LAURENCE MORONEY: Hi, everybody, and welcome to TensorFlow

Meets.

I'm absolutely delighted to have my colleague Edd Wilder-James

here with me today.

Now, Edd is Mr. Community and TensorFlow

is all about community.

So Ed, can you tell us a little bit

about what you've been up to?

EDD WILDER-JAMES: Yeah, sure.

I have a great job, which is I kind of

to reach out to the whole community that's

working with in TensorFlow.

And one of the most striking things about TensorFlow

is obviously so many places it's used in so many areas.

And for TensorFlow to continue to be a great project,

continue to grow, we need to build it

so that the community can easily be part of the thing

that we're building because there's so many use cases.

You can't just have the core team trying

to support them all, right?

It needs to be a sustainable community where everyone

can help build TensorFlow towards the use cases

that they have,

LAURENCE MORONEY: Mm-hmm and one of the things

I find amazing is that we talk about having

a dedication to community.

But the proof of dedication to community

is that we have you full time working on it.

It's like your role, right?

EDD WILDER-JAMES: It is.

And it's a tremendously fun job to be

able to kind of help the engineering teams out

by doing the other part, designing

processes and designing the groups in the same way

that an API helps you program, right?

Some of the structures I help build, help people interact.

LAURENCE MORONEY: That's a really interesting analogy.

I never thought about it that way.

That's pretty cool.

So now, you did a talk at the TensorFlow developer summit

around community.

And there was one thing that really jumped out,

to me, at that was that you had a subtitle in your slide

that was saying, I think it was, we're building TensorFlow

together--

EDD WILDER-JAMES: That's right.

LAURENCE MORONEY: Something along those lines.

So can you tell us really what it

means for us to be building TensorFlow together

with the community?

EDD WILDER-JAMES: Yeah, I think, especially in the last year,

there are two things that help us work together

as a community.

The first of these is that we started

to use an RFC request for comments process for design

changes.

So a year ago, we were at the point

where we just kind of landed design changes in code.

Or if somebody else wanted to contribute,

they just landed in big PR.

And there's not a lot of transparency or discussion.

But now we've published, I think, over 21 RFCs

where new designs for APIs are discussed ahead

of time in public.

And it's not just the discussion because, afterwards, that

accesses documentation.

So someone, in the future, can come and understand

why we made these choices.

LAURENCE MORONEY: Interesting.

EDD WILDER-JAMES: That's one way we've built together.

The second way is that we've established,

now, six special interest groups.

And these are very defined project groups.

So they work on things like new networking protocols or ways

to connect TensorFlow to other data sources.

And these work together, predominately community

led, to build parts of TensorFlow.

So now, we've increased the surface area,

increased the transparency and the communication.

LAURENCE MORONEY: Wow, great stuff.

So one of the things that I always hear with--

it's easy to talk about community.

It's hard to build community.

And one of the things to make building community

is to try and make it easy as possible to participate.

And I know you've been doing lots and lots

of great work in that space.

Can you share a little bit about some of the great things

that we have that will help people

to participate in the community, beyond what you've already

shared?

EDD WILDER-JAMES: Oh, well, I'll try.

Yeah, there's a lot now.

There is a lot more surface area.

And it really is about surface area, right?

You walk into a big project like TensorFlow, where do you start?

Where are the points you can get traction?

So we, obviously-- I mentioned that the six that are going on.

The modularity of the code base really matters, too.

And this is one of things we're doing in TensorFlow 2.0,

is making way more modular, having

less in this monolithic core.

So now, you could find the repo that you want to work on

or the developer who's looking after that.

It's a lot more accessible.

In addition to that in code and the six that I mentioned,

we now have a community documentation group,

which is gaining steam, people bringing translations on.

LAURENCE MORONEY: I've seen the translations.

Isn't that incredible--

EDD WILDER-JAMES: Yeah, amazing.

LAURENCE MORONEY: --coming from the community.

EDD WILDER-JAMES: Last week, we posted up Korean and Russian

translations.

And it's fabulous to have first class resources on our website

to those communities.

And also, finally, the testing group

for TensorFlow 2.0, the page [INAUDIBLE] leading,

which is really giving people hands on time

to bash on TensorFlow 2.0 and help it,

make sure it meets all those important use

cases that everyone has.

LAURENCE MORONEY: Right, there's much there.

Are there any of the community contributions

that you've seen that particularly inspire you,

that you really like?

EDD WILDER-JAMES: Well, I think what particularly inspires

me is the way that all this is coming together

to support TensorFlow 2.0.

And in many ways, it would not be

possible to do 2.0 in the way we're

doing without the community.

Let me give you an example.

All the major design, changes we've consulted the RFC.

We now have moved a lot of stuff out of contrib

that was existing before and is being maintained

by community groups, the six.

That wouldn't have been possible before.

In addition, the TensorFlow 2.0 testing group,

which is also powered by a lot of great Google Developer

Experts, is really kind of mashing on the APIs,

making sure they work, but also creating

examples and notebooks that will demo the functionality.

LAURENCE MORONEY: One that I particularly

like is with TensorFlow data services, the fact

that we've being able to have contributions of data sets

from the community.

And so some of the data sets that have

come in-- there was one from Stanford,

an undergraduate at Stanford University

who contributed like 200,000 chest X-ray images into a data

set.

And to make that then easy for other people

to build training on.

It's like, without good community, how could--

I find it inspiring.

EDD WILDER-JAMES: Yeah, exactly.

It's one half about our attitude but also

about what we create and structures

and also how we code things.

LAURENCE MORONEY: Right, right, so let's switch gears

for a second.

Now, I know you're hard at work on something

called TensorFlow World.

EDD WILDER-JAMES: Yeah.

LAURENCE MORONEY: So it's a great name.

[LAUGHTER]

So could you tell us a little bit about that.

EDD WILDER-JAMES: Yeah, well, one

of the exciting things about TensorFlow

now is that it's so widespread.

And what we wanted to do was really

create an event that would enable everyone

in the ecosystem to come together to share

and to talk about what they're doing.

Obviously, Google does some great TensorFlow oriented

events.

But they're limited in capacity.

They're quite short.

There's a lot of the core TensorFlow

team presenting outwards.

But there's so many things in the world

where TensorFlow is being used that it's

really important for us to continue

to grow our ecosystem by having everyone come together.

LAURENCE MORONEY: I see.

I see.

EDD WILDER-JAMES: Well, let me give you

an example about some of the things we'll have in there.

So it's not just talks.

But there will be tutorials.

LAURENCE MORONEY: OK.

EDD WILDER-JAMES: There'll be training.

There'll be a chance for software vendors who interface

with TensorFlow-- out in the real world,

people keep all their data in databases and clouds and other

places--

that we want to tell their story about how

they work with TensorFlow, too.

So it'll really be something for everybody.

LAURENCE MORONEY: Can I go please?

EDD WILDER-JAMES: Well, let me tell you.

Let me tell you a good way that you could go.

Obviously, we'd love to have everyone come and attend

as an attendee.

But right now, we have a call for participation open--

LAURENCE MORONEY: Right.

EDD WILDER-JAMES: --which is open until April 23.

LAURENCE MORONEY: OK.

EDD WILDER-JAMES: And you can go to the website

URL, which is very excitingly tensorflow.world.

LAURENCE MORONEY: OK, I think I can remember.

EDD WILDER-JAMES: Yeah, right, the clue's in the name.

And submit a proposal to talk or deliver a tutorial.

And we'll be reviewing those.

And by sort of mid-May, we'll have a schedule settled.

LAURENCE MORONEY: And where and when is TensorFlow World?

EDD WILDER-JAMES: Right, the conference

itself is October 28 through the 31st of October.

And that'll be in Santa Clara.

LAURENCE MORONEY: OK, so and it's got Halloween.

EDD WILDER-JAMES: It's Halloween and TensorFlow loves orange.

So I'm psyched.

LAURENCE MORONEY: Exactly, it'll be great.

Are you going to go in fancy dress?

[LAUGHTER]

Well, thanks so much.

Oh, one last question, actually.

If people want to learn more about the community,

where can they go?

EDD WILDER-JAMES: We decided that, again, one URL

is the best idea.

So if you go to tensorflow.org/community,

if you just go to the TensorFlow home page and hit

on the community label, you'll get to all our resources.

LAURENCE MORONEY: Awesome, awesome, OK, great.

Thanks so much.

And thanks everybody for watching this episode

of TensorFlow Meets.

And if you've any questions for Edd,

if you've any questions for me, just

please leave them in the comments below.

And all the links that we discussed today,

I'll paste them in there as well.

So thanks so much.

[MUSIC PLAYING]