#### DMCA

## FTW: Fast Similarity Search under the Time Warping (2005)

Venue: | Distance, Proceedings of PODS |

Citations: | 29 - 2 self |

### Citations

3781 |
Introduction to statistical pattern recognition (2nd ed
- Fukunaga
- 1990
(Show Context)
Citation Context ...es constant-value segments to approximate sequences. While many dimensionality reduction techniques have been proposed (e.g. DFT [20, 10], Discrete Wavelet Transform [24] and Karhunen-Loeve Transform =-=[9]-=-), APCA gives especially high approximation quality. Recent applications require DTW for calculating the similarity of sequences [13, 21, 19, 12]. To reduce the matching cost, many sequence-matching t... |

2021 |
Fundamentals of Speech Recognition
- Rabiner, Juang
- 1993
(Show Context)
Citation Context ...lculate the distance between sequences whose lengths and/or sampling rates are different. Furthermore, it can be sensitive to outliers [2]. Recent applications have adopted Dynamic Time Warping (DTW) =-=[5, 22]-=- to overcome these problems [13, 21, 19, 12]. DTW is a transformation that allows sequences to be stretched along the time axis to minimize the distance between the sequences. The distance of DTW is c... |

1261 | The r*-tree: an efficient and robust access method for points and rectangles
- Beckmann
- 1990
(Show Context)
Citation Context ...at of [14]. Thus, we show the results of a performance evaluation that compared FTW with the best existing method proposed in [27]. [27]’s method is denoted by LB PAA. For LB PAA, we used the R*-tree =-=[3]-=-, and each sequence was indexed according to 16 reduced dimensions as shown in [27]. We evaluated the search performance mainly 0 50 100 150 200 250 300 350 400 25000 50000 75000 100000sW al l C lo ck... |

623 | A quantitative analysis and performance study for similarity-search methods in high dimensional spaces, in:
- Weber, Schek, et al.
- 1998
(Show Context)
Citation Context ...andomly, feature data are visited sequentially in query processing. A sequential scan of feature data should significantly boost performance because of the sequential nature of their I/O requests. In =-=[23]-=-, Weber et al. indicate two speed-up factors for the phenomenon: a practical factor of 10 and a conservative one of 5. Thus, we used these speed-up factors, SF , in our experimental evaluations. The n... |

532 | Fast subsequence matching in time-series databases
- Faloutsos, Ranganathan, et al.
- 1994
(Show Context)
Citation Context ... high enough performance for high-dimensional data because of the dimensionality curse problem [4, 23]. Their work focus on whole sequence matching. This was generalized to allow subsequence matching =-=[8, 18]-=-. Keogh et al. presented an indexing method by using the Adaptive Piecewise Constant Approximation (APCA) [15]. APCA is a dimensionality reduction technique for sequence matching based on the Euclidea... |

515 | Efficient similarity search in sequence databases
- Agrawal, Faloutsos, et al.
- 1993
(Show Context)
Citation Context ...he results of the experiments, which clearly show the effectiveness of FTW. Section 5 is a brief conclusion. 2. RELATED WORK Agrawal et al. first proposed an approach for similarity sequence matching =-=[1]-=-. Their method extracts feature vectors from sequences, and indexes them using R*-trees. Only a small number of features are extracted, since most multidimensional index structures cannot provide high... |

387 |
Adapted wavelet analysis from theory to software algorithms, A.K.Peters
- Wickerhauser
- 1994
(Show Context)
Citation Context ...lidean distance. This technique uses constant-value segments to approximate sequences. While many dimensionality reduction techniques have been proposed (e.g. DFT [20, 10], Discrete Wavelet Transform =-=[24]-=- and Karhunen-Loeve Transform [9]), APCA gives especially high approximation quality. Recent applications require DTW for calculating the similarity of sequences [13, 21, 19, 12]. To reduce the matchi... |

350 | Exact indexing of dynamic time warping,
- Keogh, Ratanamahatana
- 2005
(Show Context)
Citation Context ...ven data sequences of different lengths and/or data sequences with lengths different from that of the query sequence. 4. Support for any, as well as for no restriction on warping scope: The method in =-=[14]-=- is fast, because it cleverly exploits global constraints [22] that appear in dynamic programming (See Figure 2). We would like to have a method that can exploit the restrictions on the warping scope,... |

