13:51:38 <RRSAgent> RRSAgent has joined #webmachinelearning
13:51:38 <RRSAgent> logging to https://www.w3.org/2019/06/06-webmachinelearning-irc
13:51:45 <Zakim> Zakim has joined #webmachinelearning
13:51:49 <anssik> RRSAgent, make logs public
13:51:56 <anssik> Meeting: WebML CG Teleconference – 6 June 2019
13:52:02 <anssik> Chair: Anssi
13:52:06 <anssik> Agenda: https://github.com/webmachinelearning/meetings/blob/master/telcons/2019-06-06-agenda.md
13:58:01 <ningxin_hu> ningxin_hu has joined #webmachinelearning
13:58:47 <anssik> Scribe: Anssi
13:58:53 <RafaelCintron> RafaelCintron has joined #webmachinelearning
13:58:58 <anssik> scribeNick: anssik
13:59:04 <anssik> Present+ Anssi_Kostiainen
13:59:08 <anssik> Present+ Rafael_Cintron
13:59:10 <Rama> Rama has joined #webmachinelearning
13:59:16 <anssik> Present+ Ganesan_Ramalingam
13:59:33 <anssik> RRSAgent, draft minutes v2
13:59:33 <RRSAgent> I have made the request to generate https://www.w3.org/2019/06/06-webmachinelearning-minutes.html anssik
14:00:06 <ningxin_hu> Present+ Ningxin_Hu
14:00:14 <tomayac> tomayac has joined #webmachinelearning
14:00:37 <tomayac> Hi there!
14:00:47 <anssik> Present+ Thomas_Steiner
14:01:42 <Jonathan> Jonathan has joined #webmachinelearning
14:03:01 <anssik> TOPIC: Survey graph-building APIs from native ecosystem
14:05:17 <anssik> ningxin_hu: first input on nGraph API, in Paul's comments nGraph is one exec provider
14:05:37 <anssik> ... approach applied to try write as simple example code as possible
14:06:41 <anssik> ... for WebNN vs. other native graph APIs comparison, reproducing the same example graph
14:07:56 <anssik> ... highlights of evaluation: nGraph API uses no enums to represent op types, it uses constructors of different classes
14:08:32 <anssik> ... notably, no addOperand call similar to WebNN POC
14:10:21 <anssik> ... data provided by user code as a param at execution time
14:10:41 <anssik> -> https://github.com/webmachinelearning/webnn/issues/20 Survey graph-building APIs from native ecosystem
14:13:48 <anssik> [ningxin_hu explains the details of the example https://github.com/webmachinelearning/webnn/issues/20#issuecomment-496838734]
14:14:34 <anssik> Rama: does nGraph has a notion of a subgraph used for control flow constructs?
14:14:42 <anssik> s/has a/have a/
14:15:01 <anssik> ningxin_hu: has a concept of function, not sure about nesting, will investigate and come back
14:17:13 <anssik> anssik: is the nGraph more expressive than WebNN API?
14:18:26 <anssik> ningxin_hu: question is more about developer ergonomics, WebNN API is a C-style API so lacks object-oriented style
14:20:41 <anssik> ningxin_hu: MPSNNGraph investigation done by my colleague
14:21:02 <anssik> ... he is involved with WebNN POC implementation work, did MPS part of that
14:21:27 <anssik> ... happy to share his experience, also welcome Apple reps' insights
14:22:07 <anssik> ... highlight comparing with nGraph: no concept to distinguish between constants and inputs
14:23:00 <anssik> [ ningxin_hu discussing example https://github.com/fujunwei/MPSNNGraph/blob/simple/MPSNNGraphSample/main.mm]
14:23:23 <anssik> ... on API level not much diff between constants and inputs
14:24:43 <anssik> ... the API uses different class types to distinguish ops, also MPSNNGraph takes images as input
14:25:05 <anssik> ... nGraph OTOH has a helper for the same task
14:25:43 <anssik> ... MPSNNGraph has a class to create a graph, you provide result image as a graph output
14:26:21 <anssik> ... would like to get Apple reps insight on the MPSNNGraph
14:27:37 <anssik> RafaelCintron: talking on behalf of paul
14:27:56 <anssik> ... ONNX is an open source standard, standards body of the same name
14:28:22 <anssik> ... ONNX runtime is an OSS project, implementation of the ONNX format
14:28:50 <anssik> ... execution providers run the models, types for CPU, DirectML with compute shaders
14:29:27 <anssik> anssik: shipping status?
