenda: https://github.com/webmachinelearning/meetings/blob/master/telcons/2021-09-30-agenda.md 13:57:24 Scribe: Anssi 13:57:28 scribeNick: anssik 13:57:36 Present+ Anssi_Kostiainen 13:57:47 RRSAgent, draft minutes 13:57:47 I have made the request to generate https://www.w3.org/2021/09/30-webmachinelearning-minutes.html anssik 13:59:06 ningxin_hu has joined #webmachinelearning 13:59:13 Present+ Ningxin_Hu 13:59:57 Present+ Rafael_Cintron 14:00:04 Geun-Hyung has joined #webmachinelearning 14:00:10 Present+ Geun-Hyung_Kim 14:00:12 present+ 14:01:23 Present+ 14:02:15 Present+ Ganesan_Ramalingam 14:02:33 Present+ Chai_Chaoweeraprasit 14:03:10 Rama has joined #webmachinelearning 14:03:12 chai has joined #webmachinelearning 14:03:40 scribe+ 14:04:05 Topic: WebNN API recent new feature requests 14:04:24 Subtopic: Request input layout and resize only on height and width for Resample 14:04:36 -> https://github.com/webmachinelearning/webnn/issues/200 issue #200 14:04:41 -> https://github.com/webmachinelearning/webnn/pull/205 PR #205 14:04:49 Anssi: we discussed this in our last meeting 14:04:53 ... the PR was updated accordingly 14:04:55 Jonathan has joined #webmachinelearning 14:04:57 ... it has been reviewed 14:05:16 ... I think we should be ready to merge this, after confirmation we're happy with the change, incl from Rama 14:05:22 RafaelCintron has joined #webmachinelearning 14:05:35 Rama: fine with me 14:05:38 proposed RESOLUTION: Merge Change resample to resample2d PR #205 14:06:02 RESOLUTION: Merge Change resample to resample2d PR #205 14:06:09 Subtopic: Support for configuring rounding type in pooling operations 14:06:19 -> https://github.com/webmachinelearning/webnn/issues/198 issue #198 14:06:26 -> https://github.com/webmachinelearning/webnn/pull/208 PR 208 14:06:47 Anssi: updated too following our last meeting 14:06:50 -> https://github.com/webmachinelearning/webnn/pull/208/commits/17d1fc583f03cf713e893d204ef35de04115249f commit "Support outputSizes of pooling operations" 14:07:10 Anssi: Chai, wanted to get your input on this? 14:07:34 Chai: I thought I had provided feedback on this, but maybe not directly on the PR 14:07:43 Anssi: wanted to check you're OK with the latest commit in it 14:07:58 ... wanted an explicit signal 14:08:05 Chai: I'll follow up on the PR 14:08:41 Subtopic: reduceLogSum, reduceLogSumExp, reduceSumSquare are not supported on OV/NNAPI 14:08:58 -> https://github.com/webmachinelearning/webnn/issues/209 Issue #209 14:09:11 Anssi: do we still want to include the informative text on emulation in the spec for consistency sake 14:10:01 https://webmachinelearning.github.io/webnn/#api-mlgraphbuilder-gemm 14:10:12 the pool2d PR looks good to me. I've approved the PR. 14:10:47 chai: where would the informative text be added? 14:10:58 anssi: within the context of ops that can be used to implement this 14:11:04 ... reduction operations section 14:11:39 https://webmachinelearning.github.io/webnn/#api-mlgraphbuilder-reduce 14:11:46 chai: for these 3 ops, to add this pseudo-code? 14:11:52 ... yes, we can do that 14:12:00 anssi: I think in the end it will help us 14:12:15 +1 14:12:37 Subtopic: Support iohw filterLayout for conv2d op 14:12:42 -> https://github.com/webmachinelearning/webnn/issues/212 issue 212 14:12:47 [need to step away for a minute] 14:12:51 anssi: a new request to be triaged 14:13:12 Anssi: Ninxgin submitted a PR for this 14:13:24 -> https://github.com/webmachinelearning/webnn/pull/214 PR 214 14:13:35 anssik: a new request informed by fast style transfer ONNX model 14:13:50 ... ConvTranspose op needs "iohw" filter layout: 14:13:54 ... [input_channels/groups, output_channels, height, width] 14:13:57 [back now] 14:14:17 -> https://webmachinelearning.github.io/webnn/#usecase-style-transfer Style Transfer use case 14:14:38 Anssi: in the spec use cases we quote two other style transfer models 14:14:48 ... are there concerns in expanding the layout options as proposed? 