Results 1 - 10
of
49
Impact of problem centralization in distributed constraint optimization algorithms
- In AAMAS
, 2005
"... Recent progress in Distributed Constraint Optimization Problems (DCOP) has led to a range of algorithms now available which differ in their amount of problem centralization. Problem centralization can have a significant impact on the amount of computation required by an agent but unfortunately the d ..."
Abstract
-
Cited by 47 (4 self)
- Add to MetaCart
(Show Context)
Recent progress in Distributed Constraint Optimization Problems (DCOP) has led to a range of algorithms now available which differ in their amount of problem centralization. Problem centralization can have a significant impact on the amount of computation required by an agent but unfortunately the dominant evaluation metric of “number of cycles ” fails to account for this cost. We analyze the relative performance of two recent algorithms for DCOP: OptAPO, which performs partial centralization, and Adopt, which maintains distribution of the DCOP. Previous comparison of Adopt and OptAPO has found that OptAPO requires fewer cycles than Adopt. We extend the cycles metric to define “Cycle-Based Runtime (CBR) ” to account for both the amount of computation required in each cycle and the communication latency between cycles. Using the CBR metric, we show that Adopt outperforms OptAPO under a range of communication latencies. We also ask: What level of centralization is most suitable for a given communication latency? We use CBR to create performance curves for three algorithms that vary in degree of centralization, namely Adopt, OptAPO, and centralized Branch and Bound search.
Privacy loss in distributed constraint reasoning: A quantitative framework for analysis and its applications
, 2006
"... It is critical that agents deployed in real-world settings, such as businesses, offices, universities and research laboratories, protect their individual users ’ privacy when interacting with other entities. Indeed, privacy is recognized as a key motivating factor in the design of several multiagent ..."
Abstract
-
Cited by 28 (2 self)
- Add to MetaCart
(Show Context)
It is critical that agents deployed in real-world settings, such as businesses, offices, universities and research laboratories, protect their individual users ’ privacy when interacting with other entities. Indeed, privacy is recognized as a key motivating factor in the design of several multiagent algorithms, such as in distributed constraint reasoning (including both algorithms for distributed constraint optimization (DCOP) and distributed constraint satisfaction (DisCSPs)), and researchers have begun to propose metrics for analysis of privacy loss in such multiagent algorithms. Unfortunately, a general quantitative framework to compare these existing metrics for privacy loss or to identify dimensions along which to construct new metrics is currently lacking. This paper presents three key contributions to address this shortcoming. First, the paper presents VPS (Valuations of Possible States), a general quantitative framework to express, analyze and compare existing metrics of privacy loss. Based on a state-space model, VPS is shown to capture various existing measures of privacy created for specific domains of DisCSPs. The utility of VPS is further illustrated through analysis of privacy loss in DCOP algorithms, when such algorithms are used by personal assistant agents to schedule meetings
Bumping strategies for the multiagent agreement problem
- In Proceedings of Autonomous Agents and Multi-Agent Systems, (AAMAS
, 2005
"... We introduce the Multiagent Agreement Problem (MAP) to represent a class of multiagent scheduling problems. MAP is based on the Distributed Constraint Reasoning (DCR) paradigm and requires agents to choose values for variables to satisfy not only their own constraints, but also equality constraints ..."
Abstract
-
Cited by 27 (1 self)
- Add to MetaCart
(Show Context)
We introduce the Multiagent Agreement Problem (MAP) to represent a class of multiagent scheduling problems. MAP is based on the Distributed Constraint Reasoning (DCR) paradigm and requires agents to choose values for variables to satisfy not only their own constraints, but also equality constraints with other agents. The goal is to represent problems in which agents must agree on scheduling decisions, for example, to agree on the start time of a meeting. We investigate a challenging class of MAP – private, incremental MAP (piMAP) in which agents do incremental scheduling of activities and there exist privacy restrictions on information exchange. We investigate a range of strategies for piMAP, called “bumping ” strategies. We empirically evaluate these strategies in the domain of calendar management where a personal assistant agent must schedule meetings on behalf of its human user. Our results show that bumping decisions based on scheduling difficulty models of other agents can significantly improve performance over simpler bumping strategies.
Experimental analysis of privacy loss in dcop algorithms
- in AAMAS
, 2006
"... Abstract.Distributed Constraint Optimization (DCOP) is rapidly emerging as a prominent technique for multiagent coordination. Unfortunately, rigorous quantitative evaluations of privacy loss in DCOP algorithms have been lacking despite the fact that agent privacy is a key motivation for applying DCO ..."
