13:34:44 RRSAgent has joined #webmachinelearning 13:34:44 logging to https://www.w3.org/2021/09/02-webmachinelearning-irc 13:34:47 RRSAgent, make logs Public 13:34:47 please title this meeting ("meeting: ..."), anssik 13:34:49 Meeting: WebML WG Teleconference – 2 Sep 2021 13:34:53 Chair: Anssi 13:35:03 Agenda: https://github.com/webmachinelearning/meetings/blob/master/telcons/2021-09-02-agenda.md 13:35:03 Scribe: Anssi 13:35:07 scribeNick: anssik 13:35:12 Present+ Anssi_Kostiainen 13:35:17 RRSAgent, draft minutes 13:35:17 I have made the request to generate https://www.w3.org/2021/09/02-webmachinelearning-minutes.html anssik 13:56:24 RafaelCintron has joined #webmachinelearning 13:59:14 ningxin_hu has joined #webmachinelearning 13:59:46 Present+ Ningxin_Hu 13:59:52 Present+ Rafael_Cintron 14:01:35 Present+ Wonsuk_Lee 14:01:52 Present+ Present+ Ganesan_Ramalingam 14:02:06 rama has joined #webmachinelearning 14:02:07 Present+ Sungpil_Shin 14:03:09 Topic: WebNN API TAG review closure 14:03:25 anssik: All TAG review issues have been addressed or have PRs in review. 14:04:00 ... thanks everyone, it is very helpful to have TAG review done at this stage of the spec development. We may reach out to TAG again for further advice as we progress. 14:04:23 ... Today's plan is to review the remaining PR and make decisions whether to close the issues 14:04:32 Subtopic: [tag-tracker] Define a common term for logical tensor changes? 14:04:44 -> https://github.com/webmachinelearning/webnn/issues/150 Issue #150 14:04:52 -> https://github.com/webmachinelearning/webnn/pull/196 PR #196 14:05:12 anssik: My assessment is that PR #196 by Rama did also address issue #150. Do folks agree with that? 14:05:30 wonsuk has joined #webmachinelearning 14:05:30 anssik: thanks Rama for this contribution! 14:05:36 +1 14:05:52 Jonathan has joined #webmachinelearning 14:05:54 Rama: added an overview section to discuss and cover semantical aspects of the API. 14:06:05 Present+ Jonathan_Bingham 14:06:48 ... TAG issue refers to the fact that operators can return a shared view to a tensor, and this is mostly an implementation notion 14:07:10 -> https://webmachinelearning.github.io/webnn/#programming-model-overview Programming Model Overview 14:07:44 anssik: any concerns in closing issue #150? 14:08:13 Subtopic: [tag-tracker] Ergonomics of the JS examples 14:08:18 -> https://github.com/webmachinelearning/webnn/issues/139 Issue #139 14:08:24 -> https://github.com/webmachinelearning/webnn/pull/202 PR #202 14:08:49 anssik: PR #202 adds an introduction to the WebNN API spec that notes ML JS frameworks are primary consumers of the API 14:09:33 ... I hope this upfront introduction will help the reader orientate oneself around the API, and reason about the API design choices made and the level of abstraction 14:09:41 ... the section also notes there's complementary work on Model Loader API that defines a higher-level abstraction targeting primarily web developers 14:10:05 ... any concerns in merging PR #202 and closing #139? 14:10:25 anssik: no concerns, will merge and close the issue 14:10:50 Subtopic: [tag-tracker] String enum for activations 14:11:03 -> https://github.com/webmachinelearning/webnn/issues/138 Issue #138 14:11:16 -> https://github.com/webmachinelearning/webnn/pull/188 PR #188 14:12:17 anssik: any further comments for issue #138 that was closed? 14:12:25 anssik: (none heard) 14:13:10 Topic: WebNN API recent new feature requests 14:13:24 anssik: As usual, let's briefly discuss and triage new feature requests 14:13:35 Subtopic: Request input layout and resize only on height and width for Resample 14:13:39 -> https://github.com/webmachinelearning/webnn/issues/200 Issue #200 14:13:44 -> https://github.com/webmachinelearning/webnn/pull/205 PR #205 14:14:21 anssik: Request for resample2d op that takes a 4-D tensor with layout information (nchw or nhwc) and only supports interpolation on spatial dimensions (h and w) 14:14:53 ningxin_hu: issue comes from implementation feedback and test perspective 14:15:04 ... from a contributor to webnn-native project 14:15:13 ... today we have DirectML and OpenVINO backends 14:15:58 ... according to implementation findings WebNN ops can be only directly map to DML, not OpenVINO, only support resize and resample in spatial dimension 14:16:50 ... I think this is a good example of implementation feedback informing spec development 14:17:02 ... we want to be compatible with major native ML APIs 14:18:11 ... I submitted PR for resample2d with options for resize the spatial dimensions and input