Marc Dufay

2papers

2 Papers

4.7DCJun 5
General Convex Agreement with Near-Optimal Communication

Marc Dufay, Diana Ghinea, Anton Paramonov

Byzantine Agreement (BA) considers a setting of $n$ parties out of which up to $t$ can be byzantine (malicious), and requires the honest parties to agree on an input subject to a condition called \emph{validity}: if all honest parties have input $v$, the output agreed upon must be $v$. Convex Agreement (CA) strengthens BA by requiring the output agreed upon to lie in the convex hull of the honest parties' inputs. This validity condition captures aggregation tasks, such as robust learning and sensor fusion, where honest inputs may differ but should still constrain the final decision. Existing protocols for CA over general convexity spaces require at least $O(L \cdot n^2)$ bits of communication for $L$-bit inputs, leaving a gap with BA's $Ω(L \cdot n)$ lower bound. We investigate this gap, and we present deterministic synchronous CA protocols with near-optimal communication complexity in the long-message regime. When $L=Ω(n\cdotκ)$, where $κ$ is a security parameter, our protocols use $\mathcal{O}(L\cdot n\log n)$ bits of communication for finite convexity spaces and $\mathcal{O}(L\cdot n^{1+o(1)})$ communication for Euclidean spaces $\mathbb{R}^d$. Our protocols also have asymptotically optimal round complexity $\mathcal{O}(n)$. If an upper bound $L$ on the honest inputs' length in bits is known in advance, we achieve near-optimal resilience $t<n/(ω+\varepsilon)$ for any constant $\varepsilon>0$, where $ω$ is the Helly number of the convexity space. When no such bound is known, we achieve resilience $t<n/(ω+\varepsilon+1)$. As a sample application, we show how our protocols can be used to obtain efficient solutions for parallel instances of BA. Our main technical contribution is the use of extractor graphs to obtain a deterministic assignment of parties to committees, which is robust against adaptive adversaries.

NEJun 22, 2022
General Univariate Estimation-of-Distribution Algorithms

Benjamin Doerr, Marc Dufay

We propose a general formulation of a univariate estimation-of-distribution algorithm (EDA). It naturally incorporates the three classic univariate EDAs \emph{compact genetic algorithm}, \emph{univariate marginal distribution algorithm} and \emph{population-based incremental learning} as well as the \emph{max-min ant system} with iteration-best update. Our unified description of the existing algorithms allows a unified analysis of these; we demonstrate this by providing an analysis of genetic drift that immediately gives the existing results proven separately for the four algorithms named above. Our general model also includes EDAs that are more efficient than the existing ones and these may not be difficult to find as we demonstrate for the OneMax and LeadingOnes benchmarks.