GTC 2020: Toward an Autonomous-and-Safe Deep Learning Framework for Time Series Data
After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
Toward an Autonomous-and-Safe Deep Learning Framework for Time Series Data
Jian Chang, Alibaba Group | Sanjian Chen, Alibaba Group
Global retailers, such as Alibaba, generate hundreds of petabytes of time-series data every day. Making intelligent decisions based upon such time-series data is essential to many business units in Alibaba. We'll discuss our approach of combining two major themes — "autonomous" and "safe" — in a unified machine-learning framework for time-series data. To improve the accuracy of time-series forecasting and classification, we leverage graph neural networks (GNN) to model the complexity of multi-variant time-series. To improve the computation efficiency, we borrow ideas from Transformer and WaveNet architecture to enable parallel forecast generation. The vulnerability of deep neural networks to adversarial attacks has also generated a lot of attention and discussion on the topic of “safe AI”. We'll also discuss the importance and challenges in developing “safe” deep-learning models for time-series, and how we can defend against those attacks.