先前就有许多针对时间序列资料进行的结果比对的研究.pptVIP

ACM SIGMOD Subsequence Matching on Structured Time Series Data Advisor : Dr. Hsu Presenter : Jia-Hao Yang Author :Huanmei Wu, Betty Salzberg, Gregory C Sharp, Steve B Jiang, Hiroki Shirato, David Kaeli motivation Although many method about time series、subsequence matching have been proposed. Few less attention pay to the internal structure within the data. Objective This paper using subsequence similarity matching : To predict tumor motion in real-time. (online) To find a correlation between moving pattern and patient conditions. (offline) To provide a general solution for all problem domain. Methods Using a finite state automaton to simulate the motion model. V→ segments→ stream→ records→ DB Vi : Online subsequence matching Dynamic query subsequence generation For real-time applications, query subsequences must be an accurate and last condition. DEFINITION 1. (Subsequence Stability) S is stable if Methods (cons.) The more stable, the shorter the query subsequence will be. and the length of the query subsequence is between the user specified Lmin and Lmax. EX : Lmin = 3 , Lmax = 8; Online subsequence similarity DEFINITION 2. (Online Subsequence Similarity) Methods( cons. ) Motion prediction Offline clustering Stream and patient similarity are important for many application. Stream similarity : for each query subsequence from R1 , the most similar γ.N2 retrieved subsequences from R2 will be used to define the distance between R1 and R2 . EX :γ=10% , at least 0.1×N2 with the same state order from R2. else will be removed. DEFINITION 3. (Stream Distance) Methods( cons. ) Patient similarity : It based on the stream similarity. The distance is the average distance between two streams. DEFINITION 4. (Patient Distance) Generalization of the method In addition to respiratory motion, there are many other applications which can be simulated and analyzed using the above framework. Experiences Direction : Evaluating the subsequence matching