Gradient Descent Only Converges to Minimizers: Non-Isolated Critical Points and Invariant Regions

From MaRDI portal
Publication:4638050

DOI10.4230/LIPIcs.ITCS.2017.2zbMath1402.90210arXiv1605.00405MaRDI QIDQ4638050

Georgios Piliouras, Ioannis Panageas

Publication date: 3 May 2018

Full work available at URL: https://arxiv.org/abs/1605.00405




Related Items

Analysis of Asymptotic Escape of Strict Saddle Sets in Manifold OptimizationModel-free Nonconvex Matrix Completion: Local Minima Analysis and Applications in Memory-efficient Kernel PCAConvergence guarantees for a class of non-convex and non-smooth optimization problemsOn initial point selection of the steepest descent algorithm for general quadratic functionsProximal methods avoid active strict saddles of weakly convex functionsLandscape analysis for shallow neural networks: complete classification of critical points for affine target functionsUnnamed ItemA proof of convergence for stochastic gradient descent in the training of artificial neural networks with ReLU activation for constant target functionsGlobal convergence of the gradient method for functions definable in o-minimal structuresPolynomial‐time universality and limitations of deep learningStatistical Inference with Local OptimaA geometric approach of gradient descent algorithms in linear neural networksRun-and-inspect method for nonconvex optimization and global optimality bounds for R-local minimizersFirst-order methods almost always avoid strict saddle pointsSufficient Conditions for Instability of the Subgradient Method with Constant Step SizeInertial Newton algorithms avoiding strict saddle pointsA Newton-Based Method for Nonconvex Optimization with Fast Evasion of Saddle PointsStochastic optimization with momentum: convergence, fluctuations, and traps avoidanceBacktracking gradient descent method and some applications in large scale optimisation. II: Algorithms and experimentsUnnamed ItemUnnamed ItemA geometric analysis of phase retrievalMultiscale sparse microcanonical modelsMutation, Sexual Reproduction and Survival in Dynamic EnvironmentsGradient Descent Only Converges to Minimizers: Non-Isolated Critical Points and Invariant RegionsNull space gradient flows for constrained optimization with applications to shape optimizationExtending the Step-Size Restriction for Gradient Descent to Avoid Strict Saddle PointsOn Gradient-Based Learning in Continuous Games



Cites Work