[{"author": "Nick Doty", "text": "

maybe the chair needs to stop sharing the current slide deck?

", "time": "2022-11-08T16:34:14Z"}, {"author": "Shan Wang", "text": "

has the noise gone away?

", "time": "2022-11-08T16:34:26Z"}, {"author": "David Oliver", "text": "

no

", "time": "2022-11-08T16:34:31Z"}, {"author": "Christopher Patton", "text": "

yes

", "time": "2022-11-08T16:34:32Z"}, {"author": "Shan Wang", "text": "

I wasn't typing though..

", "time": "2022-11-08T16:34:43Z"}, {"author": "Charlie Harrison", "text": "

Are the slides posted anywhere?

", "time": "2022-11-08T16:37:53Z"}, {"author": "Benjamin Schwartz", "text": "

https://datatracker.ietf.org/meeting/115/session/ppm

", "time": "2022-11-08T16:38:12Z"}, {"author": "Nick Doty", "text": "

I know we'll get to an in-band task configuration proposal later. but is there any rough description of what out of band task configuration would be like? how does a client know what kind of data is being collected and what calculations are being done, so it knows whether to participate?

", "time": "2022-11-08T16:42:08Z"}, {"author": "Samuel Weiler", "text": "

in the meetecho tool, look for the file folder icon

", "time": "2022-11-08T16:42:27Z"}, {"author": "Benjamin Schwartz", "text": "

Nick: I believe that's all out of band. The client just has to know.

", "time": "2022-11-08T16:43:14Z"}, {"author": "Shan Wang", "text": "

I hear some static noise, is it just me?

", "time": "2022-11-08T16:43:15Z"}, {"author": "Samuel Weiler", "text": "

tim is overmodulating.

", "time": "2022-11-08T16:43:32Z"}, {"author": "Daniel Gillmor", "text": "

yes, tim's mic is clipping

", "time": "2022-11-08T16:43:56Z"}, {"author": "Eric Rescorla", "text": "

@Mark Nottingham how does this match up to the HTTP API guidance

", "time": "2022-11-08T16:44:53Z"}, {"author": "Nick Doty", "text": "

\"Privacy Considerations: The client just has to know.\"

", "time": "2022-11-08T16:45:04Z"}, {"author": "Eric Rescorla", "text": "

The client has to know == \"The person who wrote the client knew and decided it was OK\"

", "time": "2022-11-08T16:46:13Z"}, {"author": "Martin Thomson", "text": "

Is it the case that these URLs are based on an identifier provided by the client?

", "time": "2022-11-08T16:46:45Z"}, {"author": "Martin Thomson", "text": "

That is, does the client determine the task ID or job ID?

", "time": "2022-11-08T16:47:01Z"}, {"author": "Eric Rescorla", "text": "

I believe that the task ID is pre-configured

", "time": "2022-11-08T16:48:35Z"}, {"author": "Eric Rescorla", "text": "

Trying to remember the job ID

", "time": "2022-11-08T16:48:38Z"}, {"author": "Christopher Patton", "text": "

yes ;the task ID is pre-configured

", "time": "2022-11-08T16:49:01Z"}, {"author": "Eric Rescorla", "text": "

@Tim Geoghegan your mic is killing me

", "time": "2022-11-08T16:49:04Z"}, {"author": "Tim Geoghegan", "text": "

Sorry about the feedback!

", "time": "2022-11-08T16:49:15Z"}, {"author": "Mark Nottingham", "text": "

I'd need to dig into the spec; from what I see so far it's somewhat reasonable.

", "time": "2022-11-08T16:49:41Z"}, {"author": "Nick Doty", "text": "

the room mic is very quiet, although clear

", "time": "2022-11-08T16:49:52Z"}, {"author": "Samuel Weiler", "text": "

it's not feedback; it's clipping. if you can lower mic gain, we'll be happier.

", "time": "2022-11-08T16:50:00Z"}, {"author": "Daniel Gillmor", "text": "

still clipping

", "time": "2022-11-08T16:51:58Z"}, {"author": "Tim Geoghegan", "text": "

Apologies for the clipping, and thanks for bearing with me

", "time": "2022-11-08T16:53:23Z"}, {"author": "Samuel Weiler", "text": "

all good.

