14:19:21 <RRSAgent> RRSAgent has joined #webmachinelearning
14:19:21 <RRSAgent> logging to https://www.w3.org/2019/03/14-webmachinelearning-irc
14:19:26 <Zakim> Zakim has joined #webmachinelearning
14:19:32 <anssik> RRSAgent, make logs public
14:19:38 <anssik> Meeting: WebML CG Teleconference – 14 March 2019
14:19:45 <anssik> Chair: Anssi
14:19:50 <anssik> Scribe: Anssi
14:19:55 <anssik> Agenda: https://github.com/webmachinelearning/meetings/blob/master/telcons/2019-03-14-agenda.md
14:20:06 <anssik> scribeNick: anssik
14:24:41 <gregwhitworth> present+ - on cellphone so I can't help scribe :(
14:24:53 <gregwhitworth> Sorry anssik
14:25:12 <tomoyuki> tomoyuki has joined #webmachinelearning
14:26:46 <anssik> Present+ Anssi_Kostiainen
14:27:08 <anssik> Present+ Rafael_Cintron
14:27:14 <anssik> Present+ Tomoyuki_Shimizu
14:27:34 <anssik> Present+ Greg_Whitworth
14:27:48 <anssik> Present+ Daniel_Smilkov
14:28:03 <Rafael> Rafael has joined #webmachinelearning
14:29:21 <Rama> Rama has joined #webmachinelearning
14:30:38 <ningxinhu> ningxinhu has joined #webmachinelearning
14:30:47 <anssik> Present+ Ningxin_Hu
14:30:56 <belem> belem has joined #webmachinelearning
14:31:45 <anssik> Present+ Belem_Zhang
14:32:12 <Rama> Present+ Ganesan_Ramalingam
14:32:22 <kreeger> kreeger has joined #webmachinelearning
14:32:23 <anssik> Present+ Paul_McDaniel
14:32:32 <anssik> Present+ Daniel_Smilkov
14:32:41 <anssik> Present+ Nick_Kreeger
14:33:09 <anssik> RRSAgent, draft minutes v2
14:33:09 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
14:33:49 <anssik> anssik: We have three topics on the agenda: executing operations, custom operations, industry events
14:33:49 <paulmcdaniel> paulmcdaniel has joined #webmachinelearning
14:33:55 <Nikhil> Nikhil has joined #webmachinelearning
14:33:56 <anssik> ... let's spend as much we need for the first two, time allowing discuss the last one
14:34:17 <anssik> TOPIC: Executing operations
14:34:23 <anssik> -> Executing operations https://github.com/webmachinelearning/webnn/issues/11
14:34:55 <anssik> anssik: On our path to API spec we should now derive requirements that will inform the design of the API for executing operations.
14:35:03 <tomoyuki> tomoyuki has joined #webmachinelearning
14:35:13 <anssik> ... since our last monthly sync:
14:35:30 <anssik> ... Ningxin provided a summary of platform APIs and their supported executing modes
14:36:00 <anssik> Present+ Nikhil_Thorat
14:36:07 <anssik> anssik: Nikhil proposed that for platform APIs (e.g. NNAPI) that need a graph we use an implicit graph that allow decorating functions in order to convert them to subgraphs.
14:36:35 <anssik> anssik: Rafael noted the API should enable the use of specialized ML hardware.
14:36:48 <anssik> anssik: Let's discuss. The expected outcome of this issue would be a list of requirements we can them feed into the spec to use as a checklist as we advance to the API definition phase
14:38:37 <anssik> [ revisiting discussion in the issue ]
14:47:39 <paulmcdaniel> q+
14:51:25 <anssik> ack paulmcdaniel
14:52:57 <ningxinhu> q+
14:54:01 <anssik> ack ningxinhu
14:54:19 <Daniel_Smilkov> If TF.js, operations like neg() and square() would return a logical wrapper around the resulting data, but not the data. When user calls result.data(), we transfer the data back to user-code
14:56:14 <paulmcdaniel> this is paul, i would like to propose we capture this proposal with some more details in the github issue and also iterate there.  i'm not sure this will work or not for our hardware acceleration
14:57:14 <paulmcdaniel> it might work great, but i have questions and can also show what we require via pictures/IDL in the github iteration
14:57:58 <anssik> RRSAgent, draft minutes v2
14:57:58 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
14:58:42 <anssik> s/present+ - on cellphone so I can't help scribe :(//
15:01:33 <anssik> TOPIC: Custom operations
15:01:41 <anssik> -> Custom operations https://github.com/webmachinelearning/webnn/issues/6
15:01:43 <ningxinhu> https://docs.google.com/presentation/d/1xOQtyAeQNdI70rICzLNKRxmzM-Dwjhh3MdzGKW29C8Q/edit?usp=sharing
15:01:53 <anssik> anssik: Ningxin volunteered to investigate custom operations support on CPU and will give an interim update specifically for the Wasm backend
15:01:59 <anssik> -> WebNN Investigation: Wasm Custom Ops (presentation slides) https://docs.google.com/presentation/d/1xOQtyAeQNdI70rICzLNKRxmzM-Dwjhh3MdzGKW29C8Q/
15:02:04 <anssik> [ Ningxin presents the slides ]
15:02:29 <anssik> anssik: The slides will give a recap on custom operations, WebNN POC API, perf data, summarize key findings.
