13:56:11 RRSAgent has joined #webmachinelearning 13:56:11 logging to https://www.w3.org/2021/10/14-webmachinelearning-irc 13:56:13 RRSAgent, make logs Public 13:56:15 please title this meeting ("meeting: ..."), anssik 13:56:21 Meeting: WebML WG Teleconference – 14 Oct 2021 13:56:21 Chair: Anssi 13:56:26 Agenda: https://github.com/webmachinelearning/meetings/blob/master/telcons/2021-10-14-agenda.md 13:56:31 Scribe: Anssi 13:56:35 scribeNick: anssik 13:56:40 Present+ Anssi_Kostiainen 13:56:52 RRSAgent, draft minutes 13:56:52 I have made the request to generate https://www.w3.org/2021/10/14-webmachinelearning-minutes.html anssik 13:57:16 Regrets+ 14:00:11 ningxin_hu has joined #webmachinelearning 14:01:33 Present+ Ningxin_Hu 14:01:40 Present+ Chai_Chaoweeraprasit 14:01:44 Present+ Ganesan_Ramalingam 14:02:15 Present+ Rafael_Cintron 14:02:41 rama has joined #webmachinelearning 14:04:56 Topic: WebNN API recent new feature requests 14:05:05 Subtopic: Autograd / JAX proposal for future work 14:05:13 -> https://github.com/webmachinelearning/webnn/issues/218 issue #218 14:05:36 RafaelCintron has joined #webmachinelearning 14:05:47 anssik: JAX is a research-focused framework based on composable function transformations 14:05:53 ... "JAX is Autograd and XLA, brought together for high-performance machine learning research" 14:06:25 ... it combines an updated version of Autograd (standalone Autograd is no longer developed) and XLA 14:06:40 Chai has joined #webmachinelearning 14:06:55 ... Autograd is an automatic differentiation engine that takes a computational graph, and calculates input gradients 14:07:56 ... XLA aka "Accelerated Linear Algebra" is the underlying domain-specific compiler that given a graph, compiles it into a sequence of computation kernels 14:07:56 ... this enables model-specific optimization by fusing multiple computation kernels into one 14:08:39 -> https://github.com/webmachinelearning/proposals Proposals repo for future work 14:08:54 anssik: I think a path forward could be to document this proposed future work in our proposals repo 14:09:30 Rama: I have to understand the goal, Autograd makes only sense in training, so in current WG scope it is out of scope 14:10:18 Present+ Ping_Yu 14:11:30 Chai: if we ask what is the future of training on the web, the trend is probably federated learning distributing training across multiple machines over the internet 14:11:42 ... each with different datasets 14:11:56 ... some ideas that the web is probably the best way realize this 14:12:54 ... I can see that at some point, when machine learning on the browser becomes more mainstream, federated learning could make sense 14:13:21 ... DML started with supporting inference, now we also provide backends for TF and other training frameworks 14:13:36 ... I can see that in a few years think could be considered 14:14:33 Jonathan has joined #webmachinelearning 14:14:45 Present+ Jonathan_Bingham 14:15:04 right, JAX 14:16:30 Jonathan: I'm happy to connect with Nikhil and Daniel 14:16:44 Ping: I have been working to get JAX models run for inference on top of TF.js 14:16:56 ... most ops are compatible with TF, 10 ops are more generic 14:17:06 ... and not available in TF and need to be reimplemented 14:17:29 ... Daniel and Nikhil, Ningxin did early exploration composability 14:17:57 ... we should perhaps think JAX-inspired design again, since its getting more popular 14:18:05 ... also looking at federated learning 14:18:27 ... at this moment training uses forward and backwards pass in JAX 14:18:50 ... TF has its own type to compute gradient backwards, it is different 14:19:07 ... given where WebNN is going it is possible to support grad ops or function 14:20:52 Present+ Rachel_Yaget 14:21:41 Topic: TPAC planning 14:22:00 Ping_yu has joined #webmachinelearning 14:22:23 anssik: I'd like to use the rest of the call to review our WebML WG Virtual Meeting sessions 14:22:32 -> https://github.com/webmachinelearning/meetings/issues/18 WebML WG Virtual Meetings at TPAC 2021 14:22:52 anssik: to make the TPAC discussions productive, please do not hesitate to self-nominate yourself as a contributor to any sessions of interest and share relevant resources for background reading 14:22:59 ... see "Contributors" in the meeting agenda for current volunteers 14:24:47 ... but to facilitate discussion on the topic, we'd like to use our TPAC meeting time to have a bit more open discussion on higher-level topics than we usual do on our