SIERRA - 2018 - Annual activity report

SIERRA

SIERRA - 2018

Project-Team Sierra

Team, Visitors, External Collaborators

Overall Objectives

Statement

Research Program

Supervised Learning
Unsupervised Learning
Parsimony
Optimization

Application Domains

Applications for Machine Learning

Highlights of the Year

New Software and Platforms

ProxASAGA
object-states-action

New Results

On the Global Convergence of Gradient Descent for Over-parameterized Models using Optimal Transport
Sharp Analysis of Learning with Discrete Losses
Gossip of Statistical Observations using Orthogonal Polynomials
Marginal Weighted Maximum Log-likelihood for Efficient Learning of Perturb-and-Map models
Slice inverse regression with score functions
Constant Step Size Stochastic Gradient Descent for Probabilistic Modeling
Nonlinear Acceleration of Momentum and Primal-Dual Algorithms
Nonlinear Acceleration of Deep Neural Networks
Nonlinear Acceleration of CNNs
Robust Seriation and Applications To Cancer Genomics
Reconstructing Latent Orderings by Spectral Clustering
Lyapunov Functions for First-Order Methods: Tight Automated Convergence Guarantees
Efficient First-order Methods for Convex Minimization: a Constructive Approach
Operator Splitting Performance Estimation: Tight contraction factors and optimal parameter selection
Finite-sample Analysis of M-estimators using Self-concordance
Uniform regret bounds over $R^{d}$ for the sequential linear regression problem with the square loss
Efficient online algorithms for fast-rate regret bounds under sparsity.
Exponential convergence of testing error for stochastic gradient methods
Statistical Optimality of Stochastic Gradient Descent on Hard Learning Problems through Multiple Passes
Central Limit Theorem for stationary Fleming–Viot particle systems in finite spaces
SeaRNN: Improved RNN training through Global-Local Losses
Improved asynchronous parallel optimization analysis for stochastic incremental methods
Asynchronous optimisation for Machine Learning
$M^{*}$ -Regularized Dictionary Learning
Optimal Algorithms for Non-Smooth Distributed Optimization in Networks
Relating Leverage Scores and Density using Regularized Christoffel Functions
Averaging Stochastic Gradient Descent on Riemannian Manifolds
Localized Structured Prediction
Optimal rates for spectral algorithms with least-squares regression over Hilbert spaces
Differential Properties of Sinkhorn Approximation for Learning with Wasserstein Distance
Learning with SGD and Random Features
Manifold Structured Prediction
On Fast Leverage Score Sampling and Optimal Learning
Accelerated Decentralized Optimization with Local Updates for Smooth and Strongly Convex Objectives

Bilateral Contracts and Grants with Industry

Bilateral Contracts with Industry
Bilateral Grants with Industry

Partnerships and Cooperations

National Initiatives
European Initiatives
International Initiatives
International Research Visitors

Dissemination

Promoting Scientific Activities
Teaching - Supervision - Juries
Popularization

Bibliography

Publications of the year

Previous |

Home | Next next

Section: New Results

Lyapunov Functions for First-Order Methods: Tight Automated Convergence Guarantees

In [21], we present a novel way of generating Lyapunov functions for proving linear convergence rates of first-order optimization methods. Our approach provably obtains the fastest linear convergence rate that can be verified by a quadratic Lyapunov function (with given states), and only relies on solving a small-sized semidefinite program. Our approach combines the advantages of performance estimation problems and integral quadratic constraints, and relies on convex interpolation.

Previous |

Home | Next next