Presented By
O’Reilly + Cloudera
Make Data Work
29 April–2 May 2019
London, UK
Please log in

Sequence-to-sequence modeling for time series

Arun Kejariwal (Independent), Ira Cohen (Anodot)
12:0512:45 Wednesday, 1 May 2019
Data Science, Machine Learning & AI
Location: Capital Suite 17
Average rating: ****.
(4.00, 5 ratings)



What you'll learn

  • Learn how to leverage sequence-to-sequence modeling for time series data


Sequence-to-sequence (seq2seq) modeling using neural networks has been increasingly becoming mainstream in the recent years. In particular, it has been leveraged for applications such as, but not limited to, speech recognition, language translation, and question answering. More recently, seq2seq has also been used for applications based on time series data, such as seq2seq modeling-based real-time anomaly detection and forecasting.

Arun Kejariwal and Ira Cohen offer an overview seq2seq and explore its early use cases. They then walk you through leveraging seq2seq modeling for these use cases, particularly with regard to real-time anomaly detection and forecasting. You’ll learn how to use a multilayered LSTM to encode the input time series and a deep LSTM to decode. For anomaly detection, the output is married with “traditional” statistical approaches for anomaly detection. Conceivably, any of the many variants of LSTM or RNN alternatives of LSTM can be used to trade off accuracy and speed. However, LSTMs operate sequentially and are quite slow to train. Arun and Ira shed light on how to leverage architectures such as convolutional neural networks (CNN) and self-attention networks (SAN) to achieve significant improvements in accuracy. They conclude with a concrete case study that illustrates the use of seq2seq for both real-time anomaly detection and forecasting for time series data.

Photo of Arun Kejariwal

Arun Kejariwal


Arun Kejariwal is an independent lead engineer. Previously, he was he was a statistical learning principal at Machine Zone (MZ), where he led a team of top-tier researchers and worked on research and development of novel techniques for install-and-click fraud detection and assessing the efficacy of TV campaigns and optimization of marketing campaigns, and his team built novel methods for bot detection, intrusion detection, and real-time anomaly detection; and he developed and open-sourced techniques for anomaly detection and breakout detection at Twitter. His research includes the development of practical and statistically rigorous techniques and methodologies to deliver high performance, availability, and scalability in large-scale distributed clusters. Some of the techniques he helped develop have been presented at international conferences and published in peer-reviewed journals.

Photo of Ira Cohen

Ira Cohen


Ira Cohen is a cofounder and chief data scientist at Anodot, where he’s responsible for developing and inventing the company’s real-time multivariate anomaly detection algorithms that work with millions of time series signals. He holds a PhD in machine learning from the University of Illinois at Urbana-Champaign and has over 12 years of industry experience.