On the Banach spaces associated with multi-layer ReLU networks: Function representation, approximation theory and gradient descent dynamics
This work provides a theoretical foundation for deep learning by formalizing continuous models and gradient flow dynamics, which is foundational for the ML/AI community but incremental in building on existing approximation theory.
The authors tackled the problem of understanding the function spaces and generalization properties of deep ReLU networks by developing Banach spaces for finite-depth, infinite-width networks, showing that functions in these spaces can be approximated with dimension-independent rates and that the unit ball has low Rademacher complexity for favorable generalization.
We develop Banach spaces for ReLU neural networks of finite depth $L$ and infinite width. The spaces contain all finite fully connected $L$-layer networks and their $L^2$-limiting objects under bounds on the natural path-norm. Under this norm, the unit ball in the space for $L$-layer networks has low Rademacher complexity and thus favorable generalization properties. Functions in these spaces can be approximated by multi-layer neural networks with dimension-independent convergence rates. The key to this work is a new way of representing functions in some form of expectations, motivated by multi-layer neural networks. This representation allows us to define a new class of continuous models for machine learning. We show that the gradient flow defined this way is the natural continuous analog of the gradient descent dynamics for the associated multi-layer neural networks. We show that the path-norm increases at most polynomially under this continuous gradient flow dynamics.