layout 14:19:40 Regrets+ Chai_Chaoweeraprasit 14:20:15 ningxin_hu: I'd like to request Google folks to review this from Google's API perspective 14:20:16 Chai should be back from vacation on Tues. 14:20:43 ningxin_hu: in Android NNAPI Spec, there're two similar function ops ANEURALNETWORKS_RESIZE_BILINEAR and ANEURALNETWORKS_RESIZE_NEAREST_NEIGHBOR which could resize input tensor 14:21:30 Subtopic: Change the min and max value to attributes instead of operands 14:21:33 Sure, I'll ask the NNAPI team. 14:21:37 -> https://github.com/webmachinelearning/webnn/issues/203 Issue #203 14:21:41 -> https://github.com/webmachinelearning/webnn/pull/206 PR #206 14:22:05 anssik: as the topic says, change minValue/maxValue from MLOperand to float type 14:22:12 ... Ningxin? 14:22:32 Present+ 14:22:44 ningxin_hu: this is similar to the previous request, issue reported by webnn-native contributor from implementation perspective 14:23:08 ... current clamp definition not mapping to some backends such as OpenVINO 14:24:00 ... proposing to change clamp options to float instead of MLOperand so this definition maps directly and explicitly to native ML APIs 14:24:25 ... if the framework or web app wants clamp as an operand, that is still possible, since clamp can be emulated with min and max 14:25:17 ok 14:25:55 Subtopic: Support for configuring rounding type in pooling operations 14:26:00 -> https://github.com/webmachinelearning/webnn/issues/198 Issue #198 14:26:20 anssik: This was discussed on our earlier meeting as well 14:26:35 ... Chai provided insights re ONNX pooling support 14:27:45 RafaelCintron: Chai provided feedback in the issue 14:28:06 ningxin_hu: I think Chai's comments are good, I still need some time to work on this 14:30:20 Subtopic: Timing 14:30:38 i/Subtopic: Timing/Topic: TPAC 2021 14:30:51 Anssi: TPAC2021 is happening between Oct 18 and 29 14:31:17 i/Subtopic/scribe+ 14:31:28 ... Tue 26 Oct 2021 14 UTC 14:31:28 ... Wed 27 Oct 2021 14 UTC 14:31:28 ... Thu 28 Oct 2021 14 UTC (the usual slot) 14:31:33 ... suggests re-using our regular slot, and maybe adding one or two at the same time on Tue/Wed 14:32:55 ... will follow up with a Doodle of some sort to verify ability 14:33:00 Subtopic: Topics 14:33:04 s/abi/availabi/ 14:33:10 -> https://lists.w3.org/Archives/Public/public-webmachinelearning-wg/2021Aug/0000.html Dom's TPAC guidance 14:33:23 Anssi: Dom explained TPAC meeting expectations in its guidance mail 14:33:41 ... our usual biweekly meetings tend to be fairly technical discussions to make progress on the spec 14:33:57 ... a TPAC meeting is an opportunity to take a higher level view, the bigger picutre 14:34:00 s/cut/ctu/ 14:34:15 ... - Rationale/criteria for adding new ops to the WebNN API 14:34:15 ... - Versioning and web compatibility 14:34:15 ... - How should WebNN API deal with fusion 14:34:15 ... - ML JS framework performance, focus areas for WebNN? 14:34:15 ... - Ethical issues in using Machine Learning on the Web 14:35:04 Anssi: re ops, we need an organized way to make decision on adding ops - this could make a good TPAC discussion 14:35:19 ... re versioning, it would be good to have discussion on how versioning works (or doesn't) for the Web platform 14:35:43 ... re fusion - this emerged from our recent discussions 14:36:20 ... re JS framework perf - would be interesting to get benchmark results on how well WebNN is doing 14:36:36 ... (TF.js people's input would be particularly useful) 14:36:58 ... last, ethical issues - I've been noodling on this, would be good to have discussions on it 14:37:12 q+ 14:37:33 ... any other suggestion? 14:39:11 jonathan: ChromeOS is quite interested in both model loader and WebNN 14:39:25 Jonathan: Chrome OS team interested in both Model Loader API and WebNN API and working to get perf benchmarks and security implementations 14:39:38 ... they're running perf benchmark and security review - they might be in a position to get feedback at TPAC on their work 14:39:39 ... also want to get to a point to share more with this group and share code 14:40:08 q+ 14:40:33 q? 14:40:47 ack dom 14:41:19 dom: in addition to technical topics for TPAC, we might consider more of where we are, where are we going style topics, status check of the group if you will 14:41:37 ... another thing, the week of WG meetings is the 2nd week of TPAC Oct 25-28 14:41:57 ... before that there's TPAC breakouts week for topics that do not directly map to existing groups 14:42:09 ... some of the topics might be interesting in that context as well 14:42:22 ... for example versioning 14:42:55 q? 14:42:59 ack RafaelCintron 14:43:09 RafaelCintron: couple of question to Jonathan 14:43:23 ... what features and use cases Chrome OS would like to use Model Loader API for? 14:43:31 ... timeline for code to be available for the group to comment on? 14:43:31 RafaelCintron: Jonathan, out of curiosity, what's driving the implementation in ChromeOS? any sense of the timeline when more might be shared? 