Chu, K. and Wong, M. (1999): Fast time-series searching with scaling and shifting. Proc 18th ACM Symposium on Principles of Database Systems, Philadelphia, PA, USA, 237-248.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....at the rate of several gigabytes a day [148] The most promising similarity search methods are techniques that perform dimensionality reduction on the data, then use a multidimensional index structure to index the data in the transformed space. The technique was introduced in [5] and extended in [119, 32]. The original work by Agrawal et al. utilizes the Discrete Fourier Transform (DFT) to perform the dimensionality reduction, but other techniques have been suggested, including Singular Value Decomposition (SVD) 79, 76, 81] the Discrete Wavelet Transform (DWT) 29, 151, 75] and Piecewise ....

....approximately with sentence length text query) We tested on two datasets, one chosen because it is very heterogeneous and one chosen because it is very homogeneous. Homogeneous Data:Electrocardiogram: This dataset is taken from the MIT Research Resource for Complex Physiologic Signals [32]. It is a relatively clean and uncomplicated electrocardiogram. The total size of the data is 100,000 objects. Heterogeneous Data: Mixed Bag: This dataset we created by combining 7 datasets with widely varying properties of shape, structure, noise etc. The only preprocessing performed was to ....

K. Chu and M. Wong. Fast time-series searching with scaling and shifting. Proceedings of PODS, 1999.

....patterns, to process user queries in a fast and an accurate manner, and to compute statistics on data streams in real time. 1.1 Related work There has been a substantial body of work on similarity search in sequence databases. Various high di mensional index structures have been proposed in [2, 3, 11, 16, 27, 29, 33, 37] to achieve fast query response time and a good quality of answers. Theoretical methods have been developed for com paring data streams under various Lp distances [12] for clustering and computing the k median [22, 32] and for computing aggregates over data streams [14, 18] Various ....

K. K. W. Chu and M. H. Wong. Fast time-series searching with scaling and shifting. In PODS, pages 237-248, 1999.

....student at University of California at Irvine The most promising similarity search methods are techniques that perform dimensionality reduction on the data, then use a multidimensional index structure to index the data in the transformed space. The technique was introduced in [1] and extended in [39, 31,11]. The original work by Agrawal et al. utilizes the Discrete Fourier Transform (DFT) to perform the dimensionality reduction, but other techniques have been suggested, including Singular Value Decomposition (SVD) 28, 24, 23] the Discrete Wavelet Transform (DWT) 9, 49, 22] and Piecewise ....

Chu, K & Wong, M. (1999). Fast time-series searching with scaling and shifting. Proceedings of the 18 th A CM Symposium on Principles of Database Systems, Philadelphia.

....data are examples of sequence data. The disrance similarity measure for these data varies based on the application and the data type. Some of the distance measures currently in use are Euclidean distance [1, 12, 16, 26] other vector norms like L and L [2] shift scale invariant distance measures [13, 27], edit distance [8, 25, 35] and score matrix based similarity [21] The sizes of spatial and sequence datasets are growing rapidly. Excessive amounts of data make similarity search challenging, incurring a large amount of disk I O and CPU cost. An important primitive on such datasets is the ....

K.K.W. Chu and M.H. Wong. Fast time-series searching with scaling and shifting. In PODS, Philadelphia, PA, 1999.

....indexing by a dimensionality reduction technique [3, 29, 13] On the other hand, the model cannot deal well with outliers and is very sensitive to small distortions in the time axis. There are a number of interesting extensions to the above model to support various transformations such as scaling [9, 25], shifting [9, 14] normalization [14] and moving average [25] Other recent works on indexing time series data for similarity queries assuming the Euclidean model include [17, 16] Another approach is based on the time warping technique that first has been used to match signals in speech ....

....reduction technique [3, 29, 13] On the other hand, the model cannot deal well with outliers and is very sensitive to small distortions in the time axis. There are a number of interesting extensions to the above model to support various transformations such as scaling [9, 25] shifting [9, 14], normalization [14] and moving average [25] Other recent works on indexing time series data for similarity queries assuming the Euclidean model include [17, 16] Another approach is based on the time warping technique that first has been used to match signals in speech recognition [26] Berndt ....

