Multivariate Time-Series Classification Using the Hidden-Unit Logistic Model

Abstract

Abstract-We present a new model for multivariate time series classification, called the hidden-unit logistic model, that uses binary stochastic hidden units to model latent structure in the data. The hidden units are connected in a chain structure that models temporal dependencies in the data. Compared to the prior models for time series classification such as the hidden conditional random field, our model can model very complex decision boundaries because the number of latent states grows exponentially with the number of hidden units. We demonstrate the strong performance of our model in experiments on a variety of (computer vision) tasks, including handwritten character recognition, speech recognition, facial expression, and action recognition. We also present a state-of-the-art system for facial action unit detection based on the hidden-unit logistic model.

Early Classification of Time Series by Simultaneously Optimizing the Accuracy and Earliness

“…Time series classification [1,2] is a supervised learning task which deals with predicting the class labels of time series as accurately as possible by using a training set of completely labeled full length time series. An example application of this task is trying to identify which household device is working at a given time by using the electricity usage profiles [3].…”

“…Also, Kadous and Sammut [4] use electrocardiography (ECG) data to predict whether a patient has heart disease or not and also to recognize sign language signs. Pei et al [2] use image sequences for facial expression detection, Ye and Keogh [5] use time series classification methods to identify coffee and wheat varieties by using spectrography series, and Pei et al [2] and Li et al [6] perform motion identification by using multivariate sensor readings, which can also be interpreted as time series.…”

“…. , (T S n , CL n )}, the aim is to build a model that is able to predict the class label of new unlabeled time series as accurately as possible [1,2,9,14].…”

Multivariate LSTM-FCNs for time series classification

“…To overcome this, HULM proposes using H binary stochastic hidden units to model 2 H latent structures of the data with only O(H) parameters. Results indicate HULM outperforming HCRF on most datasets [18].…”

“…For this reason, distance-based approaches are more successful in classifying multivariate time series data [17]. Hidden State Conditional Random Field (HCRF) and Hidden Unit Logistic Model (HULM) are two successful feature-based algorithms which have led to state-of-the-art results on various benchmark datasets, ranging from online character recognition to activity recognition [18]. HCRF is a computationally expensive algorithm that detects latent structures of the input time series data using a chain of k-nominal latent variables.…”