Developing cognitive IoT solutions for anomaly detection using deep learning

Overview

Anomaly detection is a process in machine learning that identifies data points, events, or observations in time-series data that deviate from the majority of the data and does not conform to normal behavior. To build an anomaly detection solution for time series data, a cognitive IoT solution, you can use Keras and TensorFlow to create an unsupervised deep learning neural network for anomaly detection on time-series data.

This learning path will introduce deep learning and long-short term memory networks (neural networks) and autoencoders, show you how to create a test physical model based data generator for anomaly detection, and finally show you how to create a deep learning neural network for anomaly detection on time-series data using Keras and TensorFlow.

Skill level

Intermediate.

Estimated time to complete

Approximately 2 hours

Learning objectives

Upon completion of this learning path, you will:

• Understand how deep learning and an LSTM network can outperform state-of-the-art anomaly detection algorithms on time-series sensor data – or any type of sequence data in general.
• Use Node-RED and the Lorenz Attractor Model to generate realistic test data by sampling from a physical model.
• How to use TensorFlow to accelerate linear algebra operations by optimizing executions and understand how Keras provides an accessible framework on top of TensorFlow.
• Create an unsupervised LSTM autoencoder neural network using TensorFlow and Keras.
• How to containerize that neural network and put it behind an HTTP service API for easy consumption and scaling on Docker, Kubernetes, and Knative using IBM Cloud Code Engine.