What Is Behind Deep Reinforcement Learning and Transfer Learning with TensorFlow?

by Sophia TurolNovember 9, 2016
This blog post discusses deep reinforcement learning, the use of artificial neurons in TensorFlow, as well as training a model with little data through transfer learning.


At the recent TensorFlow meetup in Madrid, the speakers explored the concept of deep reinforcement learning and learnt how to train a model with little data available.


The concept of deep reinforced learning

Gema Parreño Piqueras of Tetuan Valley started her session with explaining what deep reinforced learning is. According to her, it comprises two subfields of artificial intelligence. Implying “teaching through rewards,” reinforcement learning is responsible for decision making, while deep learning is based on a combination of mathematical possibilities.

”Artificial intelligence uses the advantages of both the reward method and computational power maths gives you.” —Gema Piqueras, Tetuan Valley

Gema also introduced the audience to the concept of an artificial neuron, which is an input with an assigned transfer function to it. Then, an activation function is assigned to a transfer to one and you get an output.

Moving to TensorFlow’s architecture, she explored its major components:

  • Tensors that embody data structure
  • Graphs responsible for graphic representation of the computational process
  • Variables that help to define the structure of the neural model
  • Neural networks dealing with complex tasks

Finally, Gema provided a demo to exemplify how image classification works with scikit-learn and TensorFlow.

Watch the video for more details.



Below is Gema’s full presentation from the meetup.


Training a model with little data

Training a model with little data available may pose certain issues with the output accuracy. Gorka Bengochea of Ibermática demonstrated how to address this through transfer learning.

Watch the video to get more insights.



You can also look through Gorka’s presentation below.


Join our group to stay tuned with the upcoming events.


Further reading


About the experts

Gema Parreño Piqueras is Software Architect at Tetuan Valley. She is an awarded product designer with two years of experience in artificial intelligence, specifically natural language understanding. Gema is currently developing recursive neural networks and clustering classifications with TensorFlow.


Gorka Bengochea is Deep Learning Developer at Ibermática. He is passionate about artificial intelligence, big data, and automation. Gorka is currently focused on developing artificial intelligence–based computer vision using TensorFlow and training models through transfer learning. He is also studying robotics and automation at the Technical University of Madrid.