", "time": "2022-11-08T16:53:35Z"}, {"author": "Tim Geoghegan", "text": "

Oh and I forgot to mention: I want to circulate this design to the HTTP WG mailing list to get expert feedback. Thanks Mark Nottingham for the advice you've offered thus far!

", "time": "2022-11-08T16:54:34Z"}, {"author": "Martin Thomson", "text": "

the problem here with DP is the unit that provides contributions

", "time": "2022-11-08T16:54:45Z"}, {"author": "Benjamin Schwartz", "text": "

Tim: You might do even better at HTTPAPI

", "time": "2022-11-08T16:54:58Z"}, {"author": "Charlie Harrison", "text": "

haha \"long time\" aka 2006

", "time": "2022-11-08T16:55:58Z"}, {"author": "Mark Nottingham", "text": "

We may have a somewhat more formal way to get those reviews soon - stay tuned.

", "time": "2022-11-08T16:57:38Z"}, {"author": "Martin Thomson", "text": "

Are we talking L_1, L_2, ...? Would we need to choose?

", "time": "2022-11-08T16:57:56Z"}, {"author": "Charlie Harrison", "text": "

Martin: I think this depends on the deployment, even with a fixed protocol

", "time": "2022-11-08T16:58:28Z"}, {"author": "Martin Thomson", "text": "

What guarantees are you looking to gain w.r.t. different helpers and their contribution to the output? This is a place where helpers might (again) spoil results. That doesn't matter much for Prio and Poplar, but it might matter for other protocols.

", "time": "2022-11-08T16:59:11Z"}, {"author": "Charlie Harrison", "text": "

This is discussed a bit in https://github.com/ietf-wg-ppm/draft-ietf-ppm-dap/issues/19#issuecomment-1143667405

", "time": "2022-11-08T16:59:55Z"}, {"author": "Martin Thomson", "text": "

@Charlie Harrison yup, lots of choices, no answers

", "time": "2022-11-08T17:00:07Z"}, {"author": "Charlie Harrison", "text": "

In that comment we had some alignment to ignore that threat since the helpers are already in a position to spoil the correctness to the output

", "time": "2022-11-08T17:00:48Z"}, {"author": "Martin Thomson", "text": "

that assumes a lot, but I think that - for the protocols in scope currently- that is entirely sensible

", "time": "2022-11-08T17:01:16Z"}, {"author": "Eric Rescorla", "text": "

I think that what Jonathan was suggesting was what would happen if there were two measurements which were related

", "time": "2022-11-08T17:02:08Z"}, {"author": "Eric Rescorla", "text": "

Like, say number of web pages versus number of URLs

", "time": "2022-11-08T17:02:18Z"}, {"author": "Christopher Patton", "text": "

I agree @ekr that getting DAP done has to be the highest priority

", "time": "2022-11-08T17:03:06Z"}, {"author": "Charlie Harrison", "text": "

I think it should be straightforward to support \"simple\" DP deployments

", "time": "2022-11-08T17:05:23Z"}, {"author": "Martin Thomson", "text": "

@Christopher Patton you can turn your video off now, I think

", "time": "2022-11-08T17:05:43Z"}, {"author": "Charlie Harrison", "text": "

But yeah once you start getting to complicated use-cases (overlapping batches, etc) things get off the rails quickly

", "time": "2022-11-08T17:05:43Z"}, {"author": "Tim Geoghegan", "text": "

IMO local DP can be totally orthogonal to DAP. You apply DP before you secret share, you compensate after you aggregate. Server/central DP does require DAP to say something.

", "time": "2022-11-08T17:05:57Z"}, {"author": "Martin Thomson", "text": "

n query per batch is not that dissimilar to 1 query per batch

", "time": "2022-11-08T17:06:16Z"}, {"author": "Martin Thomson", "text": "

both require some amount of state

", "time": "2022-11-08T17:06:26Z"}, {"author": "Christopher Patton", "text": "

It would be useful @charlie to spell out what the requirements are that DAP has to meet in order to be suitable for a given DP swcheme.

