Publications related to Statistical learning theory

Boosting likelihood learning with event reweighting

Andrea Wulzer, Siyu Chen, Alfredo Glioti

Extracting maximal information from experimental data requires access to the likelihood function, which however is never directly available for complex experiments like those performed at high energy colliders. Theoretical predictions are obtained in this ...

Springer Nature2024

Data-Driven Control and Optimization under Noisy and Uncertain Conditions

Baiwei Guo

Control systems operating in real-world environments often face disturbances arising from measurement noise and model mismatch. These factors can significantly impact the perfor- mance and safety of the system. In this thesis, we aim to leverage data to de ...

EPFL2023

Towards Robust Monitoring of the Laser Powder Bed Fusion Process based on Acoustic Emission combined with Machine Learning Solutions

Rita Drissi Daoudi

Laser Powder Bed Fusion (LPBF) is an Additive Manufacturing (AM) process consolidating parts layer by layer, from a metallic powder bed. It allows no limitation in terms of geometry and is therefore of particular interest to various industries. Metallic LP ...

EPFL2023

A Statistical Framework to Investigate the Optimality of Signal-Reconstruction Methods

Michaël Unser, Pakshal Narendra Bohra, Pol del Aguila Pla

We present a statistical framework to benchmark the performance of reconstruction algorithms for linear inverse problems, in particular, neural-network-based methods that require large quantities of training data. We generate synthetic signals as realizati ...

2023

A Theory of Finite-Width Neural Networks: Generalization, Scaling Laws, and the Loss Landscape

Berfin Simsek

Deep learning has achieved remarkable success in various challenging tasks such as generating images from natural language or engaging in lengthy conversations with humans. The success in practice stems from the ability to successfully train massive neural ...

EPFL2023

The very knotty lenser: Exploring the role of regularization in source and potential reconstructions using Gaussian process regression

Georgios Vernardos

Reconstructing lens potentials and lensed sources can easily become an underconstrained problem, even when the degrees of freedom are low, due to degeneracies, particularly when potential perturbations superimposed on a smooth lens are included. Regulariza ...

OXFORD UNIV PRESS2022

Practical Byzantine-resilient Stochastic Gradient Descent

Sébastien Louis Alexandre Rouault

Algorithms are everywhere. The recipe for the frangipane cake is an algorithm. If all the listed ingredients are available and the cook is sufficiently deft, after a finite number of small, well-defined steps a delicious dessert will exit the oven. Now, wh ...

EPFL2022

Semantic Perturbations with Normalizing Flows for Improved Generalization

Martin Jaggi, Tatjana Chavdarova, Sebastian Urban Stich

Data augmentation is a widely adopted technique for avoiding overfitting when training deep neural networks. However, this approach requires domain-specific knowledge and is often limited to a fixed set of hard-coded transformations. Recently, several work ...

IEEE2021

Last iterate convergence of SGD for Least-Squares in the Interpolation regime

Nicolas Henri Bernard Flammarion, Aditya Vardhan Varre, Loucas Pillaud-Vivien

Motivated by the recent successes of neural networks that have the ability to fit the data perfectly \emph{and} generalize well, we study the noiseless model in the fundamental least-squares setup. We assume that an optimum predictor fits perfectly inputs ...

Advances in Neural Information Processing Systems 20212021

Landscape and training regimes in deep learning

Matthieu Wyart, Mario Geiger, Leonardo Petrini

Deep learning algorithms are responsible for a technological revolution in a variety oftasks including image recognition or Go playing. Yet, why they work is not understood.Ultimately, they manage to classify data lying in high dimension – a feat generical ...

2021

AL2: Progressive Activation Loss for Learning General Representations in Classification Neural Networks

Sabine Süsstrunk, Majed El Helou, Frederike Dümbgen

The large capacity of neural networks enables them to learn complex functions. To avoid overfitting, networks however require a lot of training data that can be expensive and time-consuming to collect. A common practical approach to attenuate overfitting i ...

IEEE2020

Wasserstein Distributionally Robust Learning

Soroosh Shafieezadeh Abadeh

Many decision problems in science, engineering, and economics are affected by uncertainty, which is typically modeled by a random variable governed by an unknown probability distribution. For many practical applications, the probability distribution is onl ...

EPFL2020

Wasserstein Distributionally Robust Optimization: Theory and Applications in Machine Learning

Daniel Kuhn, Viet Anh Nguyen, Soroosh Shafieezadeh Abadeh, Peyman Mohajerin Esfahani

Many decision problems in science, engineering and economics are affected by uncertain parameters whose distribution is only indirectly observable through samples. The goal of data-driven decision-making is to learn a decision from finitely many training s ...

2019

Regularization via Mass Transportation

Daniel Kuhn, Soroosh Shafieezadeh Abadeh, Peyman Mohajerin Esfahani

The goal of regression and classification methods in supervised learning is to minimize the empirical risk, that is, the expectation of some loss function quantifying the prediction error under the empirical distribution. When facing scarce training data, ...

2019

A Machine-Learning Framework for Automating Well-Log Depth Matching

Timon Zimmermann

Depth matching well logs acquired from multiple logging passes in a single well has been a longstanding challenge for the industry. The existing approaches employed in commercial platforms are typically based on classical cross-correlation and covariance m ...

SOC PETROPHYSICISTS & WELL LOG ANALYSTS-SPWLA2019

Non-intrusive reduced order modeling of nonlinear problems using neural networks

Jan Sickmann Hesthaven, Stefano Ubbiali

We develop a non-intrusive reduced basis (RB) method for parametrized steady-state partial differential equations (PDEs). The method extracts a reduced basis from a collection of high-fidelity solutions via a proper orthogonal decomposition (POD) and emplo ...

2018

Early Stages of Precipitation In Aluminum Alloys by First-Principles and Machine-Learning Atomistic Simulations

Daniele Giofré

Age hardening induced by the formation of (semi)-coherent precipitate phases is crucial for the processing and final properties of the widely used Al-6000 alloys despite the early stages of precipitation are still far from being fully understood. This cruc ...

EPFL2018

Optimized Distributed Hyperparameter Search and Simulation for Lung Texture Classification in CT Using Hadoop

Adrien Raphaël Depeursinge

Many medical image analysis tasks require complex learning strategies to reach a quality of image-based decision support that is sufficient in clinical practice. The analysis of medical texture in tomographic images, for example of lung tissue, is no excep ...

2016

Dendritic tree extraction from noisy maximum intensity projection images in C-elegans

Pascal Fua, Raphael Sznitman, Josué Sznitman

Background: Maximum Intensity Projections (MIP) of neuronal dendritic trees obtained from confocal microscopy are frequently used to study the relationship between tree morphology and mechanosensory function in the model organism C. elegans. Extracting den ...

Biomed Central Ltd2014

Fast Geodesic Active Fields for Image Registration Based on Splitting and Augmented Lagrangian Approaches

Jean-Philippe Thiran, Xavier Bresson, Dominique Zosso

In this paper, we present an efficient numerical scheme for the recently introduced geodesic active fields (GAF) framework for geometric image registration. This framework considers the registration task as a weighted minimal surface problem. Hence, the da ...

Ieee-Inst Electrical Electronics Engineers Inc2014