Add workshop agenda for Live Sessions #2 and #3

[anssiko]

No description provided.

[anssiko]

Heads up:

If you see your GitHub name associated with topic(s) below, that means you've been identified as a key contributor for W3C Workshop on Web and Machine Learning discussions for those topics and we may ask you to share your perspective on the Zoom call. We'll send a reminder with Zoom meeting logistics via email.

September 22, 2020, 2pm UTC 🗓️

🔎 Scope: Web Platform Foundations for Machine Learning

✅ Goal: Understand how machine learning fits into the Web technology stack

Considerations for creating and deploying models

ℹ️ ML model format #74 - @cynthia @jbingham @wchao1115

❓ There is no standard format for packaging and shipping ML models, model formats evolve rapidly.
✔️ Proposal: Initially focus on defining a Web API for accelerating established reusable ML operations instead of standardizing a model format.

ℹ️ Protecting ML models #67 - @jasonmayes @tidoust @pyu10055 @jbingham

❓ Some ML providers need to ensure their ML models cannot be extracted from a browser app.
✔️ Proposal: Investigate existing access control mechanisms for video, learnings from 3D assets.

ℹ️ In-browser training #82 - @irealva @hapticdata @cynthia

❓ The current in-browser efforts are focused on inference rather than training.
✔️ Proposal: Understand successful real-world usages (e.g. Teachable Machine) and target transfer learning as the initial training use case for related browser API work.

Training across devices #83 - @wmaass @Nov1102 @EmmaNingMS @zolkis @jaykishigami

❓ Understand the role of edge computing in training and interactions with the web platform.
✔️ Proposal: Work with Web & Networks IG to understand edge computing use cases and ensure input from ML usages is considered.

Extending the web foundations for ML

ℹ️ Targeting WASI-NN and WebNN together #96 - @mehmetoguzderin @mingqiusun @abrown

❓ Should libraries for browsers and/or Wasm execution environments be able to target WebNN and WASI-NN together?
✔️ Proposal: TBD

ℹ️ Heterogeneous parallel computing for the web #92 - @jeffhammond @Kangz

❓ How do the heterogeneous parallel computing abstractions fit in with the web platform?
✔️ Proposal: TBD

September 23, 2020, 2pm UTC 🗓️

🔎 Scope: Machine Learning Experiences on the Web: A Developer’s Perspective

✅ Goal: Authoring ML experiences on the Web; challenges and opportunities of reusing existing ML models on the Web; on-device training, known technical solutions, gaps

Applying web design principles to ML

ℹ️ Progressive Enhancement / Graceful degradation #68 - @dontcallmedom @jbingham @wchao1115 @huningxin

❓ How to bring more ML features as optional improvements on more powerful devices and browsers without breaking web compatibility?
✔️ Proposal: TBD

ℹ️ Conformance testing of ML APIs for the Web #80 - @wchao1115 @Kangz

❓ Robust conformance testing is a cornerstone of the interoperable web platform, how to scale that to the ML APIs and formats?
✔️ Proposal: TBD

Improving web developer ergonomics

ℹ️ JS Operator overloading for Machine Learning #73 - @cynthia @huningxin

❓ Limitations in ECMAScript expressiveness impose ergonomics limitations for JS APIs on the web platform e.g. in vector matrix or tensor operations.
✔️ Proposal: TBD

ℹ️ WebGL garbage collection #63 - @jasonmayes @Kangz @wchao1115 @huningxin

❓ Garbage collection in the WebGL API affects multiple ML libraries through side effects.
✔️ Proposal: Identify any improvements in graphics APIs to alleviate the GC issue, ensure purpose-built APIs designed around computational graph abstraction (e.g. WebNN) optimize GC from library usage perspective.

ℹ️ Neural network-oriented graph database #102 - @WenheLI

❓ Understand model storage issues on the client, research the feasibility of a neural network-oriented graph database for the web.
✔️ Proposal: TBD

Developing interactive web experiences with ML

ℹ️ Action-Response Cycle bottlenecks in interactive music apps #97 - @teropa

❓ Action-Response Cycle in interactive (music) apps must execute within 20 ms. Today, web developers need to do some API gymnastics to meet the requirement.
✔️ Proposal: Investigate inference in AudioWorklet context and media integration e.g. fast streaming inputs from MediaStream.

ℹ️ Noise suppression with DSP+DNN, WebNN and Web Audio API feature gaps #100 - @jmvalin @teropa @huningxin

❓ What areas needs work on the web platform to ensure noise suppression models perform? The need for primitives like Basic Linear Algebra Subprograms, Web Audio API enhancements to allow better analysis of waveforms?
✔️ Proposal: TBD