14:20:03 RRSAgent has joined #webmachinelearning 14:20:03 logging to https://www.w3.org/2019/02/14-webmachinelearning-irc 14:20:08 Zakim has joined #webmachinelearning 14:20:13 RRSAgent, make logs public 14:20:21 Meeting: WebML CG Teleconference – 14 February 2019 14:20:26 Chair: Anssi 14:20:31 Scribe: Anssi 14:20:35 Agenda: https://github.com/webmachinelearning/meetings/blob/master/telcons/2019-02-14-agenda.md 14:20:40 scribeNick: anssik 14:20:58 Regrets+ Nikhil_Thorat 14:21:04 Regrets+ Daniel_Mazurkiewicz 14:21:09 Regrets+ Sangwhan_Moon 14:21:21 RRSAgent, draft minutes v2 14:21:21 I have made the request to generate https://www.w3.org/2019/02/14-webmachinelearning-minutes.html anssik 14:22:23 present+ gregwhitworth 14:25:46 Present+ Anssi_Kostiainen 14:26:09 Present+ Tomoyuki_Shimizu 14:26:13 Present+ Rafael_Cintron 14:28:44 dsmilkov has joined #webmachinelearning 14:30:00 Present+ Daniel_Smilkov 14:30:12 kreeger has joined #webmachinelearning 14:30:23 Present+ Nick_Kreeger 14:30:40 RafaelCintron has joined #webmachinelearning 14:30:47 NingxinHu has joined #webmachinelearning 14:31:10 Present+ Ningxin_Hu 14:31:22 belem has joined #webmachinelearning 14:31:46 present+ 14:31:57 present+ 14:32:44 Present+ G_Ramalingam 14:32:58 RRSAgent, draft minutes v2 14:32:58 I have made the request to generate https://www.w3.org/2019/02/14-webmachinelearning-minutes.html anssik 14:33:39 TOPIC: High level vs low level revisited 14:33:54 -> #3 High level vs low level https://github.com/webmachinelearning/webnn/issues/3 14:34:17 anssik: a lot of discussion in this mega issue #3, hard to resolve since we're branching to multiple directions 14:34:38 ... proposal to split into self-contained issues that we can resolve one at a time and make progress with 14:34:52 rgesteve has joined #webmachinelearning 14:34:53 anssik: my key takeaways from the mega issue #3 are the following positions, quoting: 14:35:10 ... @nsthorat: Our preference from TensorFlow is for this specification to focus on operations. 14:35:19 ... @RafaelCintron: WebML should include both a "loadModel-style" and "operator-style" APIs. 14:35:32 -> @nsthorat's position https://github.com/webmachinelearning/webnn/issues/3#issuecomment-453129272 14:35:37 -> @RafaelCintron's position https://github.com/webmachinelearning/webnn/issues/3#issuecomment-457886464 14:35:48 ... I observed one question Rafael asked from Nikhil is unanswered, maybe we can discuss it now? Specifically: 14:36:00 "An operator-style API should allow web developers to build their graph using Javascript function calls. Once the graph is in place, graph-rewriting and fusing of operations like you describe could be possible by the user agent. Am I missing something or do you have a different notion of what an operator-style API entails than I do?" 14:36:32 -> quote from @RafaelCintron's comment https://github.com/webmachinelearning/webnn/issues/3#issuecomment-454250090 14:37:39 dsmilkov: I like the idea that you could execute a single operation, or a chain of operation, if the API allows makes it much more flexible 14:37:55 ... the reason why more flexible, the model you try to exec need to be done in user code 14:38:11 ... being able to exec some using built-in ops and go back and forth from user code 14:38:44 Rama has joined #webmachinelearning 14:41:01 [ agreement that there's a common ground, follow up on GH ] 14:41:18 TOPIC: Eager and graph execution mode mapping to native 14:41:26 anssik: next subtopic, implications of eager and graph execution mode mapping 14:41:37 -> eager and graph execution mode mapping https://github.com/webmachinelearning/webnn/issues/3#issuecomment-459755470 14:42:35 NingxinHu: related to the previous discussion, single operation is eager execution, raphael's graph is execution model 14:43:31 [ NingxinHu describing the findings documented in the mapping table ] 14:45:03 +1 to new issue 14:45:12 Too much discussion on #3 14:46:53 dsmilkov: two discussion: 1) exec of ops 2) exec models 14:47:26 s/discussions/issues to be created/ 14:48:38 TOPIC: Splitting the mega issue 14:49:50 proposed RESOLUTION: split #3 into "Executions ops" and "Execution models" 14:49:58 [ agreement ] 14:50:03 RESOLUTION: split #3 into "Executions ops" and "Execution models" 14:50:12 TOPIC: Custom operations 14:50:28 anssik: Daniel Smilkov raised this issue on our previous call and kindly opened an issue for it so we can follow up. I see positions for and against. 14:50:49 ... I believe everyone would be fine if we'd commence with some of the investigation and do a decision after we're more informed of the problem space. 14:51:11 ... I heard two possible investigation tasks: 1) browser implementation feasibility of built-in ops, 2) survey ops needed to write custom ops 14:52:13 RafaelCintron: OK to investigate, but could ship v1 w/o them and defer to v2 14:53:07 dsmilkov: I wasn't clear when filed the issue that for TF fast data exchange crucial 14:54:46 NingxinHu: I think if it helps I can investigate as mentioned in the issue, possible investigation is for the CPU side, since based on our current setup 14:55:30 ... specifically, investigate Wasm implementation path with CPU backend on macOS or Linux 14:56:32 ... as for the WebGPU and TF WebGL backend, is it reasonable to investigate WebGL exchange with MPS etc? 14:57:36 nick: we've discussed this on our end, op level accelerated endpoints, curious how the spec falls down to multiple long tail devices 14:57:49 s/nick/kreeger/ 14:58:13 kreeger: open-ended questions on how ops are accelerated on specific devices 14:59:10 RafaelCintron: browser could take hints 15:01:15 re: model loading in browser - lots of work for browser vendors to maintain - questions about quantization 15:01:58 I'd be very grateful to hear about the limitations of NNAPI as briefly described by Nikhil 15:02:31 as part of the initial WebML API design was based on devices declaring capabilities inspired in the NNAPI design 15:03:41 TOPIC: Any other business 15:03:55 [ hearing nothing ] 15:03:59 Thanks! We're Adjourned. Our next monthly call is scheduled 14 March 2019. 15:04:09 RRSAgent, draft minutes v2 15:04:09 I have made the request to generate https://www.w3.org/2019/02/14-webmachinelearning-minutes.html anssik 15:04:44 belem has left #webmachinelearning 15:06:23 s/Executions ops/Executing ops/ 15:06:42 s/Execution models/Executing models/ 15:06:44 RRSAgent, draft minutes v2 15:06:45 I have made the request to generate https://www.w3.org/2019/02/14-webmachinelearning-minutes.html anssik 15:12:03 s/two discussion:/two issues:/ 15:12:06 RRSAgent, draft minutes v2 15:12:06 I have made the request to generate https://www.w3.org/2019/02/14-webmachinelearning-minutes.html anssik 15:12:35 s/exec models/exec of models/ 15:12:38 RRSAgent, draft minutes v2 15:12:38 I have made the request to generate https://www.w3.org/2019/02/14-webmachinelearning-minutes.html anssik 15:13:46 s/exchange crucial/exchange is crucial/ 15:13:50 RRSAgent, draft minutes v2 15:13:50 I have made the request to generate https://www.w3.org/2019/02/14-webmachinelearning-minutes.html anssik 15:14:29 s/since based on/based on/ 15:14:30 RRSAgent, draft minutes v2 15:14:30 I have made the request to generate https://www.w3.org/2019/02/14-webmachinelearning-minutes.html anssik 16:38:04 Zakim has left #webmachinelearning 16:59:37 myles has joined #webmachinelearning 17:04:39 myles has joined #webmachinelearning 18:06:43 myles has joined #webmachinelearning 19:23:59 zolkis has joined #webmachinelearning 20:01:16 zolkis has joined #webmachinelearning