14:29:46 <anssik> RafaelCintron: DirectML shipping in upcoming Windows release
14:29:52 <anssik> ... both public APIs
14:30:10 <anssik> ... ONNX runtime has a CPU exec model only as of now
14:30:57 <anssik> ningxin_hu: ONNX Runtime's main functionality to load the model and do the inference, is that correct?
14:31:01 <anssik> RafaelCintron: correct
14:31:20 <anssik> ningxin_hu: in Paul's comments, he provided a graph and edge reference, how that works with exec providers
14:31:48 <anssik> ... question re the header file, is this an API for developers to call into, or an internal API for execution boundary
14:32:19 <anssik> RafaelCintron: API that an exec provider would call into
14:32:48 <anssik> ningxin_hu: to clarify, it is an API for an exec provider to use, not an application developer
14:32:54 <anssik> RafaelCintron: correct
14:33:24 <anssik> anssik: are there any web bindings to these APIs?
14:33:28 <anssik> RafaelCintron: not in browsers no
14:35:11 <anssik> Present+ Kai_Ninomiya
14:35:21 <anssik> Present+ Jonathan_Bingham
14:35:28 <anssik> Present+ Gabe_Esteven
14:35:45 <anssik> RRSAgent, draft minutes v2
14:35:45 <RRSAgent> I have made the request to generate https://www.w3.org/2019/06/06-webmachinelearning-minutes.html anssik
14:36:36 <tomayac> Sorry, I don't know the answer
14:40:24 <ningxin_hu> Android NNAPI has a C++ wrapper which should simply the code to use NN functionalities. My question is what's the status of that? Any code example available to compare with existing WebNN API (derived from NN C API)?
14:41:24 <RafaelCintron> I agree with Anssi said earlier that we should combine all of the APIs into one.
14:41:51 <ningxin_hu> The link of NNAPI C++ wrapper: https://android.googlesource.com/platform/frameworks/ml/+/refs/heads/master/nn/runtime/include/NeuralNetworksWrapper.h
14:44:08 <anssik> ningxin_hu: based on the survey, my take is maybe we can focus on the aspects of the API itself, functionality of the APIs one by one
14:44:31 <anssik> ... for each we can refer to each native API feature, and see what to absorb for the WebNN API improvement
14:44:55 <anssik> ... goals to simplify web developers work, follow web API design conventions
14:46:12 <anssik> ... propose continue discussion in that issue https://github.com/webmachinelearning/webnn/issues/16
14:46:36 <anssik> a/that//
14:46:42 <anssik> s/that//
14:47:19 <anssik> ningxin_hu: in #16 there are already multiple design points, maybe split to focus discussion, then map one PR to one issue
14:48:28 <anssik> TOPIC: Graph-building syntax simpler for web developers
14:48:33 <anssik> -> https://github.com/webmachinelearning/webnn/issues/16  Graph-building syntax simpler for web developers
14:49:07 <anssik> ningxin_hu: we should summarize discussion points in #16 and create separate issues for further discussion and eventual resolution and PRs
14:49:20 <anssik> Rama: support splitting where it makes sense
14:49:56 <anssik> anssik: please open issues as you see fit
14:50:40 <anssik> ningxin_hu: I will summarize my position in that issues
14:51:26 <anssik> TOPIC: Call schedule for the upcoming summer
14:51:41 <anssik> anssik: unable to chair 20 June 2019, July holiday period, next telcons:
14:51:48 <anssik> 27 June 2019
14:52:01 <anssik> 15 August 2019
14:52:33 <anssik> https://www.w3.org/2019/09/TPAC/schedule.html
14:52:50 <anssik> WebML CG has F2F Tuesday 17 September https://www.w3.org/2019/09/TPAC/schedule.html#Tuesday
14:54:01 <anssik> TOPIC: Adjust
14:54:23 <anssik> RRSAgent, draft minutes v2
14:54:23 <RRSAgent> I have made the request to generate https://www.w3.org/2019/06/06-webmachinelearning-minutes.html anssik
17:09:36 <Zakim> Zakim has left #webmachinelearning