GTC 2020: When Neural Networks Meet Conventional Learning and Optimization Methods: Learning Radio Maps for Enhanced Physical-Layer Security
After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
When Neural Networks Meet Conventional Learning and Optimization Methods: Learning Radio Maps for Enhanced Physical-Layer Security
Slawomir Stanczak, Heinrich-Hertz-Institute and Technical University of Berlin
We'll consider neural network architectures inspired by iterative algorithms that are widely used to solve convex optimization problems in many fields, including wireless communications. We refer to such neural networks as iterative neural networks based on Mann-type iteration. In particular, we'll focus on optimization problems where the objective function can be written as the sum of a convex smooth function and a non-smooth regularizer function. This problem formulation includes the important special case of the reconstruction of sparse signals where the non-smooth function is related to the l1-norm. Sparse signals occur in many optimization problems in wireless networks, such as channel estimation or reconstruction of quality-of-service (QoS) maps. We'll focus on the problem of reconstructing secrecy maps by learning any-to-any radio maps.