14:00:42 RRSAgent has joined #webmachinelearning 14:00:42 logging to https://www.w3.org/2020/08/20-webmachinelearning-irc 14:00:48 Zakim has joined #webmachinelearning 14:00:52 RRSAgent, make logs public 14:00:57 Meeting: WebML CG Teleconference – 20 August 2020 14:01:00 ningxin_hu has joined #webmachinelearning 14:01:03 Chair: Anssi 14:01:12 scribeNick: anssik 14:01:17 Present+ Anssi_Kostiainen 14:01:17 Rama has joined #webmachinelearning 14:01:27 Present+ Chai_Chaoweeraprasit 14:01:31 Present+ Rafael_Cintron 14:01:36 Present+ Ningxin_Hu 14:01:41 Present+ Ganesan_Ramalingam 14:03:04 chai has joined #webmachinelearning 14:04:59 TOPIC: Welcome back after the break! 14:05:28 TOPIC: Announcement: WebML CG meeting at TPAC 14:05:42 anssik: Proposed WebML CG Meeting at TPAC 2020 on October 22 15:00 UTC: 14:05:48 -> https://www.w3.org/wiki/TPAC/2020/GroupMeetings Overall TPAC 2020 schedule 14:06:12 anssik: TPAC is an opportunity to interact with the broader W3C community. 14:06:30 ... Suggestions for topics for joint meetings with other W3C groups: WebGPU, Wasm, Audio? 14:07:12 TOPIC: Announcement: Web & ML workshop presentations published 14:07:26 anssik: This week we published the first wave of presentations, thanks to all the speakers! 14:07:33 -> https://www.w3.org/2020/06/machine-learning-workshop/presentations.html Workshop presentations 14:07:57 anssik: Discussion on presentations happens via GitHub, speakers are tagged, but everyone is encouraged to take part in the discussions 14:08:15 -> https://github.com/w3c/machine-learning-workshop/issues Workshop GH 14:08:32 anssik: The workshop culminates in September into four interactive sessions that will summarize the finding and chart out a path for future W3C efforts. 14:08:45 anssik: August 14, 2020 was the deadline to register as workshop participant, if you missed the DL but think you should attend, get in touch and we'll look into it; we got 800 registrants overall but had to filter the list down to 150 to have a reasobly sized room for discussions 14:09:39 TOPIC: WebNN polyfill and samples 14:09:50 anssik: The group agreed to add WebNN API polyfill and samples to its deliverables: 14:09:58 https://github.com/webmachinelearning/webnn/issues/81 14:10:01 https://github.com/webmachinelearning/webnn/issues/70 14:10:13 RRSAgent, draft minutes v2 14:10:13 I have made the request to generate https://www.w3.org/2020/08/20-webmachinelearning-minutes.html anssik 14:11:17 anssik: I've asked Ningxin to give an introduction to the initial contribution from Intel proposed to be used as a starting point, latest status, issues, plan, areas that welcome further contributions etc. 14:11:58 https://github.com/webmachinelearning/webnn-polyfill/pull/1 14:12:07 ningxin_hu: I created a PR to add the polyfill foundation 14:12:24 ... this PR implements 10 ops from the latest WebNN API spec 14:12:33 ... required by 1st wave models 14:12:59 ... the ops supported are listed in the PR comments 14:13:40 ... the ops are included because they are used by our basic examples, we also agreed to add an advanced example handwriting detection with a convnet 14:14:01 ... 10 ops implemented by the polyfill are required by the two examples included 14:14:30 ... TFJS WebGL kernels used, if not available CPU kernels are used, but that's slow at this point 14:14:54 ... licensed under a permissive Apache 2.0 14:15:42 ningxin_hu: exact implementation of the polyfill uses TypeScript transpiled into JS 14:16:26 anssik: are there open issues folks can help with? 14:16:51 ningxin_hu: we need to add the ops required by 1st wave ops, some new are being specified so those need to be added to the polyfill 14:17:15 ... the spec contains 30+ ops, polyfill supports 10 ops, so need to bridge this gap 14:17:56 ... another TODO, compilation options are not implemented by the polyfill, always using OpenGL kernels 14:20:24 ningxin_hu: we develop spec and polyfill in parallel, it allows others to get a feel of the API early on 14:20:54 chai: WebNN API spec is getting bigger, and some of the ops are not trivial, so having some unit tests will be helpful 14:21:41 ... it should ideally follow the models we're targeting with the API spec, currently the polyfill is in a catch up mode 14:22:03 ningxin_hu: I also added some unit tests for the ops in the PR 14:22:47 ... not a full conformance test 14:24:19 ningxin_hu: as chai mentioned, we want to improve certain areas in the polyfill and will open new issues for those 14:25:50 anssik: polyfill repo for discussing polyfill implementation topics, webnn repo for spec topics, don't worry if you create an issue in the wrong repo, they're easy to move around 14:27:38 anssik: what is browser / OS support? 