A deterministic gradient-based approach to avoid saddle points
From MaRDI portal
Recommendations
- On Nonconvex Optimization for Machine Learning
- A Newton-based method for nonconvex optimization with fast evasion of saddle points
- Extending the Step-Size Restriction for Gradient Descent to Avoid Strict Saddle Points
- Gradient descent provably escapes saddle points in the training of shallow ReLU networks
- First-order methods almost always avoid strict saddle points
Cites work
- A geometric analysis of phase retrieval
- A Newton-based method for nonconvex optimization with fast evasion of saddle points
- A trust region algorithm with a worst-case iteration complexity of \(\mathcal{O}(\epsilon ^{-3/2})\) for nonconvex optimization
- Cubic regularization of Newton method and its global performance
- Exploiting negative curvature in deterministic and stochastic optimization
- Finding approximate local minima faster than gradient descent
- First-order methods almost always avoid strict saddle points
- Gradient descent finds the cubic-regularized nonconvex Newton step
- scientific article; zbMATH DE number 5060482 (Why is no real title available?)
- Laplacian smoothing stochastic gradient Markov chain Monte Carlo
- Learning deep architectures for AI
- Learning representations by back-propagating errors
This page was built for publication: A deterministic gradient-based approach to avoid saddle points
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6622959)
