Publications associées à Stochastic Second-Order Methods Improve Best-Known Sample Complexity of SGD for Gradient-Dominated Function

On the Generalization of Stochastic Gradient Descent with Momentum

While momentum-based accelerated variants of stochastic gradient descent (SGD) are widely used when training machine learning models, there is little theoretical understanding on the generalization error of such methods. In this work, we first show that th ...

Brookline2024

Understanding generalization and robustness in modern deep learning

Maksym Andriushchenko

In this thesis, we study two closely related directions: robustness and generalization in modern deep learning. Deep learning models based on empirical risk minimization are known to be often non-robust to small, worst-case perturbations known as adversari ...

EPFL2024

Leveraging Continuous Time to Understand Momentum When Training Diagonal Linear Networks

Nicolas Henri Bernard Flammarion, Hristo Georgiev Papazov, Scott William Pesme

In this work, we investigate the effect of momentum on the optimisation trajectory of gradient descent. We leverage a continuous-time approach in the analysis of momentum gradient descent with step size

\gamma

and momentum parameter

\beta

that allows u ...

2024

Universal and adaptive methods for robust stochastic optimization

Ali Kavis

Within the context of contemporary machine learning problems, efficiency of optimization process depends on the properties of the model and the nature of the data available, which poses a significant problem as the complexity of either increases ad infinit ...

EPFL2023

Augmented Lagrangian Methods for Provable and Scalable Machine Learning

Mehmet Fatih Sahin

Non-convex constrained optimization problems have become a powerful framework for modeling a wide range of machine learning problems, with applications in k-means clustering, large- scale semidefinite programs (SDPs), and various other tasks. As the perfor ...

EPFL2023

End-to-End Learning for Stochastic Optimization: A Bayesian Perspective

Daniel Kuhn, Yves Rychener, Tobias Sutter

We develop a principled approach to end-to-end learning in stochastic optimization. First, we show that the standard end-to-end learning algorithm admits a Bayesian interpretation and trains a posterior Bayes action map. Building on the insights of this an ...

2023

Stochastic adaptive resampling for the estimation of discrete choice models

Michel Bierlaire, Nicola Marco Ortelli, Matthieu Marie Cochon de Lapparent

In the field of choice modeling, the availability of ever-larger datasets has the potential to significantly expand our understanding of human behavior, but this prospect is limited by the poor scalability of discrete choice models (DCMs): as sample sizes ...

2023

The statistical complexity of early-stopped mirror descent

Tomas Vaskevicius, Varun Kanade

Recently there has been a surge of interest in understanding implicit regularization properties of iterative gradient-based optimization algorithms. In this paper, we study the statistical guarantees on the excess risk achieved by early-stopped unconstrain ...

Oxford2023

Byzantine Fault-Tolerance in Federated Local SGD Under 2f-Redundancy

Nirupam Gupta

In this article, we study the problem of Byzantine fault-tolerance in a federated optimization setting, where there is a group of agents communicating with a centralized coordinator. We allow up to

f

Byzantine-faulty agents, which may not follow a prescr ...

Piscataway2023

Alternation makes the adversary weaker in two-player games

Volkan Cevher, Efstratios Panteleimon Skoulakis, Luca Viano, Ali Kavis

Motivated by alternating game-play in two-player games, we study an altenating variant of the Online Linear Optimization (OLO). In alternating OLO, a learner at each round t ∈[n] selects a vector xt and then an adversary selects a cost-vector ct ∈[−1,1]n. ...

2023

Stochastic Second-Order Methods Improve Best-Known Sample Complexity of SGD for Gradient-Dominated Function

Graph Chatbot

Chattez avec Graph Search

On the Generalization of Stochastic Gradient Descent with Momentum

Understanding generalization and robustness in modern deep learning

Leveraging Continuous Time to Understand Momentum When Training Diagonal Linear Networks

Universal and adaptive methods for robust stochastic optimization

Augmented Lagrangian Methods for Provable and Scalable Machine Learning

End-to-End Learning for Stochastic Optimization: A Bayesian Perspective

Stochastic adaptive resampling for the estimation of discrete choice models

The statistical complexity of early-stopped mirror descent

Byzantine Fault-Tolerance in Federated Local SGD Under 2f-Redundancy

Alternation makes the adversary weaker in two-player games

End-to-End Learning for Stochastic Optimization: A Bayesian Perspective

Stochastic adaptive resampling for the estimation of discrete choice models

The statistical complexity of early-stopped mirror descent

Understanding generalization and robustness in modern deep learning

Universal and adaptive methods for robust stochastic optimization

Augmented Lagrangian Methods for Provable and Scalable Machine Learning

On the Generalization of Stochastic Gradient Descent with Momentum

Byzantine Fault-Tolerance in Federated Local SGD Under 2f-Redundancy

Alternation makes the adversary weaker in two-player games

Leveraging Continuous Time to Understand Momentum When Training Diagonal Linear Networks