Nonparametric Extrema Analysis in Time Series for Envelope Extraction, Peak Detection and Clustering

Abstract

In this paper, we propose a nonparametric approach that can be used in envelope extraction, peak-burst detection and clustering in time series. Our problem formalization results in a naturally defined splitting/forking of the time series. With a possibly hierarchical implementation, it can be used for various applications in machine learning, signal processing and mathematical finance. From an incoming input signal, our iterative procedure sequentially creates two signals (one upper bounding and one lower bounding signal) by minimizing the cumulative L1 drift. We show that a solution can be efficiently calculated by use of a Viterbi-like path tracking algorithm together with an optimal elimination rule. We consider many interesting settings, where our algorithm has near-linear time complexities.

Second Order Regret Bounds Against Generalized Expert Sequences under Partial Bandit Feedback

“…Similar to [64], [65], we update the weights w λt using the following two-step approach. First, we define an intermediate variable z λt (which incorporates the exponential update as in the exponential weighting [42]) such that z λt w λt e −ηt−1φ t,λ t (1) , (10) where φ t,λt(1) is a measure of the performance (λ t (1)) th expert at time t, which we discuss more in the next section. Secondly, we create a probability sharing network among the equivalence classes (which also represents and assigns a weight to every individual sequence S t implicitly) at time t as…”

“…In machine learning literature [1], [2], the field of online learning [3] is heavily studied in a myriad of fields from control theory [4] and computational learning theory [5], [6] to decision theory [7], [8]. Especially, algorithms pertaining to the universal prediction perspective [9] have been utilized in many signal processing applications [10]- [13], and in sequential estimation/detection problems [14]- [17] such as density estimation or anomaly detection [18]- [22]. Some of its most popular applications are in multi-agent systems [23]- [27] and more prominently in reinforcement learning problems [28]- [39], where the famous exploration and exploitation trade off is commonly encountered [40].…”

Smoothing with the Best Rectangle Window is Optimal for All Tapered Rectangle Windows

“…The decision of what is considered short or long term, determines the moving average's input parameters (e.g., length) [31]. It is most prominently utilized in the field of mathematical finance, to analyze stock prices and trading volumes [32], [33]. It is also utilized in the analysis of many macroeconomic metrics such as GDP, import, export and unemployment [34].…”

Natural Hierarchical Cluster Analysis by Nearest Neighbors with Near-Linear Time Complexity

“…In the problem of clustering, the aim is to group together the data points in meaningful ways such that similar data points belong to the same groups [1], [2]. This data analysis is commonly used in many distinct fields of research [3], [4] including signal processing [5]- [10], bioinformatics [11]- [15], machine learning [16]- [18], recommender systems [19]- [25], anomaly detection [26]- [32]. In general, clustering is a form of unsupervised classification problem, where the samples are distributed to classes according to a measure of similarity (generally a metric or distance function) [33].…”