INRIA a CCSD electronic archive server
Not a member yet
122212 research outputs found
Sort by
Virtual reality-augmented differentiable simulations for digital twin applications in surgical planning
No full textInternational audienceThis paper presents a novel approach for digital twin applications in surgical planning, integrating a differentiable simulator for trajectory generation within segmented medical images and a virtual reality (VR) platform for navigating an overlay of medical images and generated trajectories. The first section of this study delineates a path-planning method utilizing Langevin random walkers subjected to both local and non-local trajectory constraints. This stochastic process effectively navigates the complex topography of medical imaging data without necessitating comprehensive surgical instrument modeling or replicating the environment's physical properties. The second section introduces a custom-developed shader that seamlessly integrates raw and segmented medical images with potential surgical trajectories within a VR environment. This VR integration provides surgeons an immersive and intuitive platform, facilitating interactive exploration and selecting optimal surgical paths. The system extends beyond predefined trajectory generation criteria, allowing real-time adjustments based on the surgeon's expertise and situational assessment. The approach combines these advanced computational techniques with state-of-the-art visualization methods to enhance surgical planning precision and efficiency. This integrated digital twin strategy can potentially improve preoperative decision-making significantly and, consequently, surgical outcomes
Assessing simulation-based supervised machine learning for demographic parameter inference from genomic data
No full textInternational audienceThe ever-increasing availability of high-throughput DNA sequences and the development of numerous computational methods have led to considerable advances in our understanding of the evolutionary and demographic history of populations. Several demographic inference methods have been developed to take advantage of these massive genomic data. Simulation-based approaches, such as approximate Bayesian computation (ABC), have proved particularly efficient for complex demographic models. However, taking full advantage of the comprehensive information contained in massive genomic data remains a challenge for demographic inference methods, which generally rely on partial information from these data. Using advanced computational methods, such as machine learning, is valuable for efficiently integrating more comprehensive information. Here, we showed how simulation-based supervised machine learning methods applied to an extensive range of summary statistics are effective in inferring demographic parameters for connected populations. We compared three machine learning (ML) methods: a neural network, the multilayer perceptron (MLP), and two ensemble methods, random forest (RF) and the gradient boosting system XGBoost (XGB), to infer demographic parameters from genomic data under a standard isolation with migration model and a secondary contact model with varying population sizes. We showed that MLP outperformed the other two methods and that, on the basis of permutation feature importance, its predictions involved a larger combination of summary statistics. Moreover, they outperformed all three tested ABC algorithms. Finally, we demonstrated how a method called SHAP, from the field of explainable artificial intelligence, can be used to shed light on the contribution of summary statistics within the ML models
A fast implementation of 3D wavelet compression on GPU
No full textInternational audienceGPUs are widely used in high-performance computing for performing computationally intensive tasks. However, GPUs have fixed memory capacities that are much smaller than those of CPUs, often limiting the full exploitation of their computational power. Moreover, the memory bus between CPUs and GPUs is an order of magnitude slower, hindering the efficient feeding of data to GPUs during computation. To overcome these limitations, we developed an efficient compression kernel for GPUs. Our implementation relies on the wavelet method used in the JPEG2000 format and is highly optimized in CUDA using the GPU's shared memory. We demonstrate that our kernel can compress and decompress data faster than transferring data over the PCIe bus, allowing to store more data and transferring them faster.</div
Automatic generation of input-aware approximate arithmetic circuits
No full textInternational audienceApproximate Computing (AxC) is systematically applied across various abstraction levels to reduce overheads and enhance the performance of applications such as image processing and machine learning. However, AxC does not typically consider the specific workload (i.e., data input) of a given application. For instance, in signal processing applications like filters, some inputs are constants (filter coefficients), which allows for an additional level of approximation by considering the specific input distribution. This method is known as "Input-Aware Approximation" (IAA) and has shown potential advantages in previous studies. Unfortunately, existing input-aware design methodologies lack scalability as they mostly depend on ad-hoc, non-automatic design approaches, limiting their applicability. In this paper, we investigate how the input-aware approximate design approach can be integrated into a systematic, generic, and automatic design flow. We employ state-of-the-art approximation and multi-objective optimization techniques to achieve input awareness. Our experimental results, focusing on classical signal processing applications like FIR filters, demonstrate that the input-aware approach can provide significant savings in both area and power consumption
A genetic approach for automatic AxC design exploration at RTL based on assertion mining and fault analysis
