Review assignment is a common task that many people such as conference organizers, journal editors, and grant administrators would have to do routinely. As a computational problem, it involves matching a set of candidate reviewers with a paper or proposal to be reviewed. A common deficiency of all existing work on solving this problem is that they do not consider the multiple aspects of topics or expertise and all match the entire document to be reviewed with the overall expertise of a reviewer. As a result, if a document contains multiple subtopics, which often happens, existing methods would not attempt to assign reviewers to cover all the subtopics; instead, it is quite possible that all the assigned reviewers would cover the major subtopic quite well, but not covering any other subtopic. In this paper, we study how to model multiple aspects of expertise and assign reviewers so that they together can cover all subtopics in the document well. We propose three general strategies for solving this problem and propose new evaluation measures for this task. We also create a multi-aspect review assignment test set using ACM SIGIR publications. Experiment results on this data set show that the proposed methods are effective for assigning reviewers to cover all topical aspects of a document.

3D neuro-anatomical images and other volumetric data sets are important in many scientific and biomedical fields. To achieve a high compression rate, near-lossless or lossless compression algorithm is applied which uses a Hilbert traversal to produce a data stream from the original image. An extremely fast linear DPCM is used. The linear DPCM takes the average of the previous two voxels' intensity to predict the current voxel's intensity. If near-lossless compression is desired, the prediction error can be uniformly quantized. For lossless mode, no action is taken on the prediction error. The prediction error is further encoded using Huffman code. In order to provide efficient data access, the source image is divided into blocks and indexed by an octree data structure. The points are binned using a novel binning method. All the error distributions that fall into the same bin are summed together to form a summed error distribution. Although our compression method is designed for performance-critical digital brain atlas applications, it would be suitable for other applications with fast data access.

The instability problem of decision tree classification algorithms is that small changes in input training samples may cause dramatically large changes in output classification rules. Different rules generated from almost the same training samples are against human intuition and complicate the process of decision making. In this paper, we present fundamental theorems for the instability problem of decision tree classifiers. The first theorem gives the relationship between a data change and the resulting tree structure change (i.e. split change). The second theorem, Instability Theorem, provides the cause of the instability problem. Based on the two theorems, algorithmic improvements can be made to lessen the instability problem. Empirical results illustrate the theorem statements. The trees constructed by the proposed algorithm are more stable, noise-tolerant, informative, expressive, and concise. Our proposed sensitivity measure can be used as a metric to evaluate the stability of splitting predicates. The tree sensitivity is an indicator of the confidence level in rules and the effective lifetime of rules.

In a client/server architecture, replicated copies of an object are assigned to different servers (possibly client sites) in order to increase the availability of the database in case some but not all servers fail. A highly efficient, highly reliable quorum consensus protocol, the Quorum Rings protocol (QRP), specifically designed to take advantage of the client/server architecture, is introduced. The high resiliency of QRP is achieved at a very low cost in communication messages. More specifically, the number of communication messages required for a QRP read or write operation is O(1) in normal mode (when there are no failures in the system) and only O(/spl radic/N) in failure mode (when some but not all servers with replicated objects are faulty). Finally, QRP exhibits the property of graceful degradation: more communication cost is incurred only when the number of failures increases in the system.>

In a real-time database system, it is difficult to meet all timing constraints due to the consistency requirements of the underlying database. Real-time database transaction scheduling requires the development of efficient concurrency control protocols that try to maximize the number of transactions satisfying their real-time constraints. Several distributed, lock-based, real-time, concurrency control protocols are described. The relative performance of the protocols in a nonreplicated database environment is reported. The protocols take the real-time requirements of the transactions into account in ordering data accesses, while maintaining data consistency by enforcing serializability.>

The authors propose an object-based index structure for manipulating spatial objects with non-zero size. They introduce the main ideas of the proposed index structure. A detailed description of the algorithms is given for searching, insertion, and deletion in a database system with a high frequency of retrievals and a low frequency of insertions and deletions. The algorithms are then described for retrievals, insertions, and deletions for a database system with a nearly equal frequency of retrievals, insertions and deletions.>

