However, it is well known that deep learning models are data- intensive, and getting access to labeled training data is expensive. For time series, deep recurrent neural networks (RNNs) have been shown to perform hierarchical processing with different layers tackling different time scales. These features or representations have even been shown to outperform models heavily tuned for the specific tasks. Noticeably, deep Convolutional Neural Networks (CNNs) trained on millions of images from 1000 object classes have been used as off-the-shelf feature extractors to yield powerful generic image descriptors for a diverse range of tasks such as image classification, scene recognition and image retrieval. 1 Introduction Recently, fixed-dimensional vector representations for sequences of words in the form of sentences, paragraphs, and documents have been successfully used for natural language processing tasks such as machine translation and sentiment analysis. For several publicly available datasets from UCR TSC Archive and an industrial telematics sensor data from vehicles, we observe that a classifier learned over the TimeNet embeddings yields significantly better performance compared to (i) a classifier learned over the embeddings given by a domain-specific RNN, as well as (ii) a nearest neighbor classifier based on Dynamic Time Warping. The representations or embeddings given by a pre-trained TimeNet are found to be useful for time series classifica- tion (TSC). Once trained, TimeNet can be used as a generic off-the-shelf feature extractor for time series. Rather than relying on data from the problem domain, TimeNet attempts to generalize time series representa- tion across domains by ingesting time series from several domains simul- taneously. Inspired by the tremendous success of deep Convolutional Neu- ral Networks as generic feature extractors for images, we propose TimeNet : a deep recurrent neural network (RNN) trained on diverse time series in an unsupervised manner using sequence to sequence (seq2seq) models to extract features from time series. TimeNet: Pre-trained deep recurrent neural network for time series classification Pankaj Malhotra, Vishnu TV, Lovekesh Vig, Puneet Agarwal, Gautam Shroff TCS Research, New Delhi, India Abstract.
0 Comments
Leave a Reply. |