
New Instability Results for High Dimensional Nearest Neighbor Search

In%2DNetwork Outlier Detection in Wireless Sensor Networks
To address the problem of unsupervised outlier detection in wireless sensor networks, we develop an algorithm that (1) is flexible with respect to the outlier definition, (2) works in-network with a communication load proportional to the outcome, and (3) reveals its outcome to all sensors. We examine the algorithm’s performance using simulation with real sensor data streams. Our results demonstrate that the algorithm is accurate and imposes a reasonable communication load and level of power consumption. 1.
On the Privacy of Euclidean Distance Preserving Data Perturbation

Approximate Distributed K%2DMeans Clustering over a Peer%2Dto%2DPeer Network.

New instability results for high%2Ddimensional nearest neighbor search.

Distributed Decision%2DTree Induction in Peer%2Dto%2DPeer Systems.
This paper offers a scalable and robust distributed algorithm for decision tree induction in large Peer-to-Peer (P2P) environments. Computing a decision tree in such large distributed systems using standard centralized algorithms can be very communication-expensive and impractical because of the synchronization requirements. The problem becomes even more challenging in the distributed stream monitoring scenario where the decision tree needs to be updated in response to changes in the data distribution. This paper presents an alternate solution that works in a completely asynchronous manner in distributed environments and offers low communication overhead, a necessity for scalability. It also seamlessly handles changes in data and peer failures. The paper presents extensive experimental results to corroborate the theoretical claims. Index Terms peer-to-peer, data mining, decision trees I.
Distributed Top%2DK Outlier Detection from Astronomy Catalogs using the DEMAC System.

In%2DNetwork Outlier Detection in Wireless Sensor Networks.
To address the problem of unsupervised outlier detection in wireless sensor networks, we develop an algorithm that (1) is flexible with respect to the outlier definition, (2) works in-network with a communication load proportional to the outcome, and (3) reveals its outcome to all sensors. We examine the algorithm’s performance using simulation with real sensor data streams. Our results demonstrate that the algorithm is accurate and imposes a reasonable communication load and level of power consumption. 1.
An Attacker%27s View of Distance Preserving Maps for Privacy Preserving Data Mining.
Abstract. We examine the effectiveness of distance preserving transformations in privacy preserving data mining. These techniques are potentially very useful in that some important data mining algorithms can be efficiently applied to the transformed data and produce exactly the same results as if applied to the original data e.g. distance-based clustering, k-nearest neighbor classification. However, the issue of how well the original data is hidden has, to our knowledge, not been carefully studied. We take a step in this direction by assuming the role of an attacker armed with two types of prior information regarding the original data. We examine how well the attacker can recover the original data from the transformed data and prior information. Our results offer insight into the vulnerabilities of distance preserving transformations. 1
An Attacker%27s View of Distance Preserving Maps for Privacy Preserving Data Mining.
Abstract. We examine the effectiveness of distance preserving transformations in privacy preserving data mining. These techniques are potentially very useful in that some important data mining algorithms can be efficiently applied to the transformed data and produce exactly the same results as if applied to the original data e.g. distance-based clustering, k-nearest neighbor classification. However, the issue of how well the original data is hidden has, to our knowledge, not been carefully studied. We take a step in this direction by assuming the role of an attacker armed with two types of prior information regarding the original data. We examine how well the attacker can recover the original data from the transformed data and prior information. Our results offer insight into the vulnerabilities of distance preserving transformations. 1
K%2DMeans Clustering Over a Large%2C Dynamic Network.

Distributed Data Mining in Peer%2Dto%2DPeer Networks.
Distributed data mining deals with the problem of data analysis in environments with distributed data, computing nodes, and users. Peer-to-peer (P2P) computing is emerging as a new distributed computing paradigm for many novel applications that involve exchange of information among a large number of peers with little centralized coordination. P2P file sharing, P2P electronic commerce, and P2P monitoring based on a network of sensors are some examples. This paper offers an overview of distributed data mining applications and algorithms for P2P environments. It describes both exact and approximate distributed data mining algorithms that work in a decentralized manner.
Clustering distributed data streams in peer%2Dto%2Dpeer environments.