K. Chu and M. Wong. Fast Time-Series Searching with Scaling and Shifting. ACM Principles of Database Systems, pages 237--248, June 1999.

....In a separate work, Park, Chu, and Hsu [16] use the idea of time warping distance. This distance metric compares sequences of different lengths by stretching them. The distance between two time series data can be made shift and scale invariant by transforming them onto shift eliminated plane [10, 5]. Another distance metric tric 23 is defined by Lee, Chun, Kim, Lee, and Chung [15] for multidimensional sequences. Although this metric has a high recall, it again allows false dismissals, when determining candidate solution intervals. Vlachos, Kollios and Gunopulos [27] proposed to use the ....

....Subsequence (LCSS) technique in order to find similar trajectories. The authors show that LCSS is more robust to noise than both Euclidean and time warping distances. The distance between two time series data can be made shift and scale invariant by transforming them onto shift eliminated plane [10, 5]. Range searches and nearest neighbor searches in whole matching and subsequence matching have been the principal queries of interest for time series data. Whole matching corresponds to the case when the query sequence and the sequences in the database have the same length. Agrawal, Faloutsos, ....

K.K.W. Chu and M.H. Wong. Fast time-series searching with scaling and shifting. In PODS, Philadelphia, PA, 1999.

....the need for database support. Databases have not accommodated such data in the past, as their design paradigm always was one of snapshot representation . Their present support for spatial time series is at best rudimentary. General time series support is an active field of research, however [3, 7, 8, 11, 13]. The central theme of this paper concerns the process of turning raw data streams into object trajectories, and turning a collection of the latter into an aggregate result.Thus, we study the problem of aggregating moving object trajectories. There are many applications of trajectory ....

....defining and determining moving object trajectory similarity: di#erences in sequence length and sampling rates, as well as inherent spatial uncertainty. The simplest approach to define similarity between two sequences is viewing them as vector and using Euclidean distance as similarity measure [2, 7, 9, 12, 13, 15, 16, 24, 25]. Such techniques cannot be easily applied to sequences having di#erent length or sampling rate. Also, they are not e#ective in the presence of noise in the data, as is common in spatial time series. To remedy some of these problems, 17, 20, 30] used a timewarpingtechniquetomeasure similarity, ....

K.K.W.ChuandM.H.Wong.Fasttime-series searching with scaling and shifting. In V. Vianu and

....of sequence data. The distance similarity measure for these data varies based on the application and the data type. Some of the distance measures currently in use are Euclidean distance [1, 13, 17, 26, 28, 39, 43] other vector norms like L 1 and L1 [2] shift scale invariant distance measures [14, 27], edit distance [9, 25, 35] and score matrix based similarity [22] The sizes of spatial and sequence datasets are growing rapidly. Excessive amount of data makes similarity search challenging, incurring large amount of disk I O and CPU cost. An important primitive on such datasets is the join, ....

K.K.W. Chu and M.H. Wong. Fast time-series searching with scaling and shifting. In PODS, Philadelphia, PA, 1999.

....classic first paper by Agrawal et al. 1 ] More than 68 of the indexing approaches surveyed here use the original GEMINI framework of Faloutsos [17] but suggest a different approach to the dimensionality reduction stage. The proposed representations include the Discrete Fourier Transform (DFT) [1, 11, 16, 28, 49, 50], several kinds of Wavelets (DWT) 10, 27, 45, 51, 57, 60] Singular Value Decomposition [32, 35] Adaptive Piecewise Constant Approximation [32] Inner Products [18] and Piecewise Aggregate Approximation (PAA) 61] The majority of work has focused solely on performance issues, however some ....

....of a proposed approach, vs. the quality of implementation of the completing approaches. Implementing fairly complex indexing techniques allows many opportunities for implementation bias. For example, suppose you hope to demonstrate that DWT is superior to DFT. With shiftnormalized data [11, 28] the first DWT coefficient is zero so you could take advantage of that fact bff indexing the 2 na to N I th coefficients, rather than the 1 to N coefficients. However, you might neglect doing a similar optimization for DFT, whose first real coefficient is also zero for normalized data. Another ....

