GTC 2020 S21454
Presenters: Stephane Rion,Teradata; Amgad Muhammad,Teradata
A vast amount of aircraft sensor data still remains unexploited due to its volume and shear complexity. In 2019, Airbus released, as part of a four-month long challenge, a 65+ gigabyte dataset recorded from real aircraft systems. The goal was to detect a set of anomalies from an unlabeled multidimension time series dataset. We’ll describe our solution, which is based on a two-stage approach using auto encoders and long short-term memory neural networks, and how we reached third place out of 160 teams competing. You’ll learn the benefits of using autoencoder to achieve dimension reduction in a clustering problem and how LSTM-based neural networks can be applied to detect anomalies in an unsupervised way. We’ll dive into the technical details of the solution and discuss the results obtained, as well as potential next steps. Basic knowledge of deep learning-techniques such as Autoencoder or LSTM will help, but isn’t required.
Watch this session
Join in the conversation below.