Time Series Segmentation through Automatic Feature Learning

Abstract

Internet of things (IoT) applications have become increasingly popular in recent years, with applications ranging from building energy monitoring to personal health tracking and activity recognition. In order to leverage these data, automatic knowledge extraction -whereby we map from observations to interpretable states and transitions -must be done at scale. As such, we have seen many recent IoT data sets include annotations with a human expert specifying states, recorded as a set of boundaries and associated labels in a data sequence.ese data can be used to build automatic labeling algorithms that produce labels as an expert would. Here, we refer to human-specified boundaries as breakpoints. Traditional changepoint detection methods only look for statistically-detectable boundaries that are defined as abrupt variations in the generative parameters of a data sequence. However, we observe that breakpoints occur on more subtle boundaries that are non-trivial to detect with these statistical methods. In this work, we propose a new unsupervised approach, based on deep learning, that outperforms existing techniques and learns the more subtle, breakpoint boundaries with a high accuracy. rough extensive experiments on various real-world data sets -including human-activity sensing data, speech signals, and electroencephalogram (EEG) activity traces -we demonstrate the effectiveness of our algorithm for practical applications. Furthermore, we show that our approach achieves significantly be er performance than previous methods.

Segmenting Hybrid Trajectories using Latent ODEs

“…However, training data is required to fit the base model and, as such, its total runtime is significantly higher. Other methods have incorporated deep architectures with CPD search methods (Lee et al, 2018), but use a sliding window search with predefined window size, and use a feature distance metric to determine boundaries as opposed to marginal likelihood used by LatSegODE.…”

Hybrid deep neural networks to infer state models of black-box systems

“…We applied their approach on our data in early stages of the research but it could not perform as others. Lee et al, trained deep auto encoder networks that learns latent features in the data to detect change points [45]. Ebrahimzadeh et al, proposed what they call a pyramid recurrent neural network architecture, which is resilient to missing to detect patterns that are warped in time [22].…”

“…There are also a family of methods based on Bayesian models that focus on finding changes in parameters of underlying distributions of the data [1,5,7,24,45,62].…”

“…Making assumptions about the data such as its distribution or the distribution of change points across the time and relying on basic statistical properties are the two major short comings of traditional CPD methods [45], which our proposed approach has overcome.…”

Active multi-fidelity Bayesian online changepoint detection

“…The Bayesian approach is appealing due to the ability to specify priors and represent posterior uncertainty(Chib, 1998;Fearnhead, 2006;Chopin, 2007). For streaming applications, exact filtering algorithms allow for online Bayesian detection of changepoints without retrospective smoothing(Fearnhead and Liu, 2007;Adams and MacKay, 2007).Many applications of online changepoint detection are in real-time settings with limited resources for sensing and computation, such as content delivery networks(Akhtar et al, 2018), autonomous vehicles(Ferguson et al, 2015), and smart home and internet-of-things devices(Aminikhanghahi et al, 2018;Lee et al, 2018;Munir et al, 2019). In such resourceconstrained settings, the observations for a changepoint detector are typically environmental measurements, for example heart-rate data(Villarroel et al, 2017).…”