-
Federated Evaluation and Tuning for On-Device Personalization: System Design & Applications
Authors:
Matthias Paulik,
Matt Seigel,
Henry Mason,
Dominic Telaar,
Joris Kluivers,
Rogier van Dalen,
Chi Wai Lau,
Luke Carlson,
Filip Granqvist,
Chris Vandevelde,
Sudeep Agarwal,
Julien Freudiger,
Andrew Byde,
Abhishek Bhowmick,
Gaurav Kapoor,
Si Beaumont,
Áine Cahill,
Dominic Hughes,
Omid Javidbakht,
Fei Dong,
Rehan Rishi,
Stanley Hung
Abstract:
We describe the design of our federated task processing system. Originally, the system was created to support two specific federated tasks: evaluation and tuning of on-device ML systems, primarily for the purpose of personalizing these systems. In recent years, support for an additional federated task has been added: federated learning (FL) of deep neural networks. To our knowledge, only one other…
▽ More
We describe the design of our federated task processing system. Originally, the system was created to support two specific federated tasks: evaluation and tuning of on-device ML systems, primarily for the purpose of personalizing these systems. In recent years, support for an additional federated task has been added: federated learning (FL) of deep neural networks. To our knowledge, only one other system has been described in literature that supports FL at scale. We include comparisons to that system to help discuss design decisions and attached trade-offs. Finally, we describe two specific large scale personalization use cases in detail to showcase the applicability of federated tuning to on-device personalization and to highlight application specific solutions.
△ Less
Submitted 16 February, 2021;
originally announced February 2021.
-
Optimal Strategies for Simultaneous Vickrey Auctions with Perfect Substitutes
Authors:
Enrico H. Gerding,
Rajdeep Kumar Dash,
Andrew Byde,
Nicholas Robert Jennings
Abstract:
We derive optimal strategies for a bidding agent that participates in multiple, simultaneous second-price auctions with perfect substitutes. We prove that, if everyone else bids locally in a single auction, the global bidder should always place non-zero bids in all available auctions, provided there are no budget constraints. With a budget, however, the optimal strategy is to bid locally if this b…
▽ More
We derive optimal strategies for a bidding agent that participates in multiple, simultaneous second-price auctions with perfect substitutes. We prove that, if everyone else bids locally in a single auction, the global bidder should always place non-zero bids in all available auctions, provided there are no budget constraints. With a budget, however, the optimal strategy is to bid locally if this budget is equal or less than the valuation. Furthermore, for a wide range of valuation distributions, we prove that the problem of finding the optimal bids reduces to two dimensions if all auctions are identical. Finally, we address markets with both sequential and simultaneous auctions, non-identical auctions, and the allocative efficiency of the market.
△ Less
Submitted 14 January, 2014;
originally announced January 2014.
-
Stratified B-trees and versioning dictionaries
Authors:
Andy Twigg,
Andrew Byde,
Grzegorz Milos,
Tim Moreton,
John Wilkes,
Tom Wilkie
Abstract:
A classic versioned data structure in storage and computer science is the copy-on-write (CoW) B-tree -- it underlies many of today's file systems and databases, including WAFL, ZFS, Btrfs and more. Unfortunately, it doesn't inherit the B-tree's optimality properties; it has poor space utilization, cannot offer fast updates, and relies on random IO to scale. Yet, nothing better has been developed s…
▽ More
A classic versioned data structure in storage and computer science is the copy-on-write (CoW) B-tree -- it underlies many of today's file systems and databases, including WAFL, ZFS, Btrfs and more. Unfortunately, it doesn't inherit the B-tree's optimality properties; it has poor space utilization, cannot offer fast updates, and relies on random IO to scale. Yet, nothing better has been developed since. We describe the `stratified B-tree', which beats all known semi-external memory versioned B-trees, including the CoW B-tree. In particular, it is the first versioned dictionary to achieve optimal tradeoffs between space, query and update performance.
△ Less
Submitted 30 March, 2011; v1 submitted 22 March, 2011;
originally announced March 2011.
-
Optimal query/update tradeoffs in versioned dictionaries
Authors:
Andrew Byde,
Andy Twigg
Abstract:
External-memory dictionaries are a fundamental data structure in file systems and databases. Versioned (or fully-persistent) dictionaries have an associated version tree where queries can be performed at any version, updates can be performed on leaf versions, and any version can be `cloned' by adding a child. Various query/update tradeoffs are known for unversioned dictionaries, many of them with…
▽ More
External-memory dictionaries are a fundamental data structure in file systems and databases. Versioned (or fully-persistent) dictionaries have an associated version tree where queries can be performed at any version, updates can be performed on leaf versions, and any version can be `cloned' by adding a child. Various query/update tradeoffs are known for unversioned dictionaries, many of them with matching upper and lower bounds. No fully-versioned external-memory dictionaries are known with optimal space/query/update tradeoffs. In particular, no versioned constructions are known that offer updates in $o(1)$ I/Os using O(N) space. We present the first cache-oblivious and cache-aware constructions that achieve a wide range of optimal points on this tradeoff.
△ Less
Submitted 12 April, 2011; v1 submitted 13 March, 2011;
originally announced March 2011.
