Grenoble Images Parole Signal Automatique

Assessing cocoa bean quality using spectral information offers a noninvasive and objective alternative to traditional, often subjective and destructive, methods. However, progress has been limited by the lack of comprehensive datasets across multiple spectral resolutions. This work presents a new dataset capturing the spectral properties of cocoa beans at different spatiospectral resolutions, enabling non-invasive quality assessment and scalable evaluation methodologies. It comprises 19 scenes acquired with four imaging devices under both open (invasive) and closed (non-invasive) conditions, along with corresponding physicochemical measurements. Data collection follows the Colombian standard NTC 1252:2021, which labels beans as well, partially, or poorly fermented. Global physicochemical properties-moisture, polyphenols, and cadmium-were measured using gravimetric analysis, UV-visible spectroscopy, and atomic absorption spectroscopy with microwave digestion. Hyperspectral images were obtained using four devices covering up to the 350–1000 nm spectral range. Statistical analysis shows the dataset distinguishes between cocoa quality levels under both open and closed conditions, supporting the development of automated classification methods.

The present work introduces curvature-based rejection sampling (CURS). This is a method for sampling from a general class of probability densities defined on Riemannian manifolds. It can be used to sample from any probability density which ``depends only on distance". The idea is to combine the statistical principle of rejection sampling with the geometric principle of volume comparison. CURS is an exact sampling method and (assuming the underlying Riemannian manifold satisfies certain technical conditions) it has a particularly moderate computational cost. The aim of the present work is to show that there are many applications where CURS should be the user's method of choice for dealing with relatively low-dimensional scenarios.

This paper suggests a new formulation of a Model Predictive Control (MPC) strategy, allowing to design propofol and remifentanil infusion profiles in order to control the Bispectal Index (BIS). This new formulation, based on a range cost, allows to reduce the sensitivity of the control profiles with respect to the BIS measurement noise. Furthermore, it allows to better represent the anesthesiologists behavior in practice, who usually do not over-react to small changes in the measured health indicators. The paper assesses numerically this formulation, by comparing its performance, in an uncertain setting, to standard set-point tracking based MPC strategies.

Standard single-turn, static benchmarks fall short in evaluating the nuanced capabilities of Large Language Models (LLMs) on complex tasks such as software engineering. In this work, we propose a novel interactive evaluation framework that assesses LLMs on multi-requirement programming tasks through structured, feedback-driven dialogue. Each task is modeled as a requirement dependency graph, and an ``interviewer'' LLM, aware of the ground-truth solution, provides minimal, targeted hints to an ``interviewee'' model to help correct errors and fulfill target constraints. This dynamic protocol enables fine-grained diagnostic insights into model behavior, uncovering strengths and systematic weaknesses that static benchmarks fail to measure. We build on DevAI, a benchmark of 55 curated programming tasks, by adding ground-truth solutions and evaluating the relevance and utility of interviewer hints through expert annotation. Our results highlight the importance of dynamic evaluation in advancing the development of collaborative code-generating agents.

For quasilinear parabolic partial differential equations (PDEs) that exhibit finite-time blow up in open loop, i.e., under null boundary conditions, we provide an estimate of the region of attraction under cubic feedback laws applied at the boundary, using boundary measurements. We guarantee: 1-L 2 and H 1 exponential stability of the origin with an estimate of the region of attraction. 2-Convergence of the H 2 and the C 1 norms of the solutions to zero. 3-Existence and uniqueness of complete classical solutions. 4-Positivity of the solutions starting from positive initial conditions. Unlike existing approaches, our framework handles nonlinear state-dependent diffusion, convection, and (destabilizing) reaction. The cubic terms are used to enlarge our estimate of the region of attraction. The size of the region of attraction is shown, in many cases, to grow unboundedly as diffusion increases. Finally, our controllers can be implemented as Neumann, Dirichlet, or mixed-type boundary conditions.

ABSTRACT Recent neurocomputational speech perception models include a segmentation process delimiting speech segments, typically syllable-long, before decoding. This process is conceived as purely bottom-up, detecting temporally salient events in the speech signal. Beyond the scope of current models, the COSMO-Onset model incorporates in the segmentation process top-down predictions based on linguistic knowledge of speech rhythms. We present an adaptation of the model, previously studied on artificial stimuli, to process real speech signals. The model is used to simulate syllable-type recognition in noise, with two main results. Firstly, bottom-up segmentation based on resonant processes favours isochrony, since sentences with a more regular structure of syllabic events provide better segmentation and recognition scores. Secondly, top-down lexical predictions of syllable duration make detection and syllable recognition more robust in noise. The combination of bottom-up resonant and top-down predictive processes yields simulations in line with a recent behavioural experiment on speech comprehension in noise.