", "time": "2022-11-08T17:06:32Z"}, {"author": "Tim Geoghegan", "text": "

Or I think maybe VDAF would need to specify how you apply DP to an aggregate share, so that DAP could do it generically

", "time": "2022-11-08T17:06:34Z"}, {"author": "Tim Geoghegan", "text": "

@Martin Thomson: Yeah, we have to support n queries per batch for Poplar1

", "time": "2022-11-08T17:06:48Z"}, {"author": "Christopher Patton", "text": "

@ekr, do you think it's worth parallelizing work on DP, if we can get folks to work on this?

", "time": "2022-11-08T17:07:32Z"}, {"author": "Charlie Harrison", "text": "

@Tim, yes I think e.g. ENPA-style DP does not need anything more

", "time": "2022-11-08T17:07:34Z"}, {"author": "Martin Thomson", "text": "

My understanding is that the leakage in Poplar means that your epsilon with respect to the collector is very much larger than the epsilon with respect to helpers; does that bother anyone?

", "time": "2022-11-08T17:08:23Z"}, {"author": "Charlie Harrison", "text": "
\n

epsilon with respect to the collector is very much larger than the epsilon with respect to helpers

\n
\n

Isn't it the other way around? Helpers see more than the collector

", "time": "2022-11-08T17:09:11Z"}, {"author": "Martin Thomson", "text": "

Oh, yeah, sorry back to front.

", "time": "2022-11-08T17:09:31Z"}, {"author": "Charlie Harrison", "text": "

It does bother me yes :) but I can imagine a hypothetical deployment that is OK with this leakage

", "time": "2022-11-08T17:10:00Z"}, {"author": "Martin Thomson", "text": "

This design violates the basic precept I outlined earlier, where the server determines the structure of URLs.

", "time": "2022-11-08T17:10:55Z"}, {"author": "Martin Thomson", "text": "

Maybe the server can still be in a position to choose the URL, which means not including the task id in the URL proper.

", "time": "2022-11-08T17:11:24Z"}, {"author": "Charlie Harrison", "text": "

Chris: If you bug me enough I can help with a guidance draft to spell out what requirements need to be met to achieve DP

", "time": "2022-11-08T17:11:56Z"}, {"author": "Robin Wilton", "text": "

Technically I like this distributed architecture. Financially, a lot of GDPR compliance consultancies are going to love it.

", "time": "2022-11-08T17:12:06Z"}, {"author": "Christopher Patton", "text": "

:)) thanks!

", "time": "2022-11-08T17:12:13Z"}, {"author": "Martin Thomson", "text": "

Why can't the author create the task?

", "time": "2022-11-08T17:12:45Z"}, {"author": "Tim Geoghegan", "text": "

I asked about that on the ppm@ietf.org list yesterday and Shan responded.

", "time": "2022-11-08T17:13:22Z"}, {"author": "Martin Thomson", "text": "

Sorry @Tim Geoghegan that was a big tl;dr

", "time": "2022-11-08T17:14:04Z"}, {"author": "Tim Geoghegan", "text": "

Yeah I can understand that folks who are in London this week don't have time to read the novels I post to mailing lists

", "time": "2022-11-08T17:15:01Z"}, {"author": "Christopher Patton", "text": "

I think malice

", "time": "2022-11-08T17:15:09Z"}, {"author": "Robin Wilton", "text": "

collusion...

", "time": "2022-11-08T17:15:17Z"}, {"author": "Mark Nottingham", "text": "

'Malice or incompetence' is a great band name

", "time": "2022-11-08T17:16:22Z"}, {"author": "Martin Thomson", "text": "

I just read the thread: I didn't find the answer particularly responsive to the concern. It seems like an automated task creation process has some value (in terms of reducing overhead and error risk), but this design doesn't work for me.

", "time": "2022-11-08T17:18:20Z"}, {"author": "Martin Thomson", "text": "

It's inefficient and indirect.