316 | Locally adaptive dimensionality reduction for indexing large time series databases. - Keogh, Chakrabarti, et al. - 2001 |

240 | Dimensionality reduction for fast similarity search in large time series databases.
- Keogh, Chakrabarti, et al.
- 2001
(Show Context)
Citation Context ...ment over the one of [14]. The search methods of [14] and [27] compute the envelope of the query sequence from the scope of warping paths; they then derive the PAA (Piecewise Aggregate Approximation) =-=[16, 25]-=- of the envelope (See Figures 3 and 4). The lower bounding distance between each data sequence and the query sequence is defined as the Euclidean distance between the PAA of the envelope and the MBR (... |

238 |
Bioinformatics: Sequence and Genome Analysis,
- Mount
- 2001
(Show Context)
Citation Context ...ces whose lengths and/or sampling rates are different. Furthermore, it can be sensitive to outliers [2]. Recent applications have adopted Dynamic Time Warping (DTW) [5, 22] to overcome these problems =-=[13, 21, 19, 12]-=-. DTW is a transformation that allows sequences to be stretched along the time axis to minimize the distance between the sequences. The distance of DTW is calculated by dynamic programming (See Figure... |

236 | Fast similarity search in the presence of noise, scaling, and translation in time-series databases
- Agrawal, Lin, et al.
- 1995
(Show Context)
Citation Context ...n treats sequence elements independently, it cannot be used to calculate the distance between sequences whose lengths and/or sampling rates are different. Furthermore, it can be sensitive to outliers =-=[2]-=-. Recent applications have adopted Dynamic Time Warping (DTW) [5, 22] to overcome these problems [13, 21, 19, 12]. DTW is a transformation that allows sequences to be stretched along the time axis to ... |

215 | Efficient retrieval of similar time sequences under time warping.
- Yi, Jagadish, et al.
- 1998
(Show Context)
Citation Context ...ion, which varies its accuracy during the course of query processing. Although this method is efficient, it does not guarantee no false dismissals. Yi et al. proposed a lower bounding measure for DTW =-=[26]-=-. The distance from a query sequence to each data sequence is evaluated by using a lower bounding measure, after which a candidate set is constructed. The current data sequence for each candidate is v... |

208 | The pyramid-technique: towards breaking the curse of dimensionality
- Berchtold
- 1998
(Show Context)
Citation Context ...nly a small number of features are extracted, since most multidimensional index structures cannot provide high enough performance for high-dimensional data because of the dimensionality curse problem =-=[4, 23]-=-. Their work focus on whole sequence matching. This was generalized to allow subsequence matching [8, 18]. Keogh et al. presented an indexing method by using the Adaptive Piecewise Constant Approximat... |

183 |
Fast Time Sequence Indexing for Arbitrary Lp Norms.
- Yi, Faloutsos
- 2000
(Show Context)
Citation Context ...ment over the one of [14]. The search methods of [14] and [27] compute the envelope of the query sequence from the scope of warping paths; they then derive the PAA (Piecewise Aggregate Approximation) =-=[16, 25]-=- of the envelope (See Figures 3 and 4). The lower bounding distance between each data sequence and the query sequence is defined as the Euclidean distance between the PAA of the envelope and the MBR (... |

129 |
Finding Patterns in Time Series: A Dynamic Programming Approach.
- Berndt, Clifford
- 1996
(Show Context)
Citation Context ...lculate the distance between sequences whose lengths and/or sampling rates are different. Furthermore, it can be sensitive to outliers [2]. Recent applications have adopted Dynamic Time Warping (DTW) =-=[5, 22]-=- to overcome these problems [13, 21, 19, 12]. DTW is a transformation that allows sequences to be stretched along the time axis to minimize the distance between the sequences. The distance of DTW is c... |