15:02:38 <anssik> Ningxin_Hu: Background: in ML field, the ops are evolving quickly
15:02:49 <anssik> ... Problem: big framework’s ops set (e.g. TF.js has ~200) VS. small WebNN’s ops set (focusing on hardware acceleration)
15:03:25 <anssik> ... Status: frameworks ops are implemented in WebGL and WASM, later will add WebGPU
15:03:53 <anssik> ... Solutions under investigation:
15:04:05 <anssik> ... - Framework partitions the NN graph based on WebNN’s ops set
15:04:23 <anssik> ... - Framework offloads supported sub-graphs to WebNN execution
15:04:29 <anssik> ... - Framework executes other sub-graphs by own ops
15:04:35 <anssik> ... - Require a high-performance data exchange between framework ops and WebNN
15:04:46 <anssik> ... - Device affinity may matter: WASM ops interop with WebNN/CPU, WebGL/WebGPU ops interop with WebNN/GPU
15:04:51 <anssik> ... - Start with WASM ops
15:05:59 <anssik> [ Using WebNN POC slide walkthrough ]
15:06:38 <anssik> [ WASM ops and WebNN graph execution ]
15:08:17 <anssik> [ Workload: MobileNet V1 ]
15:09:04 <anssik> [ Graph partition configurations ]
15:09:17 <paulmcdaniel> question on the IRC:    these slides are great !  thank you ningxin !   is the POC source code link somewhere?    would love to see how some of the ops where connected to BNNS.
15:10:52 <anssik> [ Results: performance ]
15:11:19 <anssik> [ Results: efficiency ]
15:13:55 <ll> ll has joined #webmachinelearning
15:14:01 <anssik> RRSAgent, draft minutes v2
15:14:01 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
15:15:56 <anssik> Ningin_Hu: Summary:
15:16:01 <anssik> ... Offloading key WASM ops to WebNN can get good speedup
15:16:08 <anssik> ... Interop between WASM ops and WebNN graph has overhead
15:16:23 <anssik> ... Need to explore the browser-side optimizations, e.g. by
15:16:25 <gregwhitworth> I have to go catch my shuttle. I'll follow up on GH.
15:16:39 <anssik> ... Reduce memory copies: CPU backend inside WASM process? Shared WASM heap?
15:17:05 <anssik> ... Reduce memory relayout: expose native memory layout?
15:17:38 <anssik> ... Need to explore the framework-side optimizations, e.g. by implementing overhead-aware graph partition and delegate algorithm
15:17:55 <anssik> anssik: thanks you Ningxin, there's a question from paul
15:18:22 <jdarpinian> jdarpinian has joined #webmachinelearning
15:18:51 <anssik> ningxinhu: the source will be published soon
15:18:52 <paulmcdaniel> thanks !
15:21:12 <anssik> TOPIC: Industry events
15:21:23 <anssik> [ skipping in the interest time ]
15:22:57 <anssik> TOPIC: Adjourn
15:23:04 <anssik> anssik: thank you for your active participation.
15:23:11 <anssik> ... Next call is scheduled 11 April 2019, so we're back to the original time, i.e. US/Canada an hour earlier than today.
15:23:17 <anssik> RRSAgent, draft minutes v2
15:23:17 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
15:23:56 <Daniel_Smilkov> Have a nice day!
15:24:10 <belem> belem has left #webmachinelearning
15:24:39 <anssik> s/skipping in the interest time/skipping in the interest of time/
15:24:45 <anssik> RRSAgent, draft minutes v2
15:24:45 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
15:25:18 <anssik> s/thanks you/thank you/
15:29:39 <anssik> RRSAgent, draft minutes v2
15:29:39 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
15:44:48 <anssik> s/Sorry anssik//
15:44:50 <anssik> RRSAgent, draft minutes v2
15:44:50 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
15:46:36 <anssik> s/-> Executing operations https://‌github.com/‌webmachinelearning/‌webnn/‌issues/‌11/-> https://‌github.com/‌webmachinelearning/‌webnn/‌issues/‌11 Executing operations
15:46:40 <anssik> RRSAgent, draft minutes v2
15:46:40 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
15:47:23 <anssik> s//‌github.com/‌webmachinelearning/‌webnn/‌issues/‌11/Executing operations//github.com/webmachinelearning/webnn/issues/11/-> https://‌github.com/‌webmachinelearning/‌webnn/‌issues/‌11 Executing operations
15:47:43 <anssik> RRSAgent, draft minutes v2
15:47:43 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
15:48:30 <anssik> s//‌github.com/‌webmachinelearning/‌webnn/‌issues/‌11///
15:48:31 <anssik> RRSAgent, draft minutes v2
15:48:31 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
15:51:24 <anssik> s/-> https://‌github.com/‌webmachinelearning/‌webnn/‌issues/‌11 Executing operations/https://github.com/webmachinelearning/webnn/issues/11
15:51:29 <anssik> RRSAgent, draft minutes v2
15:51:29 <RRSAgent> I have made the request to generate https://www.w3.org/2019/03/14-webmachinelearning-minutes.html anssik
17:14:54 <Zakim> Zakim has left #webmachinelearning
19:22:34 <zolkis> zolkis has joined #webmachinelearning
19:40:00 <zolkis> zolkis has joined #webmachinelearning
19:57:49 <myles> myles has joined #webmachinelearning
22:23:41 <zolkis> zolkis has joined #webmachinelearning