Renormalization insight into deep learning
Seminar
|
The analogy between machine learning in deep neural networks and renormalization is critically discussed beyond the qualitative level. Formal similarities and differences are examined to the aim of bridging the two disciplines, and thus make progress in the understanding and optimization of neural networks with the assistance of methods and insights developed in physics. The renormalization framework and other techniques, mainly borrowed from statistical mechanics and information theory, prove useful in applications, even before being neatly established or suitably generalized to the machine learning context. The talk does not assume previous knowledge in machine learning. Join Zoom Meeting https://cern.zoom.us/j/65443866152?pwd=elRUMWJXMU5zMUVIcm1JcFlRV2E1UT09
Meeting ID: 654 4386 6152 Passcode: 730462 |