Distributed asynchronous search with private constraints (extended abstract)

Abstract

Many problem-solving tasks can be formalized as constraint saris/action problems (CSP). In a multi-agent setting, information about constraints and variables may be distributed among different agents and kept confidential. Existing algorithms for distributed constraint satisfaction consider mainly the case where access to variables is restricted to certain agents, but information about constraints can be exchanged among them. In this paper, we propose methods where constraints are private but variables can be manipulated by any agent. We describe a new algorithm for distributed CSPs, called asynchronous aggregation search (AAS). It differs from existing methods in that it treats simultaneously sets of partial solutions and it exchanges information about aggregated valuations for combinations of variables. While the approach we propose provides a more general framework for dealing with privacy requirements, its overall performance is comparable or better than that of existing methods, as shown by the experiments.

Privacy of Existence of Secrets: Introducing Steganographic DCOPs and Revisiting DCOP Frameworks

“…While this definition assumes that the number of agents equals the number of variables, with each agent keeping secret the domain of exactly one variable, other versions studied the case of unequal numbers of variables and agents with no secret domains but with secret constraints (Silaghi, Sam-Haroud, and Faltings 2000b).…”

“…There are several efforts to classify privacy concepts in distributed constraint optimization (Léauté and Faltings 2013) This interpretation is assumed in many subsequent works, where agents' constraints are assumed to be extensionally defined using mutual knowledge of the values in domains of each others' variables (Silaghi, Sam-Haroud, and Faltings 2000b;Modi et al 2002).…”

“…While this is also seen as an extension of the CSP family, the efficiency was no longer a sufficient criteria and direct comparison with traditional centralized constraints was seen as irrelevant, despite a lack of formal specification of the impediments to additional constraint sharing in terms of either technical cost or quantified privacy constraints. This lack of specification gave birth to a flurry of works exploiting various amounts of constraint information without quantifica-tion and with efficiency gains whose trade-off was not always clear (Silaghi, Sam-Haroud, and Faltings 2000a) in terms of the privacy or in terms of the technical cost of constraint data centralization.…”

Coordination of intelligent agents in real‐time search

“…19, No. 2 There are several application areas of search algorithms like game playing, learning, planning, pattern recognition, robotics and theorem proving (Kanal & Kumar, 1988;Silaghi et al, 2000). Traditional methods like A* (Nilsson, 1971) are guaranteed to find optimal solutions, but they are all off-line search algorithms which compute the entire solution path before executing the first step in the path.…”

Arithmetic circuit for the first solution of distributed CSPs with cryptographic multi-party computations

“…Privacy has been recently stressed in [8,4,9,7,11] as an important goal in SisCSP algorithm design. We presented a technique where agents that need to cooperate and whose problems can be modeled as CSPs can find a first solution without any leak of additional information about their constraints.…”