Exit Time Analysis for Approximations of Gradient Descent Trajectories Around Saddle Points
نویسندگان
چکیده
Abstract This paper considers the problem of understanding exit time for trajectories gradient-related first-order methods from saddle neighborhoods under some initial boundary conditions. Given ‘flat’ geometry around points, can struggle to escape these regions in a fast manner due small magnitudes gradients encountered. In particular, while it is known that strict-saddle neighborhoods, existing analytic techniques do not explicitly leverage local points order control behavior gradient trajectories. It this context puts forth rigorous geometric analysis gradient-descent method using matrix perturbation theory. doing so, provides key result be used generate an approximate trajectory any given addition, leads linear exit-time solution certain necessary conditions, which bring out dependence on dimension, conditioning neighborhood, and more, class functions.
منابع مشابه
Accelerated Gradient Descent Escapes Saddle Points Faster than Gradient Descent
Nesterov's accelerated gradient descent (AGD), an instance of the general family of"momentum methods", provably achieves faster convergence rate than gradient descent (GD) in the convex setting. However, whether these methods are superior to GD in the nonconvex setting remains open. This paper studies a simple variant of AGD, and shows that it escapes saddle points and finds a second-order stat...
متن کاملGradient Descent Can Take Exponential Time to Escape Saddle Points
Although gradient descent (GD) almost always escapes saddle points asymptotically [Lee et al., 2016], this paper shows that even with fairly natural random initialization schemes and non-pathological functions, GD can be significantly slowed down by saddle points, taking exponential time to escape. On the other hand, gradient descent with perturbations [Ge et al., 2015, Jin et al., 2017] is not...
متن کاملEscaping From Saddle Points - Online Stochastic Gradient for Tensor Decomposition
We analyze stochastic gradient descent for optimizing non-convex functions. In many cases for non-convex functions the goal is to find a reasonable local minimum, and the main concern is that gradient updates are trapped in saddle points. In this paper we identify strict saddle property for non-convex problem that allows for efficient optimization. Using this property we show that from an arbit...
متن کاملBiased gradient squared descent saddle point finding method.
The harmonic approximation to transition state theory simplifies the problem of calculating a chemical reaction rate to identifying relevant low energy saddle points in a chemical system. Here, we present a saddle point finding method which does not require knowledge of specific product states. In the method, the potential energy landscape is transformed into the square of the gradient, which c...
متن کاملHamiltonian trajectories and saddle points in mathematical economics
Abstract: Infinite-horizon problems of kinds that arise in macroeconomic applications present a challenge in optimal control which has only partially been met. Results from the theory of convex problems of Lagrange can be utilized, to some extent, the most interesting feature being that in these problems the analysis revolves about a rest point of the Hamiltonian, which is at the same time a sa...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
ژورنال
عنوان ژورنال: Information and Inference: A Journal of the IMA
سال: 2022
ISSN: ['2049-8772', '2049-8764']
DOI: https://doi.org/10.1093/imaiai/iaac025