On the omnipresence of spurious local minima in certain neural network training problems

DOI10.1007/s00365-023-09658-wMaRDI QIDQ6629537

Unnamed Author, Constantin Christof

Publication date: 30 October 2024

Published in: Constructive Approximation (Search for Journal in Brave)

zbMATH Keywords

stability analysisbest approximationHadamard well-posednesstraining problemloss landscapedeep artificial neural networklocal affine linearityspurious local minimum

Mathematics Subject Classification ID

Artificial neural networks and deep learning (68T07) Sensitivity, stability, well-posedness (49K40) Sensitivity, stability, parametric optimization (90C31) Variants of convex sets (star-shaped, ((m, n))-convex, etc.) (52A30)

Related Items (1)

On the existence of minimizers in shallow residual ReLU neural network optimization landscapes

Cites Work

• Unnamed Item
• Unnamed Item
• Unnamed Item
• Unnamed Item
• Unnamed Item
• Unnamed Item
• Lower bounds for approximation by MLP neural networks
• Topological properties of the set of functions generated by neural networks of fixed size
• On the nonuniqueness and instability of solutions of tracking-type optimal control problems
• Inverse and ill-posed problems. Theory and applications.
• Almost convex and Chebyshev sets
• On a Property of Metric Projections Onto Closed Subsets of Hilbert Spaces
• An Introduction to Banach Space Theory
• Training a Single Sigmoidal Neuron Is Hard
• On the Benefit of Width for Neural Networks: Disappearance of Basins
• Introduction to Functional Analysis
• Plateau Phenomenon in Gradient Descent Training of RELU Networks: Explanation, Quantification, and Avoidance
• Spurious Valleys in Two-layer Neural Network Optimization Landscapes
• Real Analysis
• Approximation by superpositions of a sigmoidal function
• Continuity of approximation by neural networks in \(L_p\) spaces

This page was built for publication: On the omnipresence of spurious local minima in certain neural network training problems