Abstract
-
Cited by 20 (4 self)
- Add to MetaCart
(Show Context)
Abstract.Distributed Constraint Optimization (DCOP) is rapidly emerging as a prominent technique for multiagent coordination. Unfortunately, rigorous quantitative evaluations of privacy loss in DCOP algorithms have been lacking despite the fact that agent privacy is a key motivation for applying DCOPs in many applications. Recently, Maheswaran et al. [1, 2] introduced a framework for quantitative evaluations of privacy in DCOP algorithms, showing that early DCOP algorithms lose more privacy than purely centralized approaches and questioning the motivation for applying DCOPs. Do state-of-the art DCOP algorithms suffer from a similar shortcoming? This paper answers that question by investigating several of the most efficient DCOP algorithms, including both DPOP and ADOPT. Furthermore, while previous work investigated the impact on efficiency of distributed contraint reasoning design decisions, e.g. constraint-graph topology, asynchrony, message-contents, this paper examines the privacy aspect of such decisions, providing an improved understanding of privacy-efficiency tradeoffs. Finally, this paper augments previous work on system-wide privacy loss, by investigating inequities in individual agents ’ privacy loss. 1
Privacy-Preserving Multi-agent Constraint Satisfaction
- INTERNATIONAL CONFERENCE ON COMPUTATIONAL SCIENCE AND ENGINEERING
, 2009
"... Constraint satisfaction has been a very successful paradigm for solving problems such as resource allocation and planning. Many of these problems pose themselves in a context involving multiple agents, and protecting privacy of information among them is often desirable. Secure multiparty computation ..."
Abstract
-
Cited by 15 (6 self)
- Add to MetaCart
Constraint satisfaction has been a very successful paradigm for solving problems such as resource allocation and planning. Many of these problems pose themselves in a context involving multiple agents, and protecting privacy of information among them is often desirable. Secure multiparty computation (SMC) provides methods that in principle allow such computation without leaking any information. However, it does not consider the issue of keeping agents’ decisions private from one another. In this paper, we show an algorithm that uses SMC in distributed computation to satisfy this objective.
Privacy Guarantees through Distributed Constraint Satisfaction
- IEEE/WIC/ACM INTERNATIONAL CONFERENCE ON WEB INTELLIGENCE AND INTELLIGENT AGENT TECHNOLOGY
, 2008
"... The reason for using distributed constraint satisfaction algorithms is often to allow agents to find a solution while revealing as little as possible about their variables and constraints. So far, most algorithms for DisCSP do not guarantee privacy of this information. This paper describes some simp ..."
Abstract
-
Cited by 11 (6 self)
- Add to MetaCart
The reason for using distributed constraint satisfaction algorithms is often to allow agents to find a solution while revealing as little as possible about their variables and constraints. So far, most algorithms for DisCSP do not guarantee privacy of this information. This paper describes some simple techniques that can be used with DisCSP algorithms such as DPOP, and provide sensible privacy guarantees based on the distributed solving process without sacrificing its efficiency.
Zero-Knowledge Proofs for Mix-nets of Secret Shares and a Version of ElGamal with Modular Homomorphism
, 2005
"... Mix-nets can be used to shuffle vectors of shared secrets. This operation can be an important building block for solving combinatorial problems where constraints depend on secrets of different participants. A main contribution of this paper is to show how participants in the mix-net can provide Ze ..."
Abstract
-
Cited by 10 (6 self)
- Add to MetaCart
Mix-nets can be used to shuffle vectors of shared secrets. This operation can be an important building block for solving combinatorial problems where constraints depend on secrets of different participants. A main contribution of this paper is to show how participants in the mix-net can provide Zero-Knowledge proofs to convince each other that they do not tamper with the shuffled secrets, and that inverse permutations are correctly applied at unshu#ing. The approach is related to the proof of knowing an isomorphism between large graphs. We also make a detailed review and comparison with rationales and analysis of Chaum's and Merritt's mix-nets. Another
Multiply-constrained distributed constraint optimization
- In AAMAS
, 2006
"... Distributed constraint optimization (DCOP) has emerged as a useful technique for multiagent coordination. While previous DCOP work focuses on optimizing a single team objective, in many domains, agents must satisfy additional constraints on resources consumed locally (due to interactions within thei ..."