14:28:30 ningxin_hu: for development, I use Windows laptop, test with Chrome and Edge 14:29:07 ... using TFJS WebGL kernels, so WebGL perf critical 14:29:28 ... did not test with other browsers or OSes yet, contributions welcome 14:30:06 TOPIC: First-wave models and ops delta with WebNN API definition 14:30:21 s/TOPIC: First-wave models and ops delta with WebNN API definition// 14:30:54 ningxin_hu: I added more advanced samples in addition to the polyfill 14:30:56 https://github.com/webmachinelearning/webnn-samples/pull/1 14:31:17 ningxin_hu: using NIST dataset, LeNet arch in this sample 14:31:22 https://huningxin.github.io/webnn-samples/lenet/ 14:31:29 ... to facilitate review, I host this sample live in my fork 14:32:28 ... two modes: use MNIST dataset, or use mouse to draw your own 14:33:14 https://github.com/huningxin/webnn-samples/blob/lenet/lenet/lenet.js#L22-L96 14:33:25 ... this sample uses the polyfill, the code to build the network ^ 14:33:44 ... uses multiple ops from the 1st wave 14:33:57 RRSAgent, draft minutes v2 14:33:57 I have made the request to generate https://www.w3.org/2020/08/20-webmachinelearning-minutes.html anssik 14:35:40 chai: having the actual code will help uncover integration issues, polyfill is kind of a reference implementation 14:36:31 ningxin_hu: agree with chai, samples TODO includes adding support for 1st wave models, ResNet etc. 14:37:08 anssik: any questions? 14:37:30 TOPIC: First-wave models and ops delta with WebNN API definition 14:37:39 Review the delta between the first-wave models ops and WebNN API surface: 14:37:48 - clamp 14:37:48 - globalAveragePool 14:37:48 - gru 14:37:48 - sigmoid 14:37:48 - split 14:37:49 - squeeze 14:38:00 anssik: Discuss which ops to add to the spec definition considering e.g. major platform support and performance implications. 14:38:25 https://github.com/webmachinelearning/webnn/pull/83 14:39:56 chai: the set of models we're targeting, and ops we define should go hand in hand, many models use a shared set of ops, but not exactly the same set necessarily 14:40:30 ... adding model support as we go has been successful approach in some of our projects 14:42:10 ... that is, all models in 1st wave we should support and define required ops as we go, iterate with polyfill to understand what is testable and verifiable 14:43:54 TOPIC: Noise suppression 14:44:07 https://github.com/webmachinelearning/webnn/issues/66 14:44:28 q+ 14:45:48 ack chai 14:47:35 chai: I spend time looking at jmvalin's models, PR sent out for GRU to partially address the issue: https://github.com/webmachinelearning/webnn/pull/83 14:48:07 ... RNNoise relies on GRU and looking forward to fully support it, with couple of extra ops we'd be good to go 14:50:06 ... in terms of issue #66, once we define all the ops we should be able to close it 14:50:45 chai: related to this, questions re 1st wave models? 14:50:59 ... how to add models to 1st wave models? 14:51:47 q+ 14:51:55 ack RafaelCintron 14:52:15 RafaelCintron: I'm fine having multiple waves of models 14:52:58 ... Teams folks would like to have some models to have them work on their web app, maybe their models can be done with 1st wave already 14:54:27 chai: it is useful to define models in waves, we can qualify a wave as complete, makes it more trackable 14:54:34 ningxin_hu: agree with chai 14:54:48 ... we can make wave a fixed timeframe cadence, like quarterly? 14:55:28 ... another things I'd like to raise, we can link these models with out use cases defined in the spec 14:55:51 ... the flow from use cases -> models -> ops makes sense to document 14:57:09 +1 14:58:01 q? 14:58:33 TOPIC: Adjourn 14:59:01 RRSAgent, draft minutes v2 14:59:01 I have made the request to generate https://www.w3.org/2020/08/20-webmachinelearning-minutes.html anssik 16:59:56 Zakim has left #webmachinelearning