Augmented neural networks

A novel model for supervised learning based on the multi-layer feedforward neural network, but where the activation functions are encoded via penalties in the training problem, effectively adapting activations to data. The new framework is particularly amenable to block-coordinate algorithms where each step is composed of simple, convex problems.

ERC Sector:

  • PE7_1 Control engineering
  • PE6_11  Machine learning, statistical data processing and applications using signal processing

Keywords:

  • Neural Networks
  • Optimization

Research groups

Contact