14:15:14 chai: style transfer can work with any type of layout 14:15:26 ... the proposal makes it easier to implement *a* version of style transfer 14:15:52 Anssi: Ningxin, have you talked with Wanming about the specifics of the proposal 14:15:58 Ningxin: I did 14:16:11 ... he's been using the ONNX format of the said model 14:16:20 ... and has been trying to map that to WebNN ops 14:16:28 ... as part of development of WebNN Samples 14:17:12 ... the ONNX ConvTranspose uses a filterLayout that isn't supported by WebNN Conv2D 14:17:29 ... he proposes to support that with a WebNN Conv2D specifically for the ConvTranspose usage 14:17:40 ... and to have better support for the ONNX ConvTranpose op 14:17:47 RRSAgent, draft minutes 14:17:47 I have made the request to generate https://www.w3.org/2021/09/30-webmachinelearning-minutes.html anssik 14:18:20 ... The PR itself is fairly straightforward 14:18:32 ... the underlying implementation can support that with not much effort 14:18:48 ... it can be transposed to the native transpose operation 14:18:52 q? 14:18:56 ... would like feedback from the users-side of the API 14:19:21 Rama: this seems fine, we'll take a closer look 14:19:36 +q 14:19:56 ack chai 14:20:22 chai: thinking to our decision on convolution transpose a while back, this topic was raised 14:20:35 ... I would like to see what our resolution was 14:20:51 ... I want to make sure that manipulating the filterLayout is a good to do it now that we have quite a few of them 14:20:57 ... I'll need to look more into it 14:21:20 anssi: let's keep the PR open until we get reviews from Chai & Rama 14:21:48 Topic: TPAC registration and agenda building 14:22:10 Anssi: first, Public Service Announcement: registration for TPAC has opened https://www.w3.org/2021/10/TPAC/#participation 14:22:20 ... it's a very quick registration process 14:22:41 ... you should receive to the virtual attendee hub that will help you join the right meeting during TPAC week 14:24:00 dom: this is pending a technical glitch in our system, should be fixed soon 14:24:22 Anssi: Another announcement: we have 3 confirmed meetings during TPAC, 26-28 at 14:00 UTC 14:24:54 ... For that meeting, please raise suggestion of topics via github https://github.com/webmachinelearning/meetings/issues/18 14:24:59 Subtopic: Integrating an open-source cross-platform implementation of the Web Neural Network API into a web engine 14:25:35 Anssi: we've received this proposal that I think should be of interest for folks working on Web engines 14:25:47 ... in particular, we would invite non-WG participants as guests to that session 14:25:56 ... e.g. folks from the Dawn project 14:26:03 ... a cross-platform webgpu implementaiton 14:26:13 s/tait/tati/ 14:26:30 ... WebNN-native shares quite a bit of infrastructure (inc a code generator) with the Dawn project 14:26:44 Present+ Jonathan 14:27:06 Anssi: who should we invite from Chromium for such a topic? Kyle? 14:27:53 q+ 14:27:58 s/Kyle/Kai 14:28:01 ack RafaelCintron 14:28:18 RafaelCintron: I would be interested to join that session 14:28:21 q+ 14:28:33 ack dom 14:28:41 Anssi: absolutely - maybe also other Microsoft folks not generally involved 14:29:24 dom: maybe this could be a breakout session? 14:29:51 anssi: great idea; logistically, this timeslot works for this WG which makes it easier 14:30:01 ... but we could have it as a breakout instead or in addition 14:30:16 ... the first step is to identify who to have at that session 14:30:52 q? 14:31:00 Anssi: we would love to get Chrome engineers involved in Dawn in this discussion 14:31:55 Ningxin: last year, during the workshop, Corentin the WebGPU co-chair gave us very good input on the WebGPU & Dawn 14:32:07 ... we can check with Corentin about his interests & bandwidth 14:32:24 ... when we started with WebNN, Corentin provided us valuable feedback from the Dawn usage perspective 14:32:51 ... another aspect for the Web specs, we actually define the interop between WebNN & WebGPU buffers with Chai's input 14:33:03 ... Wrt Dawn's usage, it might be a good discussion point for WebNN/WebGPU interop 14:33:23 ... if they use a single infrastructure, would it make it easier for implementors to provide interop between the two systems 14:33:29 ... that would be an interesting topic 14:34:00 anssi: please let us know if there are people outside this WG interested in this discussion 14:35:12 Subtopic: other TPAC topics proposals 14:35:52 Anssi: our group is set to work on ethical issues - I've started a document to bootstrap that work that I could present at TPAC 14:36:03 ... Dom & I have started to reach out to guests to join that effort 14:36:21 ... I know most companies working in this space have dedicated people looking at ethical issues 14:36:30 ... it would be much appreciated if you could connect these folks with us 14:37:09 Dom: +1 on that plan 14:37:14 dom: having something out will encourage more contributions 14:37:35 anssi: what I've learned is that I know very little about it, and it's a very cross-discipline topic 14:37:56 ... there is a lot to learn 14:38:39 ... we don't want to reinvent the wheel, but develop perspectives in the lens of the Web platform 14:39:15 ... based on its specificities - e.g. the Web has the broadest reach of all platforms, making ML/AI that much more powerful when available there 14:39:25 ... I would love to have someone outside of this WG educate us on this topic 14:39:36 ... I'll get a starting point in a repo 14:40:45 ... for our TPAC discussions, I expect we'll have someone to lead a given topic to seed the discussions with an explanation of where we are 14:40:56 ... similar (but less formally structured) as the workshop 14:42:04 Anssi: going through the list of proposed topics 14:42:06 ... "Rationale/criteria for adding new ops to the WebNN API" 14:42:25 ... this would help people to contribute to the work - anyone willing to lead this? 14:42:45 ... e.g. some frameworks have similar criteria - could they be adapted to WebNN? 14:42:58 ... next topic "Versioning and web compatibility" 14:43:13 ... it would be great to provide a Web perspective on this 14:45:01 RRSAgent, draft minutes 14:45:01 I have made the request to generate https://www.w3.org/2021/09/30-webmachinelearning-minutes.html anssik 14:45:26 dom: someone from the TAG would be good person to kick off this discussion 14:46:24 dom: interesting topic in general, with a specific angle for this technology 14:46:37 ... not entirely sure how to structure this conversation 14:46:49 anssi: getting someone from the TAG involved would be interesting 14:47:06 q? 14:47:26 ... next topic "How should WebNN API deal with fusion" 14:47:44 ... again, is someone passionate about this topic? or is there an expert we should reach out to? 14:48:28 ... IIRC Daniel Smirkow from Google raised this topic early on 14:49:10 ... another topic is "ML JS framework performance" - ensuring that WebNN fulfills its promises to that primary category of customers 14:49:16 ... i.e. WebNN as backed to JS frameworks 14:49:58 Ningxin: I've been working with Chai and Ping into ONNX-Web and TF-Lite 14:50:26 ... we're trying to develop a design doc (in progress), with use cases and running code based on WebNN-native implementations 14:50:36 ... to share some data 14:50:54 ... we should be able to share the status of that WIP and that data 14:51:00 Anssi: sounds great 14:51:48 Ningxin: Chai has a lot of experience with DirectML integration with ONNX, Ping with TF 14:52:20 q+ 14:53:41 dom: please remember to register to TPAC and if you want to propose topics outside this WG submit breakout sessions, soft close for breakouts Oct 8 14:54:25 ack ningxin_hu 14:55:11 Ningxin: I'd like to propose a topic of discussion with regard to integrating with Web platform tests 14:55:22 ... there has been progess happening in the polyfill repo 14:55:39 ... the maintainer has recently converted his mocha-based tests into WPT test harness tests 14:55:44 ... incl work on IDL harness 14:56:15 ... this may be a good topic for discussion for the WG, at TPAC or otherwise 14:57:05 Anssi: good suggestion! do you if Bruce need any specific help? 14:57:46 https://github.com/webmachinelearning/webnn-polyfill/issues/2#issuecomment-920754537 14:57:56 Ningxin: may need to invite someone from WPT to help with some of the open issues 14:58:46 ... Another aspect is based on what Chai had shared during the workshop on test granularity 14:58:51 ... op-level vs … 14:59:03 ... it may be a good time to discuss these principles 14:59:24 ... and existing gaps in the test suite as a roadmap for our work in this space 14:59:26 -> https://github.com/w3c/machine-learning-workshop/issues/80 Conformance testing of ML APIs for the Web 15:00:37 anssik: Happy National Day holiday (1-7 Oct) to our participants in China! 15:00:56 RRSAgent, draft minutes 15:00:56 I have made the request to generate https://www.w3.org/2021/09/30-webmachinelearning-minutes.html anssik 17:00:44 Zakim has left #webmachinelearning