[Article contains additional citation context not shown here]

Chu, K. & Wong, M. (1999). Fast time-series searching with scaling and shifting. In proceedings of the l g ACM Symposium on Principles of Database Systems. Philadelphia, PA, May 31Jun 2. pp 237-248.

....of functions )r, such that F, F )r. The objective is to find those F, F in )r that minimize the distance. The distance between two time series X, Y is then argminF1,F:rD(F (X) F (Y) The family of functions )r can be global scaling, local scaling [3] global scaling and different baselines [7, 12, 10], or moving averaging [a0] Similarity of generalized time series Current work on the problem of computing the similarity of generalized time series is using the LCSS similarity model ( 53, 8] This model is generally robust against outliers and allows shifting in time when matching the time ....

K.K.W. Chu, M.H. Wong. Fast Time-Series Searching with Scaling and Shifting. In the 18th ACM Syrup. on Principles of Database Systems (PODS-1999), Philadelphia, PA

....of the time sequences. A normalized time sequence has a zero mean and a unit standard deviation. Given a query Q, the authors present a search algorithm that finds all database sequences S such that the normalized Euclidean distance DN (Q; aS b) ffl for some a 0 and b. Chu and Wong [5] considered the asymmetric formulation D 2 (aQ b; S) ffl. They use a transformation to map the data sequences onto the Shift Eliminated Plane. Both of these formulations of distance are inherently asymmetric in its treatment of query and database sequences. We focus on the symmetric notion of ....

....are dependent, and in the second model they are independent. We propose a novel index structure called CS Index (Cone Slice) for shift and scale invariant comparison of time sequences. As a part of this technique, the sequences in the database are first mapped to the shift eliminated plane [5]. The transformed points are then clustered in hierarchical cone slices. These slices are stored on disk according to an in order traversal, and a pointer to each slice along with angle and spatial extent information is maintained in memory. Given any query, it is first mapped onto shift ....

[Article contains additional citation context not shown here]

K.K.W. Chu and M.H. Wong. Fast time-series searching with scaling and shifting. In PODS, Philadelphia, PA, 1999.

....The p norm distance between two n dimensional vectors x and y is defined as L p ( x; y) P n i=1 (x i y i ) p ) 1 p . For p = 2 it is the well known Euclidean distance and for p = 1 the Manhattan distance. Various approaches have used, extended and indexed this distance metric [2, 37, 18, 14, 10, 32, 10, 20, 24, 23]. Another approach is based on the time warping technique that first has been used to match signals in speech recognition [33] Berndt and Clifford [5] proposed to use this technique to measure the similarity of time series data in data mining. Recent works have also used this similarity measure ....

K. Chu and M. Wong. Fast Time-Series Searching with Scaling and Shifting. ACM Principles of Database Systems, pages 237--248, June 1999.

.... the database is called similar sequence matching [1, 7] Owing to faster computing speed and larger storage devices, there has been a number of efforts to utilize the large amount of time series data, and accordingly, similar sequence matching has become an importance research topic in data mining [1, 2, 5, 6, 7, 10, 11, 12, 19]. Various similarity models have been studied in similar sequence matching. In this paper, we use the similarity model based on the Euclidean distance [1, 5, 7, 11] Given two sequences x = fx 0 ; x 1 ; xn Gamma1 g and y = fy 0 ; y 1 ; yn Gamma1 g of the same length n, the Euclidean ....

....sequences S 1 ; S 2 ; SN of varying lengths, a query sequence Q, and the tolerance ffl, we find all the sequences S i , one or more subsequences of which are in ffl match with Q, and the offsets in S i of those subsequences. Thus, subsequence matching is a generalization of whole matching [5, 6, 7, 19]. In this paper, we focus 2 on subsequence matching. Faloutsos et al. 7] have proposed a novel solution for subsequence matching on query sequences of varying lengths (we simply call this solution FRM by taking authors initials) Subsequences similar to the query sequence can be found anywhere ....

