Posted: 21st April 2016

Professor Nando de Freitas, one of the leaders in the collaborative efforts between Oxford Computer Science and Google Deepmind, is one of the recipients of the best paper award at ICLR 2016, one of the premier conferences on deep learning. The paper introduces a compositional neural computer that learns programs and how to interpret them. It even learns to code. The video below provides an overview. The following figures show how the same neural programmer and interpreter (NPI) is able to execute several programs, including addition, sorting, and interpretation of images and planning. The figures also show the code produced by the machine, which provides a way for humans to understand what the machine is doing.