15:07:36 <RRSAgent> RRSAgent has joined #webmachinelearning
15:07:36 <RRSAgent> logging to https://www.w3.org/2020/01/09-webmachinelearning-irc
15:07:42 <Zakim> Zakim has joined #webmachinelearning
15:07:45 <anssik> RRSAgent, make logs public
15:07:50 <anssik> Meeting: WebML CG Teleconference – 9 January 2020
15:07:55 <anssik> Chair: Anssi
15:08:12 <anssik> Agenda: https://github.com/webmachinelearning/meetings/blob/master/telcons/2020-01-09-agenda.md
15:08:20 <anssik> Scribe: Anssi
15:08:25 <anssik> scribeNick: anssik
15:08:26 <Jonathan> Jonathan has joined #webmachinelearning
15:08:31 <anssik> Present+ Anssi_Kostiainen, Rafael_Cintron, Ningxin_Hu
15:08:49 <anssik> Present+ Jonathan_Bingham
15:08:54 <tomayac> Present+ Thomas_Steiner
15:09:03 <anssik> Present+ Gabe_Esteven
15:09:57 <anssik> Present+ Chai_Microsoft
15:10:31 <anssik> RRSAgent, draft minutes v2
15:10:31 <RRSAgent> I have made the request to generate https://www.w3.org/2020/01/09-webmachinelearning-minutes.html anssik
15:11:19 <anssik> RRSAgent, make logs public
15:11:40 <baul> Present+ Baul_Eun
15:11:51 <anssik> TOPIC: Buffer sharing between GPU and ML accelerator, update & next steps
15:11:53 <anssik> RRSAgent, draft minutes v2
15:11:53 <RRSAgent> I have made the request to generate https://www.w3.org/2020/01/09-webmachinelearning-minutes.html anssik
15:12:33 <anssik> -> https://github.com/webmachinelearning/webnn/issues/33#issuecomment-563913379 #33 investigation update and open questions
15:15:02 <jdarpinian> jdarpinian has joined #webmachinelearning
15:20:42 <anssik> ningxin_hu: the next step of the investigation is to run Test3 on programmable ML accelerator, e.g. VPU
15:21:12 <anssik> ... identified 3 issues that block us:
15:21:32 <anssik> Can VPU run HLSL?
15:21:32 <anssik> Can WebGPU support compute-only device (e.g. VPU) and run computer shader on it?
15:21:39 <anssik> Can WebGPU/D3D12 share buffer with WebNN/DML on VPU? (run Test3 on VPU)
15:22:18 <anssik> want to understand whether this is a driver issue, what is the root cause
15:22:42 <anssik> Present+ James_Darpinian
15:23:49 <anssik> ningxin_hu: would like to hear other participants input on this issue, now focus on VPU and DirectML
15:24:07 <anssik> ... e.g. feedback from Android NN API or other accelerators
15:24:59 <anssik> ... how about using TF Lite as a platform for this investigation?
15:25:18 <anssik> Jonathan: for TF Lite, probably yes, need to follow up with Nikhil and Daniel to follow up
15:25:25 <anssik> ... please send an email to them to connect
15:26:10 <anssik> TOPIC: conv2d native API mapping table
15:26:39 <anssik> -> https://github.com/webmachinelearning/webnn/issues/28 [op compatibility] conv2d #28
15:26:45 <anssik> -> https://github.com/webmachinelearning/webnn/blob/master/op_compatibility/conv2d.md conv2d native API mapping table
15:28:02 <anssik> ningxin_hu: Paul McDaniel wanted to contribute ONNX column to the table
15:28:22 <anssik> ... also discussed whether ONNX and TF are on framework level, and leave this for native API mapping only
15:28:58 <anssik> chai: I have discussed this with Rafael, but I haven't looked into it
15:29:21 <anssik> anssik: is adding in ONNX data reasonable?
15:29:39 <anssik> Rafael: we can add ONNX data, it seems fine
15:30:05 <anssik> Chai: DirectML and/or ONNX mapping?
15:30:43 <anssik> Rafael: both would be useful
15:31:55 <anssik> TOPIC: Float16 type support
15:32:06 <anssik> -> https://github.com/webmachinelearning/webnn/issues/26 Float16 type support #26
15:32:23 <anssik> anssik: Benjamin proposed we should add "float16" into supported tensor types
15:32:29 <anssik> ... rationale explainer in the issue #26
15:32:36 <anssik> ... any objection in doing so?
15:32:44 <anssik> ... affected spec surface https://webmachinelearning.github.io/webnn/#enumdef-operandtype
15:34:01 <anssik> Chai: want to point out, not explicit objection, but want to point our Float16 is not supported universally, old and new, many use cases in models depend on Float16, how to run models that depend on this type but hw does no support  it natively
15:35:48 <anssik> ningxin_hu: very good input! If you look at Benjamin's proposal, there's a fallback
15:36:00 <anssik> ... without native support, there's a fallback to Float32
15:36:29 <anssik> Chai: depends on how the model is created, if trained with float16, and you internally substitute higher precision
15:37:08 <anssik> ... it works, but in other cases such as convolution there are issues with accumulated loss, get different result
15:37:39 <anssik> ... if we define an API where fallbacks happen at some level, this should be OK
15:38:41 <anssik> ... internal handling that's opaque may not work for this case, the API should expose "hardware capabilities", or simply fail with a correct answer, or return "supported"
15:41:10 <anssik> proposed RESOLUTION: Add "float16" to OperandType enum in WebNN API and define a way for the API to respond when float16 is not natively supported
15:43:32 <Chai> Chai has joined #webmachinelearning
15:44:16 <anssik> ningxin_hu: sounds good to me
15:44:21 <anssik> RESOLUTION: Add "float16" to OperandType enum in WebNN API and define a way for the API to respond when float16 is not natively supported
15:44:49 <anssik> TOPIC: W3C Workshop on Web & Machine Learning (24-25 March 2020, Berlin): update & next steps
15:45:18 <anssik> 21 February 2020
15:45:18 <anssik> Registration deadline
15:45:18 <anssik> 28 February 2020
15:45:18 <anssik> Acceptance notification
15:45:19 <anssik> 6 March 2020
15:45:19 <anssik> Position statements deadline
15:45:19 <anssik> 13 March 2020
15:45:19 <anssik> Program announced
15:45:20 <anssik> 24-25 March 2020
15:45:20 <anssik> Workshop
15:47:00 <anssik> https://w3c.github.io/machine-learning-workshop/#program
15:54:37 <anssik> TOPIC: Adjourn
15:54:49 <anssik> RRSAgent, draft minutes v2
15:54:49 <RRSAgent> I have made the request to generate https://www.w3.org/2020/01/09-webmachinelearning-minutes.html anssik
17:45:35 <Zakim> Zakim has left #webmachinelearning