Learning Financial Data and Recognizing Images with TensorFlow and Neural Networks

by Sophia TurolAugust 5, 2016
Explore how to employ advancements in the recurrent neural networks and TensorFlow to learn patterns in financial markets more efficiently.


At the recent TensorFlow meetup in London, the speakers overviewed the recent advances in machine learning domain, as well as discussed what transfer learning is and how to employ it for image recognition with TensorFlow.


Using recurrent autoencoders to learn financial data

In his session, Steven Hutt of CME Group provided an introduction to learning sequences using recurrent neural networks (RNN) and TensorFlow. He described how to make use of the latest advances in RNNs to learn patterns in market data:

  • Unitary RNNs
  • RNNs as programs to address external memory
  • Deep RNNs
  • Autoencoders and recurrent autoencoders
  • Easy RNNs with TensorFlow VariableScope
  • TensorFlow encoder
  • TensorFlow deep recurrent autoencoder

Below, you can check out the full slides by Steven.


Image recognition using TensorFlow and containers

Yaz Santissi provided a live demo, highlighting the following aspects:

  • Setting up TensorFlow within Ubuntu containers
  • What is transfer learning and how to get an existing model with it?
  • Training old models with new images
  • Testing new models with new images

Below, you can check out the full slides by Yaz.


Want more detail? Watch the video!



Further reading


About the experts

Steven Hutt is a consultant for deep learning and financial risks, currently working for CME Group. He has a bachelor’s degree in mathematics from Imperial College London and a PhD from the University of Edinburgh. Steven’s job is currently focused on the application of deep learning to financial data analysis, as well as the implementation of recurrent neural networks and variational autoencoders for pattern recognition and anomaly detection.


Yaz Santissi is a freelancer, developing machine learning algorithms as proof-of-concepts for a variety of companies. Despite working in machine learning, he remains a steadfast fan of human learning. Being part of London’s Google Developer Group, Yaz has organized a large number of TensorFlow events around the world, where participants can share what they have learned and build projects together.