Hiroya Inakoshi

A Fast Algorithm for Matching Planar Maps with Minimum Fréchet Distances

In this paper, we present a fast and practical algorithm for a map-matching problem searching a path on a given graph that minimizes Fréchet distance from a given trajectory, which is a natural measurement based on the sequential order of the trajectory. However, it sometimes costs seriously to compute the Fréchet distance while making correspondences to on a path on the graph in the order from the beginning of the trajectory as a naive method (as the definition) since it often occurs to backtrack and recompute. We developed an incremental technique for updating the Fréchet distance between the trajectory and a path to overcome the problem stated above. It enables the proposed algorithm to evaluate distances for any candidate paths faster than the naive one. In addition, we can adopt Dijkstras graph searching manner due to the technique and omit to search and evaluate some useless candidates which have no relations with the solution. That also contributes to accelerate the algorithm. Experimental results show that our algorithm was more than fifty times faster than the algorithm of Alt (J. Algorithms 2003), which is formulated as a optimization problem repeating to solve decision problems with a binary search on a set of candidates of the Fréchet distance.

Mining frequent partite episodes with partwise constraints

In this paper, we study the problem of efficiently mining frequent partite episodes that satisfy partwise constraints from an input event sequence. Through our constraints, we can extract episodes related to events and their precedent-subsequent relations, on which we focus, in a short time. This improves the efficiency of data mining using trial and error processes. A partite episode of length k is of the form P=〈P1,...,Pk〉 for sets Pi(1≤i≤k) of events. We call Pi a part of P for every 1≤i≤k. We introduce the partwise constraints for partite episodes P, which consists of shape and pattern constraints. A shape constraint specifies the size of each part of P and the length of P. A pattern constraint specifies subsets of each part of P. We then present a backtracking algorithm that finds all of the frequent partite episodes satisfying a partwise constraint from an input event sequence. By theoretical analysis, we show that the algorithm runs in output polynomial time and polynomial space for the total input size. In the experiment, we show that our proposed algorithm is much faster than existing algorithms for mining partite episodes on an artificial and a real-world datasets.

Chimera: stream-oriented XML filtering/querying engine

In this paper, we study the problem of filtering and querying massive XML data against a large set of XPath patterns in Univariate XPath. Based on an efficient matching engine XSIGMA for linear XPath patterns with Boolean expression over keywords and a twig evaluator over event streams, we propose an XPath filtering/querying engine Chimera, which runs fast and stably for any XPath patterns without heavy pre- processing techniques for queried data often used by existing native XMLDBs and RDBs. Chimera also runs much faster than those engines against thousands of XPath patterns. We implemented Chimera and showed its effectiveness by several experiments on artificial and real datasets.

Discovery of emerging patterns from nearest neighbors

In this paper, we propose a scalable classifier that uses jumping emerging patterns (JEPs), which are combinations of values that occur in one class. The original classifier, DeEPs, is an instance-based classifier that operates on all instances in real-time. It discovers maximal patterns that occur throughout the entire database and identifies JEPs by using these patterns. The necessary computational effort, though, is likely to increase when DeEPs is applied to a large database. Our proposed classifier operates on the nearest neighbors of a test instance. This reduction of instances improves scalability as the database volume increases. Moreover, our classifier imposes a restriction regarding JEPs discovery, so that it excludes patterns that cannot be identified as either correct JEPs or JEPs caused by the maximal patterns missing from nearest neighbors. These probably incorrect JEPs are specialized with additional items and participate in class determination. Our classifier perform significantly faster with these two enhancements, while it remains as accurate as the original classifier.

Pyramid stack data stream mining for handling concept-drifting

Data stream mining has gained growing attentions recently. Concept drift is a particular problem in data stream mining, which is defined as the distribution of data may change over time. Most of current methods try to estimate the current distribution or reconstruct the current distribution from a mixture of old distributions. They suffer problems of estimation and reconstruction error respectively. In this paper, we found that a classifier that fits the current distribution can be obtained more directly than the current methods by ensembling classifiers trained with increasing number of recent data. This strategy guarantees that no matter when and how concept drift happens, there is always a classifier that suits the current data distribution. So our method only needs to select the current distribution classifier out of all classifiers we hold. This is much easier than estimation and reconstruction. We test our method on four real world data sets. Comparing with other methods, our method is the best algorithm in terms of average accuracy.

Discovery of Areas with Locally Maximal Confidence from Location Data

A novel algorithm is presented for discovering areas having locally maximized confidence of an association rule on a collection of location data. Although location data obtained from GPS-equipped devices have promising applications, those GPS points are usually not uniformly distributed in two-dimensional space. As a result, substantial insights might be missed by using data mining algorithms that discover admissible or rectangular areas under the assumption that the GPS data points are distributed uniformly. The proposed algorithm composes transitively connected groups of irregular meshes that have locally maximized confidence. There is thus no need to assume the uniformity, which enables the discovery of areas not limited to a certain class of shapes. Iterative removal of the meshes in accordance with the local maximum property enables the algorithm to perform 50 times faster than state-of-the-art ones.

EVIS: a fast and scalable episode matching engine for massively parallel data streams

We propose a fast episode pattern matching engine EVIS that detects all occurrences in massively parallel data streams for an episode pattern, which represents a collection of event types in a given partial order. There should be important applications to be addressed with this technology, such as monitoring stock price movements, and tracking vehicles or merchandise by using GPS or RFID sensors. EVIS employs a variant of non-deterministic finite automata whose states are extended to maintain their activated times and activating streams. This extension allows EVISs episode pattern to have 1) interval constraints that enforce time-bound conditions on every pair of consequent event types in the pattern, and 2) stream constraints by which two interested series of events are associated with each other and found in arbitrary pairs of streams. The experimental results show that EVIS performs much faster than a popular CEP engine for both artificial and real world datasets, as well as that EVIS effectively works for over 100,000 streams.