[كتاب][B] Control systems and reinforcement learning

S Meyn - 2022‏ - books.google.com
A high school student can create deep Q-learning code to control her robot, without any
understanding of the meaning of'deep'or'Q', or why the code sometimes fails. This book is …

On the optimization of deep networks: Implicit acceleration by overparameterization

S Arora, N Cohen, E Hazan - International conference on …, 2018‏ - proceedings.mlr.press
Conventional wisdom in deep learning states that increasing depth improves
expressiveness but complicates optimization. This paper suggests that, sometimes …

Acceleration methods

A d'Aspremont, D Scieur, A Taylor - Foundations and Trends® …, 2021‏ - nowpublishers.com
This monograph covers some recent advances in a range of acceleration techniques
frequently used in convex optimization. We first use quadratic optimization problems to …

Stochastic gradient descent performs variational inference, converges to limit cycles for deep networks

P Chaudhari, S Soatto - 2018 Information Theory and …, 2018‏ - ieeexplore.ieee.org
Stochastic gradient descent (SGD) is widely believed to perform implicit regularization when
used to train deep neural networks, but the precise manner in which this occurs has thus far …

[HTML][HTML] User-friendly guarantees for the Langevin Monte Carlo with inaccurate gradient

AS Dalalyan, A Karagulyan - Stochastic Processes and their Applications, 2019‏ - Elsevier
In this paper, we study the problem of sampling from a given probability density function that
is known to be smooth and strongly log-concave. We analyze several methods of …

Understanding the acceleration phenomenon via high-resolution differential equations

B Shi, SS Du, MI Jordan, WJ Su - Mathematical Programming, 2022‏ - Springer
Gradient-based optimization algorithms can be studied from the perspective of limiting
ordinary differential equations (ODEs). Motivated by the fact that existing ODEs do not …

Underdamped Langevin MCMC: A non-asymptotic analysis

X Cheng, NS Chatterji, PL Bartlett… - … on learning theory, 2018‏ - proceedings.mlr.press
We study the underdamped Langevin diffusion when the log of the target distribution is
smooth and strongly concave. We present a MCMC algorithm based on its discretization and …

Acceleration by stepsize hedging: Multi-step descent and the silver stepsize schedule

J Altschuler, P Parrilo - Journal of the ACM, 2023‏ - dl.acm.org
Can we accelerate the convergence of gradient descent without changing the algorithm—
just by judiciously choosing stepsizes? Surprisingly, we show that the answer is yes. Our …

Accelerated gradient descent escapes saddle points faster than gradient descent

C **, P Netrapalli, MI Jordan - Conference On Learning …, 2018‏ - proceedings.mlr.press
Nesterov's accelerated gradient descent (AGD), an instance of the general family of
“momentum methods,” provably achieves faster convergence rate than gradient descent …

Stochastic modified equations and adaptive stochastic gradient algorithms

Q Li, C Tai, E Weinan - International Conference on Machine …, 2017‏ - proceedings.mlr.press
We develop the method of stochastic modified equations (SME), in which stochastic gradient
algorithms are approximated in the weak sense by continuous-time stochastic differential …