14:01:31 RRSAgent has joined #webmachinelearning 14:01:31 logging to https://www.w3.org/2020/10/29-webmachinelearning-irc 14:01:33 RRSAgent, make logs Public 14:01:34 please title this meeting ("meeting: ..."), anssik 14:01:39 Meeting: WebML CG Teleconference – 29 October 2020 14:01:44 Chair: Anssi 14:01:49 Agenda: https://github.com/webmachinelearning/meetings/blob/master/telcons/2020-10-29-agenda.md 14:01:54 Scribe: Anssi 14:02:00 scribeNick: anssik 14:02:04 Present+ Ningxin_Hu 14:02:15 Present+ Anssi_Kostiainen 14:02:22 Present+ Chai_Chaoweeraprasit 14:02:27 Present+ Rafael_Cintron 14:02:40 RRSAgent, draft minutes v2 14:02:40 I have made the request to generate https://www.w3.org/2020/10/29-webmachinelearning-minutes.html anssik 14:02:54 Chai has joined #webmachinelearning 14:05:11 Topic: WebNN API explainer 14:05:17 -> https://github.com/webmachinelearning/webnn/blob/master/explainer.md WebNN API Explainer 14:05:24 -> https://github.com/webmachinelearning/webnn/pull/109 Update the webnn explainer docs (PR #109) 14:05:40 RafaelCintron has joined #webmachinelearning 14:06:05 RRSAgent, make logs public 14:06:17 Chai: Considered alternatives still missing content 14:07:51 anssik: perhaps mention Model Loader and MLIR-inspired proposals 14:08:02 Chai: those are actually already discussed in other sections 14:09:54 ... we haven't had any U-turns on the API design 14:10:08 ... I can do some cleanup and fill the acknowledgements sections 14:10:31 -> https://github.com/webmachinelearning/webnn/blob/master/explainer.md#considered-alternatives Considered alternatives 14:10:48 anssik: "One of the most important things you can do in your design process is to catalog the set of roads not taken. As you iterate on your design, you may find that major choices in your approach or API style will be revisited and enumerating the full space of alternatives can help you apply one (or more) of them later, may serve as a “graveyard” for u-turns in your design, and can give reviewers and potential users 14:10:48 confidence that you’ve got your ducks in a row." 14:12:20 Rafael: Maybe we could delete this section 14:12:39 ... only alternative we could ask here is Model Loader, or why don't we just use WebGL/GPU 14:13:28 ... refer to sections providing answers to those question 14:14:40 ningxin_hu: as discussed before, agree perf and efficiency are to be highlighted in here 14:15:31 anssik: after completing the considered alternatives and ack section, Anssi to initiate TAG review 14:15:42 Topic: Support the execution of the sub-graph scenario 14:16:11 -> https://github.com/webmachinelearning/webnn/issues/105 Support the execution of the sub-graph scenario (issue #105) 14:16:53 ningxin_hu: this should be covered in the PR #94 14:17:02 ... also related issue #105 14:17:43 Ping's comment: "With this API, the computation is tied with the compilation. should the compile method have the inputs/outputs pair and execution can only execution the graph compiled with the input/output pair? Otherwise, the compilation might not support the execution of the sub-graph scenario?" 14:20:02 Chai: two ways to do this, described in a PR: 14:21:01 https://github.com/webmachinelearning/webnn/pull/94#discussion_r498289215 14:21:24 anssik: what is Ping's reaction to the proposal? 14:21:29 Chai: no comments yet 14:22:30 ningxin_hu: Ping would need to elaborate the use case to understand the gap better 14:23:11 Topic: Proposed new ops Mirrorpad, SquaredDifference, Pow, TransposeConv 14:23:23 anssik: Please review the issue and provide feedback 14:23:28 -> https://github.com/webmachinelearning/webnn/issues/108 WebNN API is not support Mirrorpad, SquaredDifference, Pow, TransposeConv (issue #108) 14:23:44 q+ 14:23:49 ack Chai 14:24:16 Chai: this issue is about how do we know what models we need to support? 