Pilot Project 2
Neural Network Approximation
Team: R. DeVore, K. Narayanan, G. Petrova, J. Siegel (lead)
Synopsizing points
Understand the set of problems in data science/scientific computing where neural networks beat other methods of function approximation
Identify model classes for which approximation properties can be shown for various types of neural networks (ResNet, DenseNet, etc.)
Characterize the functions in a given approximation class and assess how well these classes represent realistic functions
Realize the approximation methods algorithmically and rigorously analyze approaches such as stochastic gradient and greedy methods
Related grants
NSF-DMS: Neural Networks for Stationary and Evolutionary Variational Problems, 2023-26 (PI: Stephan Wojtowytsch)
NSF-CCF CIF: Small: Interpretable Machine Learning based on Deep Neural Networks: a Source Coding Perspective, 2022-25 (local co-PI: Siegel)
NSF SCALE-MoDL: New Perspectives on Deep Learning: Bridging Approximation, Statistical, and Algorithmic Theories, 2021-24 (local PI: Petrova, local coPI: DeVore)
ONR MURI: Theoretical Foundations of Deep Learning, 2020-23 (local PI: DeVore, local coPIs: Foucart, Petrova)
Recent relevant papers
R. DeVore, R. D. Nowak, R. Parhi, J. W. Siegel. Weighted variation spaces and approximation by shallow ReLU networks. (arXiv)
J. W. Siegel. Optimal approximation of zonoids and uniform approximation by shallow neural networks. (arXiv)
J. W. Siegel, S. Wojtowytsch. A qualitative difference between gradient flows of convex functions in finite-and infinite-dimensional Hilbert spaces. (arXiv)
G. Petrova, P. Wojtaszczyk. Neural networks: deep, shallow, or in between? (arXiv)
K. Gupta, J. W. Siegel, S. Wojtowytsch. Achieving acceleration despite very noisy gradients. (arXiv)
J. W. Siegel. Sharp lower bounds on interpolation by deep ReLU neural networks at irregularly spaced data. (arXiv)
J. W. Siegel. Optimal approximation rates for deep ReLU neural networks on Sobolev and Besov spaces. Journal of Machine Learning Research, 24/357, 1-52, 2023. (link)
S. Wojtowytsch. Optimal bump functions for shallow ReLU networks: Weight decay, depth separation and the curse of dimensionality. Journal of Machine Learning Research, 25/27, 1−49, 2024. (link)
S. Wojtowytsch, J. Park. Qualitative neural network approximation over R and C: Elementary proofs for analytic and polynomial activation. (arXiv)
J. W. Siegel, J. Xu. Sharp bounds on the approximation rates, metric entropy, and n-widths of shallow neural networks. Foundations of Computational Mathematics, 24, 481-537, 2024.. (doi)
I. Daubechies, R. DeVore, N. Dym, S. Faigenbaum-Golovin, S. Kovalsky, K.-C. Lin, J. Park, G. Petrova, B. Sober. Neural network approximation of refinable functions. IEEE Transactions on Information Theory, 69/1, 482-495, 2023. (doi)
J. W. Siegel, J. Xu. High-order approximation rates for shallow neural networks with cosine and ReLUk activation functions. Applied and Computational Harmonic Analysis, 58, 1-26, 2022. (doi)
J. W. Siegel, J. Xu. Optimal convergence rates for the orthogonal greedy algorithm. IEEE Transactions on Information Theory, 68/5, 3354-3361, 2022. (doi)
A. Cohen, R. DeVore, G. Petrova, P. Wojtaszczyk. Optimal stable nonlinear approximation. Foundations of Computational Mathematics, 22, 607–648, 2022. (doi)
I. Daubechies, R. DeVore, S. Foucart, B. Hanin, G. Petrova. Nonlinear approximation and (deep) ReLU networks. Constructive Approximation, 55, 127-172, 2022. (doi)
R. DeVore, B. Hanin, G. Petrova. Neural network approximation. Acta Numerica 30, 327-444, 2021. (doi)