Chu, K. W. and Wong, M. H., "Fast Time-Series Searching with Scaling and Shifting," In Proc. the 15th ACM SIGACT-SIGMOD-SIGART Symposium on Principles of Database Systems, Philadelphia, Pennsylvania, pp. 237-248, June 1999. 27

....multidimensional space [1, 5, 7, 15, 20, 21] However, Euclidean distance by itself may be insufficient to describe the similarity. For example, if one time series is a constant multiple of the other, or is its shifted form, they can still be similar depending on the application. Chu and Wong [6] defined the Shift Eliminated Transformation, which maps the database and query sequences onto the Shift Eliminated Plane. These transformations are then used to compute the minimum distance (after all possible scaling and shiftings) between two time sequences. The authors idea is applicable to ....

....even after compressing the index to one fifth. Our notion of similarity between two time sequences is based on Euclidean distance. Such a metric is not invariant with respect to scaling and shifting, a possible shortcoming for comparison of time based data. However, as shown by Chu et al. [6], a transformation onto the shift eliminated plane achieves this invariance with respect to scaling and shifting. This transformation could be applied to all the subsequences prior to the construction of the index. In this case, each query subsequence must be mapped onto the shift eliminated plane ....

K. K. W. Chu and M. H. Wong. Fast time-series searching with scaling and shifting. In PODS, Philedelphia, PA, 1999.

....A similar case can be made for comparing time series data after eliminating shifts. Usually this is because the zero value in a time sequence differs in different measurements, and the results of a query need to be invariant under all possible shiftings. In a previous paper, Chu and Wong [6] defined the Shift Eliminated Transformation. This transformation maps data sequences onto the Shift Eliminated Plane, making the transformed data shift invariant. The authors propose to store the transformed data in an R tree [9, 3] The distance between a data sequence and a query is defined as ....

....component sequences are dependent, and in the second model they are independent. We propose a new index structure called CS Index for shift and scale invariant comparison of time sequences. As a part of this technique, the sequences in the database are first mapped to the shift eliminated plane [6]. The transformed points are then clustered in hierarchical cone slices. These slices are stored on disk in a depth firstsearch order, and a pointer to each slice along with angle and spatial extent information is maintained in memory. Given any query, it is first mapped onto shift eliminated ....

[Article contains additional citation context not shown here]

K. K. W. Chu and M. H. Wong. Fast time-series searching with scaling and shifting. In PODS, Philedelphia, PA, 1999.

....at the rate of several gigabytes a day [48] The most promising similarity search methods are techniques that perform dimensionality reduction on the data, then use a multidimensional index structure to index the data in the transformed space. The technique was introduced in [1] and extended in [39, 31,11]. The original work by Agrawal et al. utilizes the Discrete Fourier Transform (DFT) to perform the dimensionality reduction, but other techniques have been suggested, including Singular Value Decomposition (SVD) 28, 24, 23] the Discrete Wavelet Transform (DWT) 9, 49, 22] and Piecewise ....

Chu, K & Wong, M. (1999). Fast time-series searching with scaling and shifting. Proceedings of the 18 th ACM Symposium on Principles of Database Systems, Philadelphia.

....at the rate of several gigabytes a day [48] The most promising similarity search methods are techniques that perform dimensionality reduction on the data, then use a multidimensional index structure to index the data in the transformed space. The technique was introduced in [1] and extended in [39, 31,11]. The original work by Agrawal et al. utilizes the Discrete Fourier Transform (DFT) to perform the dimensionality reduction, but other techniques have been suggested, including Singular Value Decomposition (SVD) 28, 24, 23] the Discrete Wavelet Transform (DWT) 9, 49, 22] and Piecewise ....

Chu, K & Wong, M. (1999). Fast time-series searching with scaling and shifting. Proceedings of the 18 th ACM Symposium on Principles of Database Systems, Philadelphia.