88 | Warping indexes with envelope transforms for query by humming,”
- Zhu, Shasha
- 2003
(Show Context)
Citation Context ...lobal constraints that appear in dynamic programming. Global constraints (e.g. the Sakoe-Chiba Band and the Itakura Parallelogram [22]) limit the scope of the warping path. Zhu et al.’s search method =-=[27]-=- is also based on global constraints and represents an improvement over the one of [14]. The search methods of [14] and [27] compute the envelope of the query sequence from the scope of warping paths;... |

58 | An index-based approach for similarity search supporting time warping in large sequence databases.
- Kim, Park, et al.
- 2001
(Show Context)
Citation Context ...wer bounding measure also employs elements in the data sequence that are smaller than the minimum of the query sequence. Kim et al. introduced a lower bounding measure employing 4-dimensional vectors =-=[17]-=- that represent the first, last, minimum, and maximum sequence elements. The vectors can be readily indexed using any spatial access method. The lower bounding measures proposed in [26] and [17] guara... |

39 | Iterative deepening dynamic time warping for time series
- Chu, Keogh, et al.
- 2002
(Show Context)
Citation Context ...per bounds that totally enclose the sequence. sequence retrieval speed against precision. They must visit all data sequences of length N , and the time complexity is still basically O(N2). Chu et al. =-=[6]-=- proposed a search method based on distance approximation, which varies its accuracy during the course of query processing. Although this method is efficient, it does not guarantee no false dismissals... |

36 | Signature technique for similarity-based queries.
- Faloutsos, Jagadish, et al.
- 1997
(Show Context)
Citation Context ...tc−1 < tc < N That is, PAc is the coarsest, while P A 1 is the most accurate. Various algorithms have been proposed to find the optimal representation of approximate segments of each sequences (e.g., =-=[7]-=-). We will use equal-sized segments to approximate sequences for simplicity although segments of different time intervals would be acceptable to FTW. Since we operate on equal-sized segments, the time... |

29 | General match: a subsequence matching method in time-series databases based on generalized windows
- Moon, Whang, et al.
- 2002
(Show Context)
Citation Context ... high enough performance for high-dimensional data because of the dimensionality curse problem [4, 23]. Their work focus on whole sequence matching. This was generalized to allow subsequence matching =-=[8, 18]-=-. Keogh et al. presented an indexing method by using the Adaptive Piecewise Constant Approximation (APCA) [15]. APCA is a dimensionality reduction technique for sequence matching based on the Euclidea... |

23 | Hierarchical Filtering Method for Content-based Music Retrieval via Acoustic
- Jang
- 2001
(Show Context)
Citation Context ...ces whose lengths and/or sampling rates are different. Furthermore, it can be sensitive to outliers [2]. Recent applications have adopted Dynamic Time Warping (DTW) [5, 22] to overcome these problems =-=[13, 21, 19, 12]-=-. DTW is a transformation that allows sequences to be stretched along the time axis to minimize the distance between the sequences. The distance of DTW is calculated by dynamic programming (See Figure... |

10 |
Automatic modeling of a 3d city map from real-world video
- Kawasaki, Yatabe, et al.
- 1999
(Show Context)
Citation Context ...ces whose lengths and/or sampling rates are different. Furthermore, it can be sensitive to outliers [2]. Recent applications have adopted Dynamic Time Warping (DTW) [5, 22] to overcome these problems =-=[13, 21, 19, 12]-=-. DTW is a transformation that allows sequences to be stretched along the time axis to minimize the distance between the sequences. The distance of DTW is calculated by dynamic programming (See Figure... |

9 |
Fintime — a financial time series benchmark. http://cs.nyu.edu/cs/faculty/shasha/fintime.html
- Jacob, Shasha
- 2000
(Show Context)
Citation Context ...ature: Temperature measurements, from 55 sensors in buildings of Carnegie Mellon University. Each sensor gives one value every 30 seconds. 2. FinTime: This is the financial time-series benchmark from =-=[11]-=-. We used historical stock market data for 100,000 securities. 3. RandomWalk: We generated 100,000 sequences by using random-walk models [25]: pi = pi−1 + xi where the p1 of each sequence is uniformly... |

4 | Memory-based forecasting for weather image patterns
- Otsuka, Horikoshi, et al.
- 2000
(Show Context)
Citation Context |