Back to Search
Start Over
Second-Order Guarantees in Centralized, Federated and Decentralized Nonconvex Optimization
- Publication Year :
- 2020
-
Abstract
- Rapid advances in data collection and processing capabilities have allowed for the use of increasingly complex models that give rise to nonconvex optimization problems. These formulations, however, can be arbitrarily difficult to solve in general, in the sense that even simply verifying that a given point is a local minimum can be NP-hard [1]. Still, some relatively simple algorithms have been shown to lead to surprisingly good empirical results in many contexts of interest. Perhaps the most prominent example is the success of the backpropagation algorithm for training neural networks. Several recent works have pursued rigorous analytical justification for this phenomenon by studying the structure of the nonconvex optimization problems and establishing that simple algorithms, such as gradient descent and its variations, perform well in converging towards local minima and avoiding saddle-points. A key insight in these analyses is that gradient perturbations play a critical role in allowing local descent algorithms to efficiently distinguish desirable from undesirable stationary points and escape from the latter. In this article, we cover recent results on second-order guarantees for stochastic first-order optimization algorithms in centralized, federated, and decentralized architectures.
- Subjects :
- Signal Processing (eess.SP)
FOS: Computer and information sciences
Mathematical optimization
Computer Science - Machine Learning
Optimization problem
Computer science
Machine Learning (stat.ML)
Machine Learning (cs.LG)
Statistics - Machine Learning
Convergence (routing)
FOS: Electrical engineering, electronic engineering, information engineering
FOS: Mathematics
Computer Science - Multiagent Systems
Electrical Engineering and Systems Science - Signal Processing
Mathematics - Optimization and Control
convergence
Artificial neural network
learning-behavior
Stationary point
Backpropagation
Maxima and minima
consensus
Optimization and Control (math.OC)
Key (cryptography)
Gradient descent
Multiagent Systems (cs.MA)
Subjects
Details
- Language :
- English
- Database :
- OpenAIRE
- Accession number :
- edsair.doi.dedup.....c90de33ecc2123af3ee2ecdd90fa456e