bi-weekly calls that focus on making progress with WebNN API spec and discuss its details 14:25:00 ... I'll go through our session one by one to solicit your feedback 14:25:39 i did get the invites 14:26:19 Subtopic: Rationale/criteria for adding new ops to the WebNN API 14:26:29 anssik: "Discuss what makes for a good criteria to ensure the WebNN API evolves and its scope is driven by right priorities, identify and distill key learnings from related work." 14:26:37 q+ 14:27:32 ... in our current work mode, we evolve WebNN API and manage its scope based on use cases informed by customer (primarily ML JS frameworks) needs, validated by running code experiments (including polyfillable samples) and cross-platform implementation experience e.g. webnn-native 14:27:37 q? 14:27:54 Chai: is this more of a discussion session? 14:29:59 anssik: open discussion 14:30:21 Chai: I'm not closely working with ONNX, but can share perspective on big and small ops 14:30:38 ... if someone working closely on ONNX community would be good, maybe Rama might know? 14:31:29 Chai: in the past I had my opinion from the point of view of implementation of ONNX, experience I use when we work on adding WebNN ops 14:31:35 ... I can help in any way useful 14:31:49 ... for ONNX expert, we should go back and find that person 14:32:54 Google has added lots of ops to TF, TFlite and its other projects, and has plenty of experience and learnings from adding things we shouldn't have ;-) 14:34:03 Rachel: I want to suggest a general discussion on computational intelligence 14:34:27 anssik: please add a comment to the agenda https://github.com/webmachinelearning/meetings/issues/18 14:35:34 anssik: can you check if someone from Google would like to join this session? 14:35:42 Jonathan: we'll check with Ping 14:36:36 Subtopic: Versioning and web compatibility 14:36:55 Rachel has joined #webmachinelearning 14:36:55 anssik: for this talk, we're expecting W3C TAG to share insights on the general web platform story 14:37:24 ... the simple version of the story is that the web technologies are not versioned explicitly but rely on feature detection mechanisms to enable sort of defensive programming 14:37:40 ... the ML domain had specific versioning requirements 14:37:48 -> https://github.com/onnx/onnx/blob/master/docs/Versioning.md ONNX Versioning 14:38:07 ... e.g. using ONNX as an example, there are three classes of versioned entities: Intermediate Representation version, operator version (for ONNX graphs), and model version 14:38:28 anssik: we'd benefit from someone to sharing learnings from the native ML-land with respect to what works for versioning 14:39:40 anssik: any reactions? 14:40:01 Subtopic: Privacy and security discussion 14:40:06 anssik: "Revisit privacy discussion and related work happening in the WebGPU WG, discuss fingerprinting guidance from PING and evaluate its mitigations." 14:40:32 ... PING and WebGPU editors and chair invited 14:41:26 ... I encourage folks to review the references ahead the meeting, good meterial: 14:41:31 -> https://www.w3.org/TR/fingerprinting-guidance/ Fingerprinting Guidance 14:41:40 -> https://gpuweb.github.io/gpuweb/#security WebGPU Security and Privacy Considerations 14:42:18 anssik: we also have made a commitment all out specs contain a section detailing all known security and privacy implications for implementers 14:42:37 ... we're off to a good start here, and I think WebGPU has done great work documenting these considerations 14:42:43 -> https://www.w3.org/TR/webnn/#security WebNN Security and Privacy Considerations 14:42:56 -> https://github.com/webmachinelearning/webnn/labels/privacy-tracker WebNN open privacy issues 14:43:54 anssik: any other folks we should invite to this discussion on security and privacy? 14:44:21 Subtopic: ML JS framework performance, focus areas for WebNN 14:44:35 anssik: "Review ML JS framework performance data across backends, discuss learnings from backend implementation efforts." 14:44:53 ... I know Ningxin & co are working on performance data, we hope to have it available in time for our meeting. 14:45:00 ... I'd like to make sure we'd have folks for key ML JS frameworks folks around. 