An information theoretic histogram for single dimensional selectivity estimation.

Distributed data mining and agents.
Abstract. Multi-Agent Systems (MAS) offer an architecture for distributed problem solving. Distributed Data Mining (DDM) algorithms focus on one class of such distributed problem solving tasks—analysis and modeling of distributed data. This paper offers a perspective on DDM algorithms in the context of multiagents systems. It discusses broadly the connection between DDM and MAS. It provides a high-level survey of DDM, then focuses on distributed clustering algorithms and some potential applications in multi-agent-based problem solving scenarios. It reviews algorithms for distributed clustering, including privacypreserving ones. It describes challenges for clustering in sensor-network environments, potential shortcomings of the current algorithms, and future work accordingly. It also discusses confidentiality (privacy preservation) and presents a new algorithm for privacy-preserving density-based clustering.
Multi%2Dagent Systems and Distributed Data Mining.
We present a general Multi-Agent System framework for distributed data mining based on a Peer-to-Peer model. The framework adopts message-based asynchronous communication and a dynamic load balancing policy that is particularly suitable for irregular search algorithms. A modular design allows a separation of the general-purpose system protocols and software components from the specific data mining algorithm. While the general architecture has been implemented and successfully tested on a parallel frequent subgraph mining algorithm, several interesting issues still have to be explored. The present work will discuss them and will introduce the ongoing research efforts aimed at exploiting and leveraging the MAS for distributed data mining applications. 1 P2P-BASED MAS for Distributed Data Mining The last decade has seen an ever increasing availability of large amounts of data in many fields of science and in many IT applications. Data mining techniques have become popular techniques, which can reveal the valuable knowledge hidden in the rough data. However, these techniques often require high performance approaches in order to cope with the overwhelming amount of data and the complexity of algorithms. An effective approach is based on distributed and parallel computing.


On approximation measures for functional dependencies.
We investigate the problem of defining an approximation measure for functional dependencies (FDs). For fixed sets of attributes,  X and Y , an approximation measure is a function which maps relation instances to real numbers. The number to which an instance is mapped, intuitively, describes the strength of the dependency, X ! Y , in that instance. We define an approximation measure for FDs based on a connection between Shannon&#039;s information theory and relational database theory. Our measure is normalized to lie between zero and one (inclusive) , and maps a relation instance to zero if and only if X ! Y holds in the instance. Hence, the smaller the number to which an instance is mapped, the &#034;closer&#034; X ! Y is to being an FD in the instance. To put our measure in context, we compare it to a slight variation of a measure previously defined by Kivinen and Mannila, g3 . We denote the variation as  g3 , although, our results, essentially, apply unchanged to g3 . The purpose of comparing our measure with  g3 is to develop a deeper understanding of not only our measure, but also,  g3 . Moreover, we gain a deeper understanding of the natural intuitive notion of an approximate FD. We observe that our measure and  g3 agree at their extremes but are quite different in-between. As a result, we conclude that our measure and  g3 are significantly different. An interesting question emerges from this conclusion: is there a rigorous way to determine when one measure better captures the meaning of the degree to which an FD is approximate? 1 
A note on approximation measures for multi%2Dvalued dependencies in relational databases.

An Axiomatic Approach to Defining Approximation Measures for Functional Dependencies.

