[Pearg] Presentation to PEARG working group at IETF 116

[Phillipp Schoppmann <[e-mail]>]

Dear PEARG,

I would like the opportunity to present at IETF 116, on the topic of
"Secure Partitioning Protocols".

Background:
There have been many proposals in the last couple of years on protocols for
privacy-preserving measurement that use a set of aggregators to compute
aggregate statistics over the contributions from many clients. This
includes the DAP draft being developed within PPM, but also works such as
IPA <https://github.com/patcg-individual-drafts/ipa/> that is being
discussed in the W3C (and, as I was told, at the next PEARG meeting).
The advantages of DAP/VDAF is that the protocols share a structure that is
very easy to parallelize: Most computation is performed on individual
client contributions, and only the final step is aggregating across
clients. Other protocols don't have that structure and therefore the
parallelization story becomes more complicated. Taking the example of IPA
again, it requires grouping contributions by "match keys", which requires a
secure computation across multiple clients' inputs, and where it's unclear
how to parallelize it.

Proposal:
The proposed functionality is a secure partitioning protocol that can be
used to parallelize this class of protocols that require grouping
client contributions by a predefined key. After some discussion with Chris
Wood, I believe PEARG is the right place for this workstream. A starting
point could be the partitioning step implicit in this CCS 2022 paper
<https://eprint.iacr.org/2022/920.pdf>, which creates a differentially
private partitioning of the inputs using a distributed OPRF computation.

I believe 15+5 minutes would be ideal to provide some intuition about the
techniques in addition to the high-level goals.

Kind Regards,
Phillipp Schoppmann