14:46:16 ningxin_hu: I can share what progress we've made so far 14:46:40 ... we have webnn-native with DML on Windows and OpenVINO on Linux, have a prototype of TFLite Web WebNN delegate 14:46:51 ... also ONNXRuntime Web execution provider for WebNN 14:47:20 ... we have prototypes for those frameworks, also as shared before we have GSoC project OpenCV.js with WebNN backend 14:48:17 ... we plan to collect data across Windows and Linux what webnn-native supports, we don't yet have browser integration so use Node.js binding to run Electron apps with WebNN 14:49:38 ... good to invite people to review the data, we compare WebNN with Wasm, WebGL, maybe people from those groups are interested? 14:50:23 anssik: we can reach out to Wasm folks 14:51:08 Ping: also good to have Wasm and WebGPU folks in this session 14:52:16 ... need to note the perf data is without browser integration, security implications might slow things down 14:52:33 q? 14:52:42 ack Chai 14:54:25 ningxin_hu: working with ONNX Runtime Web to collect perf numbers, my ask is can we have ONNX Runtime Web folks these to have a good discussion and identify gaps we need to fill for the integration 14:54:34 Chai: we can invite Emma for this discussion 14:54:54 anssik: Emma Ning 14:55:24 q? 14:55:49 Subtopic: Integrating an open-source cross-platform implementation of the Web Neural Network API into a web engine 14:56:08 anssik: Ningxin, any opens to address for this TPAC sessions? 14:56:55 ningxin_hu: please extend the invitation to @BruceDai 14:57:30 ... we're using Dawn infra for IPC multi-process in Chromium 14:57:45 ... we're working on design doc and will share the material before the meeting 14:58:20 ... Corentin from WebGPU was interested and plans to join 14:58:46 ... Edge is another important browser, maybe Rafael has interest to join? 14:58:53 I will definitely be there, as well as for all the other sessions. 14:59:29 s/@BruceDai// 15:00:24 ningxin_hu: regarding implementation, security is an important topic for browser implementation and hardenings have performance impact 15:00:53 ... it is important to understand how sandboxing influences the implementation 15:01:25 q? 15:01:39 Subtopic: How should WebNN API deal with fusion 15:01:46 This session is looking for interested contributors, we'll drop this if no volunteers. 15:02:00 s/This session is looking for interested contributors, we'll drop this if no volunteers./anssik: This session is looking for interested contributors, we'll drop this if no volunteers./ 15:02:34 anssik: any interest for this topic? 15:02:39 ... maybe we'll drop this 15:03:02 Subtopic: Conformance testing of WebNN API 15:03:19 anssik: "Discuss the progress in conformance testing of the WebNN API, web-platform-tests migration." 15:03:31 ... @BruceDai has work in progress to convert the tests to w-p-t format 15:04:06 ... I found our workshop session on conformance insightful, is there value in revisiting that discussion at TPAC? Any new information? 15:04:26 Chai: this is important topic 15:04:33 ... due to hardware diversity 15:05:01 ... if we want WebNN to work across hardware architectures, interoperability is important 15:05:15 ... I have seen Bruce's PR, great work using w-p-t 15:05:35 ... I put my comments in there, because I was a bit worried how w-p-t support is in place today 15:05:52 ... when it comes to ensuring hw conformance, we may need to change it, to make it slightly better 15:06:09 ... my comments in the PR, with some examples of methodology we use 15:06:44 ... what we've seen in the past, we may run in the risk of not having the right behavior, or misleading results 15:07:57 ... we probably need some additional work here for WebNN 15:08:47 q? 15:08:54 Subtopic: Ethical issues in using Machine Learning on the Web 15:09:20 anssik: The Working Group is committed to develop a Working Group Note documenting ethical issues associated with using Machine Learning on the Web. 15:09:35 ... I put together a first stab that could become a W3C Note 15:09:40 -> https://webmachinelearning.github.io/ethical-webmachinelearning/ Ethical Web Machine Learning 15:09:56 anssik: I'm not an expert in ethics, so welcome someone with a passion on this topic to help out 15:10:20 ... with Dom we've reached out to some people, but need your help make those connections within your companies 15:10:37 ... based on what I learned reviewing the meta-studies, it seems almost every company has an initiative in the space of ethical principles given the importance of these issues 15:11:16 q? 15:12:23 Rachel: I'm very interested in this, will come back with suggested speakers 15:13:20 q? 15:14:18 RRSAgent, draft minutes 15:14:18 I have made the request to generate https://www.w3.org/2021/10/14-webmachinelearning-minutes.html anssik