14:24:31 ... we should look at the models people want, this is one such a request 14:24:53 ... I'm open to brainstorming what models we should look at and map into the spec 14:25:16 ... personally I don't prefer expanding the first wave models, but we need a way to accept extensions 14:26:14 ... we could have a separate repo e.g. "models" under "webmachinelearning" GH org 14:27:06 examples: ONNX model zoo https://github.com/onnx/models 14:27:30 TensorFlow model garden: https://github.com/tensorflow/models 14:27:47 anssik: What type of questions we should ask from folks proposing new ops to be added? 14:29:12 q+ 14:29:17 ack Chai 14:29:52 Chai: Style Transfer is pretty known model and supported in e.g. ONNX Model Zoo, so I think the model we might not support are those that are used in classical ML for example 14:30:03 ... any Deep Neural Network would be good to look into supporting 14:30:17 ... nothing wrong in Style Transfer 14:31:29 anssik: can you respond to Calvin maybe providing platform support information? 14:31:59 ningxin_hu: can do that, also cross-check between ONNX and XLA-HLO 14:32:27 ... re model criteria, maybe ask folks to look at the use cases in the WebNN spec, if the use case is not included need to start from there 14:32:40 https://webmachinelearning.github.io/webnn/#usecase-style-transfer 14:33:09 ningxin_hu: linking models to use cases could be the precondition 14:33:21 anssik: style transfer indeed one of the use cases 14:34:42 + 1 on ningxin's 14:35:19 anssik: you can implement only a subset of the use cases with the first-wave ops? 14:37:13 Topic: Secure Context 14:37:18 anssik: proposal to expose WebNN API only over HTTPS. 14:37:33 -> https://github.com/webmachinelearning/webnn/pull/101 Added extended attributes to Web IDL definition (PR #101) 14:38:14 anssik: ningxin can you work with Wonsuk to address the PR conflicts? 14:38:43 ningxin_hu: conflicts due to changes in deployment infra, will work with Wonsuk on this 14:39:07 Topic: WebNN API ergonomics improvements 14:39:13 Subtopic: Chained API for the Operands 14:39:18 -> https://github.com/webmachinelearning/webnn/issues/106 Chained API for the Operands (issue #106) 14:39:59 anssik: Ping's proposal 14:40:49 anssik: parking this for later to have sync discussion with Ping 14:41:03 Subtopic: Options dictionary for functions with lots of args 14:41:12 anssik: this issue #90 was fixed with PR #112, anything to report? 14:41:16 ... thanks Zoltan for the proposal, and Chai for the PR 14:41:21 -> https://github.com/webmachinelearning/webnn/issues/90 Suggestion: use an options dictionary for functions with lots of args (issues #90) 14:41:27 -> https://github.com/webmachinelearning/webnn/pull/112 Use optional dictionaries for operator's optional parameters (PR #112) 14:41:51 Chai: great suggestion, helped with versioning 14:42:14 q+ 14:43:13 ack ningxin_hu 14:43:53 ningxin_hu: want to mention re Chai's update to add the dict-based options, I'm working to update the webnn-polyfill to match 14:46:11 https://huningxin.github.io/webnn-samples/code/?example=mul_add.js 14:46:59 https://github.com/webmachinelearning/webnn-samples/pull/19 14:48:19 RRSAgent, draft minutes v2 14:48:19 I have made the request to generate https://www.w3.org/2020/10/29-webmachinelearning-minutes.html anssik 14:58:07 Topic: Adjourn 14:58:08 RRSAgent, draft minutes v2 14:58:08 I have made the request to generate https://www.w3.org/2020/10/29-webmachinelearning-minutes.html anssik 15:24:10 myles has joined #webmachinelearning 16:48:30 Zakim has left #webmachinelearning