A protocol for managing replicated data in which data copies are organized as a triangular lattice is introduced. The smallest quorum size is O( square root N), where N is the number of data copies, which is currently considered optimal for a fully distributed environment. The protocol has the property of graceful degradation. The quorum sizes increase gradually as data copy failures increase. The protocol also has the property of asymptotically high availability, i.e., the availability approaches 1 as the number of data copies goes to infinity if the probability that a data copy available is greater than 0.5.>

The authors present an approach to distributed workload analysis which can be used as a basis for the performance study of distributed deadlock detection algorithms. In particular, the expected number of times a deadlock detection algorithm is locally initiated and the subsequent number of remote invocations are derived. Simulation work was done to validate the approach.>

A highly reliable program execution environment which enables user programs to tolerate underlying hardware failures is presented. The approach is to run multiple copies of the user programs at the same time. As long as one copy survives, the user program can be completed successfully. In the meantime, the user interacts with the replicated program as if it were a normal program. The authors call this characteristic user transparent replication. In order to achieve user transparent replication, program replicas must behave consistently. Otherwise, users might get different queries or output from different running replicas. The authors identify the reasons why the inconsistent program execution problem occurs and propose algorithms to ensure that computation replicas behave consistently. With consistent running program replicas, a filter program can be easily constructed to delete duplicated I/O requests or duplicated output and thus achieve user transparency.>

The authors present an election protocol that does not assume an underlying ring structure and that tolerates failures, including lost messages and network partitioning, during the execution of the protocol itself. The major problem to be solved is that when nodes cannot communicate with one another or messages are lost, a conflict in resolving the election will often arise. In the authors' approach, the conflict is detected by the cohorts (noncandidate participants in the election). Related election protocols are discussed, and the system model is described together with assumptions about the communication subsystem. The protocol and the lost-message situations are then examined.>

Many database systems guarantee the consistency of the database for concurrent transaction processing by a standard locking protocol called two phase locking. The performance of such systems is dependent on various factors, among which are the choices of locking granularity (the amount of data controlled by a single lock) and degree of multiprogramming (the maximum number of transactions allowed to run concurrently in the system). Previous simulation and experimental studies on such systems suggest that the choices are dependent on the properties of the application.

The Report Generation Facility is a high-level interface designed and implemented to support coherent access of data stored in a number of stand-alone, independent, database systems served by it. Specifically, the Report Generation Facility provides a menu which allows a user to request one or more standard reports provided by the individual database systems and the Report Generation Facility, or to formulate a query in an English-like query language without having to specify where the requested data reside or how to retrieve them. The Report Generation Facility translates the user's request or query into subtransactions to the individual database systems and submits the substransactions in order to retrieve and format the data requested by the user. Thus, the Report Generation Facility provides a standard way to produce reports based on data stored in the database systems with minimal user effort and expertise. This paper describes its capabilities and architecture.

The optimal choices of granularity (size of a physical lock) and the degree of multiprogramming (maximum number of transactions allowed to run concurrently) for a single-site database system (SDBS) using a locking mechanism for concurrency control have been investigated by many researchers, but the case for a distributed database system (DDBS) has not been examined. Using a simulation model, we study the problem for a DDBS in various application and system environments. In particular, we have considered: the preclaim, and the claim-as-needed locking policies; three different granule placements; two different transaction sizes; and two different communication delays. Conclusions concerning the optimal choices of granularity and the degree of multiprogramming in different environments are then drawn to provide guidelines in achieving good performance in a DDBS.

This paper describes the mutual consistency maintenance mechanism used in a prototype Data Traffic Management System (DTMS), which was designed and implemented to integrate several independent and stand-alone data systems owned by the U. S. Army, Corp of Engineers (CE). The major design constraints were that no modification to the data systems involved were allowed and that the update to each duplicate copy was to be handled as automatically as possible and in a manner approved by the manager of the system containing the copy. The capabilities of the prototype DTMS and the Mutual Consistency Maintenance Subsystem are discussed. The linkage directory used to support mutual consistency maintenance and the mutual consistency maintenance mechanism are described.