Abstract
-
Cited by 9 (1 self)
- Add to MetaCart
Distributed constraint optimization (DCOP) has emerged as a useful technique for multiagent coordination. While previous DCOP work focuses on optimizing a single team objective, in many domains, agents must satisfy additional constraints on resources consumed locally (due to interactions within their local neighborhoods). Such resource constraints may be required to be private or shared for efficiency’s sake. This paper provides a novel multiply-constrained DCOP algorithm for addressing these domains which is based on mutually-intervening search, i.e. using local resource constraints to intervene in the search for the optimal solution and vice versa. It is realized through three key ideas: (i) transforming n-ary constraints to maintain privacy; (ii) dynamically setting upper bounds on joint resource consumption with neighbors; and (iii) identifying if the local DCOP graph structure allows agents to compute exact resource bounds for additional efficiency. These ideas are implemented by modifying Adopt, one of the most efficient DCOP algorithms. Both detailed experimental results as well as proofs of correctness are presented.
Secure discsp protocols - from centralized towards distributed solutions
- in DCR05 Workshop
, 2005
"... Abstract. We present new protocols for secure distributed constraint satisfaction problems (DisCSPs). The presented protocols are the first to enable an oblivious use of advanced search techniques heuristics. The first protocol is a centralized protocol, where two of the agents collect ‘encrypted’ d ..."
Abstract
-
Cited by 8 (1 self)
- Add to MetaCart
Abstract. We present new protocols for secure distributed constraint satisfaction problems (DisCSPs). The presented protocols are the first to enable an oblivious use of advanced search techniques heuristics. The first protocol is a centralized protocol, where two of the agents collect ‘encrypted’ data from all other parties, and obliviously perform a search algorithm. Our protocol improves on the previous solution of [YKH05] in several ways: It does not require introducing new agents into the protocol; it enables the use of non-trivial search techniques such as backjumping and ordering heuristics of variables and values; and, it completely eliminates information leakage to all agents. Our second protocol makes the first steps toward a feasible distributed secured protocol for solving DisCSPs. Our protocol enables agents to concurrently perform non sequential (asynchronous) algorithms. It forms an alternative network, whose nodes are small groups (e.g. pairs) of agents, that is generated from the original DisCSP. Each node group obliviously performs the roles of all its members in the search algorithm. We also identify the communication pattern of the protocol as a possible leakage source, and suggest how to eliminate this leakage. Finally, we discuss a hybrid solution that combines the centralized and distributed protocols and reduces the total communication cost. 1
Secure combinatorial optimization simulating DFS tree-based variable elimination
- In 9th Symposium on Artificial Intelligence and Mathematics, Ft
, 2006
"... are NP-hard. Using variable elimination techniques [5, 13] COPs can be solved with computation that is exponential only in the induced-width of the constraint graph (given some order on the nodes), i.e. smaller than n. Orders on nodes allowing for some parallelism are offered by Depth First Search ( ..."
Abstract
-
Cited by 7 (7 self)
- Add to MetaCart
(Show Context)
are NP-hard. Using variable elimination techniques [5, 13] COPs can be solved with computation that is exponential only in the induced-width of the constraint graph (given some order on the nodes), i.e. smaller than n. Orders on nodes allowing for some parallelism are offered by Depth First Search (DFS) trees of the constraint graph [3, 13]. Any arithmetic circuit can be compiled into a general secure multi-party computation where no participant learns anything except for the result [1, 8]. We show in [25] that a secure combinatorial problem solver must necessarily pick the result randomly among optimal solutions, to be really secure. We recently developed SMC [19], the first programming language that translates [1]’s theory into practice. SMC also supports constraint satisfaction problems (CSPs), but additional techniques were revealed needed to offer acceptably efficient support for COPs. In [24] we proposed arithmetic circuits for solving COPs but which are exponential in the number of variables, n, for any constraint graph. Here we show how to construct an arithmetic circuit with the complexity properties of DFS-based variable elimination, and that finds a random optimal solution for any COP. For forest constraint graphs, this leads to a linear cost secure solver. Developing an arithmetic circuit performing the operations of the dynamic programming step in variable elimi-1 Significant input was received from Benjamin Pflanz. 1 nation proves to be quite straightforward and similar to previous work. We encountered a more interesting scientific challenge in choosing a secure scheme for the equivalent of the decoding step. The decoding step consists of traversing the dynamic programming data structures backward to detect the assignments that generate the winning alternative. What seems to be the straightforward arithmetic circuit translation reveals results before the end of the computation, compromising security. We show how to develop an arithmetic circuit comprising all processing until the end of the computation. 1