arXiv subjects

Sylvain Rousseau

Publications and source records attributed to Sylvain Rousseau.

7 recordsLinked to original sources

arXiv2026

SPACR: Single-Pass Adaptive Training of Uncertainty-Aware Conformal Regressors

Conformal Prediction (CP) provides robust uncertainty guarantees for predictive models, but is typically applied post hoc, which misaligns model training with the conformal goal of producing efficient (i.e., narrow) intervals. We propose SPACR (Single-Pass Adaptive Conformal Regressor), a novel method for directly training uncertainty-aware regressors within a differentiable loss. SPACR jointly optimizes accuracy, efficiency, and validity without batch-splitting or a predefined confidence level during training. As a result, a single SPACR model yields valid prediction intervals at multiple confidence levels during inference, avoiding the costly retraining required by methods like Directly Optimized Inductive Conformal Regression (DOICR). Experiments on diverse tabular and image datasets show that SPACR consistently gives tighter intervals and better coverage-efficiency trade-offs compared to standard CP and DOICR, while significantly reducing computational costs relative to retraining-dependent baselines.

cs.LG↗

arXiv2025

Guaranteed prediction sets for functional surrogate models

We propose a method for obtaining statistically guaranteed prediction sets for functional machine learning methods: surrogate models which map between function spaces, motivated by the need to build reliable PDE emulators. The method constructs nested prediction sets on a low-dimensional representation (an SVD) of the surrogate model's error, and then maps these sets to the prediction space using set-propagation techniques. This results in prediction sets for functional surrogate models with conformal prediction coverage guarantees. We use zonotopes as basis of the set construction, which allow an exact linear propagation and are closed under Cartesian products, making them well-suited to this high-dimensional problem. The method is model agnostic and can thus be applied to complex Sci-ML models, including Neural Operators, but also in simpler settings. We also introduce a technique to capture the truncation error of the SVD, preserving the guarantees of the method.

cs.LG↗

arXiv2024

Harnessing Superclasses for Learning from Hierarchical Databases

In many large-scale classification problems, classes are organized in a known hierarchy, typically represented as a tree expressing the inclusion of classes in superclasses. We introduce a loss for this type of supervised hierarchical classification. It utilizes the knowledge of the hierarchy to assign each example not only to a class but also to all encompassing superclasses. Applicable to any feedforward architecture with a softmax output layer, this loss is a proper scoring rule, in that its expectation is minimized by the true posterior class probabilities. This property allows us to simultaneously pursue consistent classification objectives between superclasses and fine-grained classes, and eliminates the need for a performance trade-off between different granularities. We conduct an experimental study on three reference benchmarks, in which we vary the size of the training sets to cover a diverse set of learning scenarios. Our approach does not entail any significant additional computational cost compared with the loss of cross-entropy. It improves accuracy and reduces the number of coarse errors, with predicted labels that are distant from ground-truth labels in the tree.

cs.CV↗

arXiv2022

CHARA/SPICA: a 6-telescope visible instrument for the CHARA Array

With a possible angular resolution down to 0.1-0.2 millisecond of arc using the 330 m baselines and the access to the 600-900 nm spectral domain, the CHARA Array is ideally configured for focusing on precise and accurate fundamental parameters of stars. CHARA/SPICA (Stellar Parameters and Images with a Cophased Array) aims at performing a large survey of stars all over the Hertzsprung-Russell diagram. This survey will also study the effects of the different kinds of variability and surface structure on the reliability of the extracted fundamental parameters. New surface-brightness-colour relations will be extracted from this survey, for general purposes on distance determination and the characterization of faint stars. SPICA is made of a visible 6T fibered instrument and of a near-infrared fringe sensor. In this paper, we detail the science program and the main characteristics of SPICA-VIS. We present finally the initial performance obtained during the commissioning.

astro-ph.IM↗

arXiv2022

SPICA-FT: The new fringe tracker of the CHARA array

SPICA-FT is part of the CHARA/SPICA instrument which combines a visible 6T fibered instrument (SPICAVIS) with a H-band 6T fringe sensor. SPICA-FT is a pairwise ABCD integrated optics combiner. The chip is installed in the MIRC-X instrument. The MIRC-X spectrograph could be fed either by the classical 6T fibered combiner or by the SPICA-FT integrated optics combiner. SPICA-FT also integrates a dedicated fringe tracking software, called the opd-controller communicating with the main delay line through a dedicated channel. We present the design of the integrated optics chip, its implementation in MIRC-X and the software architecture of the group-delay and phase-delay control loops. The final integrated optics chip and the software have been fully characterized in the laboratory. First on-sky tests of the integrated optics combiner began in 2020. We continue the on-sky tests of the whole system (combiner + software) in Spring and Summer 2022. We present the main results, and we deduce the preliminary performance of SPICA-FT.

astro-ph.IM↗

arXiv2021

Copula-based conformal prediction for Multi-Target Regression

There are relatively few works dealing with conformal prediction for multi-task learning issues, and this is particularly true for multi-target regression. This paper focuses on the problem of providing valid (i.e., frequency calibrated) multi-variate predictions. To do so, we propose to use copula functions applied to deep neural networks for inductive conformal prediction. We show that the proposed method ensures efficiency and validity for multi-target regression problems on various data sets.

cs.LG↗

arXiv2020

HIRES, the high-resolution spectrograph for the ELT

HIRES will be the high-resolution spectrograph of the European Extremely Large Telescope at optical and near-infrared wavelengths. It consists of three fibre-fed spectrographs providing a wavelength coverage of 0.4-1.8 mic (goal 0.35-1.8 mic) at a spectral resolution of ~100,000. The fibre-feeding allows HIRES to have several, interchangeable observing modes including a SCAO module and a small diffraction-limited IFU in the NIR. Therefore, it will be able to operate both in seeing and diffraction-limited modes. ELT-HIRES has a wide range of science cases spanning nearly all areas of research in astrophysics and even fundamental physics. Some of the top science cases will be the detection of bio signatures from exoplanet atmospheres, finding the fingerprints of the first generation of stars (PopIII), tests on the stability of Nature's fundamental couplings, and the direct detection of the cosmic acceleration. The HIRES consortium is composed of more than 30 institutes from 14 countries, forming a team of more than 200 scientists and engineers.

astro-ph.IM↗