Convergence and concentration properties of constant step-size SGD through Markov chains.
Ibrahim MeradStéphane GaïffasPublished in: CoRR (2023)
Keyphrases
- markov chain
- step size
- convergence rate
- stochastic gradient descent
- convergence speed
- faster convergence
- variable step size
- steady state
- transition probabilities
- line search
- finite state
- stationary distribution
- markov processes
- monte carlo
- stochastic process
- markov model
- markov process
- random walk
- state space
- global convergence
- cost function
- learning rate
- importance sampling
- probabilistic automata
- transition matrix
- differential evolution
- non stationary
- machine learning
- pso algorithm
- particle swarm optimization
- convergence analysis
- image quality
- objective function