Improving Query Evaluation with Approximate Functional Dependency Based Decompositions.
We investigate how relational restructuring may be used to improve query performance.
Adding a path connectedness operator to FO poly %28linear%29.
In the constraint database community, FO+poly and FO+ linear have been proposed as foundations for spatial database query languages. One of the strengths of this approach is that these languages are a clean and natural generalization of Codd&#039;s relational model to a spatial setting. As a result, rigorous mathematical study of their expressiveness and complexity can be carried out. Along this line, important geometric queries involving connectivity have been shown to be inexpressible in FO+poly and FO+linear. To address this problem, we extend both languages with a parameterized path-connectivity predicate, Pconn. We show that: FO + linear + Pconn and FO+poly+Pconn-3D are closed and have PTIME data complexity. We also examine the expressiveness of FO+poly+Pconn and FO+ linear +Pconn and show that parity and transitive closure are expressible in each. 
FastFDs A Heuristic%2DDriven
Discovering functional dependencies (FDs) from an existing  relation instance is an important technique in data mining and database  design. To date, even the most efficient solutions are exponential in the  number of attributes of the relation (n), even when the size of the output  is not exponential in n. Lopes et al. developed an algorithm, Dep-Miner,  that works well for large n on randomly-generated integer-valued relation  instances [LPL 00a]. Dep-Miner first reduces the FD discovery problem  to that of finding minimal covers for hypergraphs, then employs a levelwise  search strategy to determine these minimal covers. Our algorithm,  FastFDs, instead employs a depth-first, heuristic driven search strategy  for generating minimal covers of hypergraphs. This type of search is  commonly used to solve search problems in Artificial Intelligence (AI)  [RN 95]. Our experimental results indicate that the levelwise strategy  that is the hallmark of many successful data mining algorithms is in  fact significantly surpassed by the depth-first, heuristic driven strategy  FastFDs employs, due to the inherent space efficiency of the search. Furthermore,  we revisit the comparison between Dep-Miner and Tane, including  FastFDs. We report several tests on distinct benchmark relation  instances, comparing the Dep-Miner and FastFDs hypergraph approaches  to Tane&#039;s partitioning approach for mining FDs from a relation  instance. At the end of the paper (appendix A) we provide experimental  data comparing FastFDs with a third algorithm, fdep [FS 99].
Useful Transformations in Answer set programming.

FastFDs A Heuristic%2DDriven
Discovering functional dependencies (FDs) from an existing  relation instance is an important technique in data mining and database  design. To date, even the most efficient solutions are exponential in the  number of attributes of the relation (n), even when the size of the output  is not exponential in n. Lopes et al. developed an algorithm, Dep-Miner,  that works well for large n on randomly-generated integer-valued relation  instances [LPL 00a]. Dep-Miner first reduces the FD discovery problem  to that of finding minimal covers for hypergraphs, then employs a levelwise  search strategy to determine these minimal covers. Our algorithm,  FastFDs, instead employs a depth-first, heuristic driven search strategy  for generating minimal covers of hypergraphs. This type of search is  commonly used to solve search problems in Artificial Intelligence (AI)  [RN 95]. Our experimental results indicate that the levelwise strategy  that is the hallmark of many successful data mining algorithms is in  fact significantly surpassed by the depth-first, heuristic driven strategy  FastFDs employs, due to the inherent space efficiency of the search. Furthermore,  we revisit the comparison between Dep-Miner and Tane, including  FastFDs. We report several tests on distinct benchmark relation  instances, comparing the Dep-Miner and FastFDs hypergraph approaches  to Tane&#039;s partitioning approach for mining FDs from a relation  instance. At the end of the paper (appendix A) we provide experimental  data comparing FastFDs with a third algorithm, fdep [FS 99].
An Empirical Study of the 4%2Dvalued Kripke%2DKleene and 4%2Dvalued Well%2DFounded Semantics in Random Propositional Logic Programs.

An Empirical Study of %22Lazy%22 Protocols for Routing Information in Dynamic Networks.
It is well known that the critical path provides an absolute lower bound on the execution time of a conservative parallel discrete event simulation. It stands to reason that optimal execution time can only be achieved by immediately executing each event on the critical path. However, dynamically identifying the critical event is difficult, if not impossible. In this paper, we examine several heuristics that might help to determine the critical event, and conduct a performance study to determine the effectiveness of using these heuristics for preferential scheduling.  1. Introduction  Ultimately, performance is the most important issue in parallel processing in general and parallel simulation in particular. There are two widely used families of parallel simulation algorithms: conservative [1] and optimistic [2]. Conservative techniques are so called because they only execute events that are known to be safe, while optimistic algorithms execute any available events, but must roll back to...
