[wbs] response to 'Call for Review: Web Machine Learning Working Group Charter'

The following answers have been successfully submitted to 'Call for Review:
Web Machine Learning Working Group Charter' (Advisory Committee) for
Microsoft Corporation by Travis Leithead.


The reviewer's organization suggests changes to this Charter, and only
supports the proposal if the changes are adopted [Formal Objection].

Additional comments about the proposal:
   Microsoft is generally supportive of forming the Web Machine Learning
Working group. On close review of the proposed charter as well as the
Community Group charter [1], we're concerned about a few things that we
would like to see clarified.

1. The Scope of the proposed WG charter is much more generic than the scope
of the CG. The proposed WG scope section describes "Web API aiming to
expose generic capabilities to the Web required to provide close-to-native
ML performance in the browser." The CG's charter is much more explicit
describing the specific capabilities of the API (construction from a NN
computational graph, with common building blocks, etc., compilation
optimized for hardware, and input from various sources including
potentially Media streams). It is unclear to us whether the proposed
working group intends to pursue these same objectives. For example: will
the group consider new NN input from WebRTC media streams (or other Stream
types), Web Audio, and Sensors? Adding to this uncertainty is the notable
absence of these related community and working groups in the Coordination
section as compared to the CG.

2. The Scope section implies in the text "it may also work on a
higher-level API to load a custom pre-trained Machine Learning model [...]"
that a formerly out-of-scope item from the CG is now in-scope for the WG,
namely "this Community Group does not attempt to mandate a specific neural
network or Machine Learning model schema or format." We understand that the
situation has evolved in the community group, and are in favor of
supporting a WG scope that includes the model schema or format, but would
like to see it listed explicitly in the Scope.
 
3. The proposed WG's Out of Scope section omits a notable item called out
in the Community Group's Out of Scope section, namely: "This [Working]
Group will not define any hardware features or algorithms." We believe this
is an important assertion to highlight in the Out of Scope section of the
proposed charter, or to clarify what aspects of hardware features or
algorithms will be pursued by this group.

[1] https://webmachinelearning.github.io/charter/


The reviewer's organization intends to participate in these groups:
   - Web Machine Learning Working Group

The reviewer's organization:
   - intends to review drafts as they are published and send comments.
   - intends to develop experimental implementations and send experience
reports.
   - intends to develop products based on this work.
   - intends to apply this technology in our operations.

Answers to this questionnaire can be set and changed at
https://www.w3.org/2002/09/wbs/33280/webml-charter/ until 2021-03-26.

 Regards,

 The Automatic WBS Mailer

Received on Thursday, 11 March 2021 01:21:06 UTC