", "time": "2022-11-08T17:18:28Z"}, {"author": "Nick Doty", "text": "

it seems like the task config would need to include data about the value/the actual data being measured, not just the batch size/query type

", "time": "2022-11-08T17:19:21Z"}, {"author": "Tim Geoghegan", "text": "

Yes, task config includes the VDAF in use, which dictates the nature of the measurement

", "time": "2022-11-08T17:19:39Z"}, {"author": "Nick Doty", "text": "

even if the client-side code wasn't going to change much, because it already decided that the URL-visited or whatever was the data being collected

", "time": "2022-11-08T17:19:50Z"}, {"author": "Martin Thomson", "text": "

@Christopher Patton be aware that there is a very long delay after you unmute before you can speak

", "time": "2022-11-08T17:20:45Z"}, {"author": "Martin Thomson", "text": "

count to 10 before you say something

", "time": "2022-11-08T17:20:57Z"}, {"author": "Martin Thomson", "text": "

In the current design, the configuration for a task determines a lot of things: is there anything in the client submission that commits to a particular configuration?

", "time": "2022-11-08T17:22:23Z"}, {"author": "Eric Rescorla", "text": "

I'm just going to say the same thing that I said previously: this is a distraction from getting DAP done

", "time": "2022-11-08T17:22:35Z"}, {"author": "Martin Thomson", "text": "

Important: The value of k changes once the threshold is hit. That's a little deceptive.

", "time": "2022-11-08T17:23:17Z"}, {"author": "Benjamin Schwartz", "text": "

How so?

", "time": "2022-11-08T17:23:52Z"}, {"author": "Martin Thomson", "text": "

For k = 100, the anonymity set for the first 100 clients is 100. The anonymity set for subsequent clients depends on the volume of submissions through the intermediary.

", "time": "2022-11-08T17:24:47Z"}, {"author": "Shan Wang", "text": "

@Nick Doty yes the task config includes the VDAF type and data required to use the VDAF on aggregators side

", "time": "2022-11-08T17:25:31Z"}, {"author": "Christopher Patton", "text": "

The commitment idea is really interesting. We're working on a security analysis for VDAF that adds credence to this.

", "time": "2022-11-08T17:25:36Z"}, {"author": "Martin Thomson", "text": "

This looks like O(k^2) to me, or did I miss something?

", "time": "2022-11-08T17:25:56Z"}, {"author": "Alex Davidson", "text": "

@martin It is O(k^2)

", "time": "2022-11-08T17:26:22Z"}, {"author": "Martin Thomson", "text": "

@Alex Davidson Good, because Shivan said O(k), but it sure didn't look like that.

", "time": "2022-11-08T17:27:31Z"}, {"author": "Benjamin Schwartz", "text": "

O(K^2) space?

", "time": "2022-11-08T17:29:24Z"}, {"author": "Alex Davidson", "text": "

both space and computation, each client sends k extra group elements, and you compute over each of them

", "time": "2022-11-08T17:30:18Z"}, {"author": "Martin Thomson", "text": "

Can you validate each as they arrive?

", "time": "2022-11-08T17:31:08Z"}, {"author": "Alex Davidson", "text": "

yes

", "time": "2022-11-08T17:31:13Z"}, {"author": "Martin Thomson", "text": "

So space is not that bad.

", "time": "2022-11-08T17:31:21Z"}, {"author": "Christopher Patton", "text": "

+1 it would be great to be able to resuse FROST

", "time": "2022-11-08T17:31:46Z"}, {"author": "Martin Thomson", "text": "

VSS as a primitive would be valuable

", "time": "2022-11-08T17:31:55Z"}, {"author": "Martin Thomson", "text": "

32 bytes times 100 for each submission at k=100 is a little scary; k=1000 starts looking pretty bad

", "time": "2022-11-08T17:34:24Z"}, {"author": "Nick Doty", "text": "

thanks all, very compact session :)

", "time": "2022-11-08T17:34:48Z"}, {"author": "Massimiliano Pala", "text": "

thank you!

", "time": "2022-11-08T17:34:59Z"}, {"author": "Tim Geoghegan", "text": "

Thanks everybody! Apologies if I mangled anyone's question or comment in the notes, it's difficult to follow cryptography while also noting.

", "time": "2022-11-08T17:35:00Z"}, {"author": "Samuel Weiler", "text": "

not exactly our choice, but I'm glad you were pleased.

", "time": "2022-11-08T17:35:04Z"}]