....elements can be traced backward in the table by choosing the previous cells with the lowest cumulative distance. This distance computation has the complexity O( X Y ) Table 1 shows the cumulative distance table for two sequences X = 4, 3#. HereD tw (X,Y ) 12 because T [ X ] Y ] T [6][3] 12. 2.2.3. Multi dimensional time warping distance Before proposing the multi dimensional time warping distance function, let us consider the weighted distance function Dmbase for any two multi dimensional elements X[i]andY [j]withk features: Dmbase (X [i] Y [j] # X[i] h] Y ....

K. W. Chu and M. H. Wong, "Fast time-series searching with scaling and shifting," Proc. ACM Symp. Principles of Database Syst. (PODS ) (1999), pp. 237--248.

....matching searches for the data sequences similar to a query sequence. Subsequence matching searches for the subsequences, contained in data sequences, that are similar to a query sequence of arbitrary length. In order to measure the similarity of any two sequences of length n, most approaches [1, 8, 12, 13, 22] map the sequences into points in an n dimensional space and compute the Euclidean distance between those points as a similarity measure. However, they often fail to search for the data sequences that are actually similar to a query sequence in users perspective when employing only the Euclidean ....

....they often fail to search for the data sequences that are actually similar to a query sequence in users perspective when employing only the Euclidean distance as a similarity measure. Therefore, recent work on similarity search tends to support various types of transformations such as scaling [2, 8, 13], shifting [2, 8, 13] normalization [9, 13] moving average [17, 22] and time warping [4, 19, 27] Time warping is a transformation that allows any sequence element to replicate itself as many times as needed without extra costs [27] For example, two sequences S = 21, 21, 20, 20, 23, 23, ....

[Article contains additional citation context not shown here]

K. W. Chu, M. H. Wong, "Fast Time-Series Searching with Scaling and Shifting", Proc. ACM PODS , pp. 237-248, 1999.

....at the rate of several gigabytes a day [23] The most promising similarity search methods are techniques that perform dimensionality reduction on the data, then use a multidimensional index structure to index the data in the transformed space. The technique was introduced in [1] and extended in [19, 6, 21, 10]. The original work by Agrawal et al. utilizes the Discrete Fourier Transform (DFT) to perform the dimensionality reduction, but other techniques have been suggested, including Singular Value Decomposition (SVD) 18, 15] the Discrete Wavelet Transform (DWT) 5, 24] and Piecewise Polynomial ....

....for indexing time series, the set of representations we could use include those listed in Table 5: Table 5. Dimensionality reduction techniques that can be used with E Index to index time series. Key: the introducing paper, extensions and follow up work. Discrete Fourier Transform 1 , [6, 10, 13, 18, 19, 21, 24]. Discrete Wavelet Transform 5 , 13, 24] Piecewise Constant approximation 15 , 26] Piecewise Linear Approximation 17 , 22] Inner Product Approximation 9 . Adaptive Piecewise Constant Approximation 16 . There are some notable omissions. We could include the Discrete ....

[Article contains additional citation context not shown here]

Chu, K & Wong, M. (1999). Fast time-series searching with scaling and shifting. Proceedings of the 18 th ACM Symposium on Principles of Database Systems, Philadelphia.

No context found.

Chu, K. and Wong, M. (1999): Fast time-series searching with scaling and shifting. Proc 18th ACM Symposium on Principles of Database Systems, Philadelphia, PA, USA, 237-248.

No context found.

K. K. W. Chu and M. H. Wong. Fast time-series searching with scaling and shifting. In PODS, 1999.

No context found.

Chu, K., Wong, M. Fast time-series searching with scaling and shifting. In Proc. of the 18th ACM Symp. on Principles of Database Systems. Philadelphia, PA, pp 237-248, May 1999.

No context found.

K. K. W. Chu and M. H. Wong. Fast time-series searching with scaling and shifting. In PODS, pages 237--248, 1999.

No context found.

K. W. Chu and M. H. Wong. Fast time-series searching with scaling and shifting. In Proc. the 15th ACM SIGACT-SIGMOD-SIGART Symposium on Principles of Database Systems, pages 237-248, 1999.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC