Artificial Intelligence and Deep Learning nowadays are ever growing subjects. People who research AI, are experimenting with various intelligence models in order to discover a strong intelligence model which is equally capable with the unknown human model. Prototyping of such models (as large as they may be) needs to be as fast as possible, in order to examine exhaustively many novel ideas as fast as it can be done. For this we need to be able to use with ease as many resources as we have possible. As Theano is contributing to this attempt, current GPU technologies need to be supported extensively in a transient way through its framework. This can be achieved by including optimized operations from current NVidia frameworks, as well as optimizing or designing GPU code which resides in the framework’s backend. Furthermore, analysis and optimization of multi-gpu code, in addition with a consideration of a multi-node/gpu design, would prove themselves useful for this cause.




  • Frédéric Bastien
  • abergeron