14:44:02 jonathan: re use case, ChromeOS only run Web Apps, and having ML acceleration would be beneficial to many apps 14:44:12 RRSAgent, draft minutes 14:44:12 I have made the request to generate https://www.w3.org/2021/09/02-webmachinelearning-minutes.html anssik 14:44:36 ... Model loader would be useful for a number of products that would rely e.g. on TF life models 14:45:04 ... re timeline, hope is to get an OT started maybe next year, which requires the code to be open 14:45:04 s/TF life/TF Lite 14:45:23 ... hope that means it will be so soon 14:45:24 q? 14:46:36 anssi: will continue collecting ideas 14:46:39 Subtopic: Demos 14:46:55 Anssi: there is an opportunity to share technical demos with the broader technical community at TPAC 14:47:04 ... I think we already have a bunch of demos available 14:47:33 dom: we use TPAC as a way for the community to learn what other groups are doing and also broader public 14:47:56 ... what we are suggesting is for group to prepare short demos 14:48:21 ... demo has two potential goals: 1) high-level illustration what the group is working on and why it matters 14:48:37 ... e.g. why ML on the Web matters and what does it change in terms of performance 14:48:51 ... 2) look at the intersection of WebNN with other web technologies 14:49:09 ... e.g. WebRTC integration to run each video frame 14:49:23 ... we may not have a such a demo, but that'd be a great discussion started with WebRTC WG 14:50:37 dom: we'd need demo description by Sep 15, and recorded demo by Sep 27 14:50:37 https://www.w3.org/wiki/TPAC2021/Demos_and_Group_updates 14:50:54 webnn-samples implement bunch of webnn use cases that can run on webnn-polyfill and webnn-native: https://github.com/webmachinelearning/webnn-samples 14:51:01 anssik: does the demo have to be a recording? 14:51:06 that could be good candidate 14:51:59 dom: having a link to the polyfill would be great, but it may not demonstrate full potential 14:52:24 ... video recording of WebNN as a browser implemented primitive can be more impactful 14:52:44 webnn-native could be used to demonstrate near-to-native performance 14:53:07 q+ 14:53:13 q? 14:53:55 dom: demo can be a group effort from multiple people contributing 14:54:25 dom: contact dom for more info on demos 14:54:26 q? 14:54:27 ack ningxin_hu 14:54:52 ningxin_hu: I'll think if we can use WebNN samples that implement a few key WebNN use cases 14:55:15 ... there demos have a nice UI and good implementation 14:55:28 ... these can run in existing browsers with webnn-polyfill 14:55:52 ... also webnn-native can be used to run these demos through node.js binding in an Electron environment 14:56:09 ... this setup might work for early demonstration and perf projection 14:56:24 dom: that sounds great 14:57:03 ... explaining the impact on use cases that are enabled by hardware acceleration would be great 14:57:51 Subtopic: Joint meetings 14:59:18 anssik: Wasm, WebGPU, WebRTC, TAG? 14:59:22 ... I checked these groups current plans: 14:59:25 ... WebRTC Oct 28, 15 UTC or Oct 29, 15 UTC 14:59:32 ... Wasm, WebGPU had not yet announced their meeting times 15:00:18 dom: my experience of joint meetings is to have a well prepared agenda around joint topics 15:00:44 ... e.g. in WebRTC the question that exists there is how and when one might start using WebNN to do some of the fancy processing on audio and video transmitted by WebRTC 15:00:57 ... what is the impact on architecture, impact on codecs etc. 15:01:22 ... I'm not sure we have even through about this questions, if we were to organize a joint meeting with WebRTC we need to look into it 15:01:25 q? 15:02:38 anssik: Dom can you start discussing with WebRTC WG chairs on joint meeting idea? 15:02:54 RRSAgent, draft minutes 15:02:55 I have made the request to generate https://www.w3.org/2021/09/02-webmachinelearning-minutes.html anssik 15:03:41 dom: only thing re TPAC we need to register interest to meet by the end of next week 15:04:04 q? 15:04:30 RRSAgent, draft minutes 15:04:30 I have made the request to generate https://www.w3.org/2021/09/02-webmachinelearning-minutes.html anssik 17:12:40 Zakim has left #webmachinelearning