No full textInternational audienceIn Approximate Computing (AxC), design exploration methods have been introduced to automatically identify approximation targets at the gate level. However, only some of them are applicable at Register Transfer Level (RTL); furthermore, the benefits of combining information from assertions and fault analysis have not been fully explored. This paper proposes a novel methodology for guiding AxC design exploration at RTL considering two approximation techniques: bit-width reduction and statement reduction. Then, it employs fault injection to mimic the approximation effect on the design under approximation. To guide the designer while assessing the approximation choices, assertions, which formally capture the behaviors implemented in the design, are dynamically generated from the RTL simulation traces. Then, the impact of fault injections on the truth values of the assertions is employed as a proxy for measuring the functional accuracy of the corresponding approximations. Based on this evaluation, a genetic algorithm is finally used to rank and cluster the approximation targets, thus providing the designer with an efficient and effective way to automatically analyze AxC variants in terms of the trade-off between accuracy and performance. The experiments carried out on state-of-the-art benchmarks show that the proposed approach represents a promising solution for the automation of AxC design exploration at RTL
Visualizing information on smartwatch faces: A review and design space
No full textInternational audienceWe present a systematic review and design space for visualizations on smartwatches and the context in which these visualizations are displayed—smartwatch faces. A smartwatch face is the primary smartwatch screen wearers see when checking the time. Smartwatch faces are small data dashboards that show a variety of data to wearers in a compact form. Yet, smartwatch faces’ usage context and form factor pose unique design challenges for visualizations. In this paper, we conducted an in-depth review and analysis of visualization designs for popular premium smartwatch faces based on their design styles, amount and types of data, as well as visualization styles and encodings they included. From our analysis, we derive a design space to provide an overview of the important considerations for new data displays for smartwatch faces and other small displays. Our design space can also serve as inspiration for design choices and grounding of empirical work on smartwatch visualization design. We end with a research agenda pointing to opportunities in this nascent research direction. Supplemental material from the study is available here: https://osf.io/p3tbj/
Repeatability of kinematics and kinetics data in cycling using wearable sensors
No full textInternational audienc
Normative Alignment of Recommender Systems via Internal Label Shift
No full textInternational audienceRecommender systems optimized solely for user engagement often fail to meet broader normative objectives such as fairness, diversity, or editorial values. We introduce NAILS (Normative Alignment of recommender systems via Internal Label Shift), a simple and scalable method for aligning recommendation outputs with target distributions over item-level attributes (e.g., categories). NAILS modifies the user-conditional item distribution to induce a specified marginal distribution over attributes, leveraging existing user–item preferences without retraining the model. To achieve this, we recast the problem as a form of label shift applied internally within a hierarchical classification framework. Adopting a stakeholder-centric perspective, NAILS enables alignment with global normative goals. Empirically, we show that NAILS consistently improves attribute-level alignment with minimal impact on user engagement, providing a practical mechanism for value-driven recommendation. Our code is available at https://github.com/johanneskruse/nails
Expédition et expansion : Exploiter les représentations sémantiques pour une exploration ciblée dans les automates cellulaires continus
No full textInternational audienceDiscovering diverse visual patterns in continuous cellular automata (CA) is challenging due to the vastness and redundancy of high-dimensional behavioral spaces. Traditional exploration methods like Novelty Search (NS) expand locally by mutating known novel solutions but often plateau when local novelty is exhausted, failing to reach distant, unexplored regions. We introduce Expedition & Expansion (E&E), a hybrid strategy where exploration alternates between local novelty-driven expansions and goal-directed expeditions. During expeditions, E&E leverages a Vision-Language Model (VLM) to generate linguistic goals-descriptions of interesting but hypothetical patterns that drive exploration toward uncharted regions. By operating in semantic spaces that align with human perception, E&E both evaluates novelty and generates goals in conceptually meaningful ways, enhancing the interpretability and relevance of discovered behaviors. Tested on Flow Lenia, a continuous CA known for its rich, emergent behaviors, E&E consistently uncovers more diverse solutions than existing exploration methods. A genealogical analysis further reveals that solutions originating from expeditions disproportionately influence long-term exploration, unlocking new behavioral niches that serve as stepping stones for subsequent search. These findings highlight E&E's capacity to break through local novelty boundaries and explore behavioral landscapes in human-aligned, interpretable ways, offering a promising template for open-ended exploration in artificial life and beyond.La découverte de motifs visuels variés dans les automates cellulaires continus (AC) est complexe en raison de l'immensité et de la redondance des espaces comportementaux à haute dimension. Les méthodes d'exploration traditionnelles, comme la recherche de nouveauté (NS), s'étendent localement en mutant des solutions novatrices connues, mais stagnent souvent lorsque la nouveauté locale est épuisée, ne parvenant pas à atteindre des régions lointaines et inexplorées. Nous présentons l'Expédition et Expansion (E&E), une stratégie hybride où l'exploration alterne entre des expansions locales axées sur la nouveauté et des expéditions orientées vers des objectifs. Lors des expéditions, l'E&E s'appuie sur un modèle vision-langage (VLM) pour générer des descriptions d'objectifs linguistiques de motifs intéressants mais hypothétiques qui orientent l'exploration vers des régions inexplorées. En opérant dans des espaces sémantiques alignés sur la perception humaine, l'E&E évalue la nouveauté et génère des objectifs de manière conceptuellement significative, améliorant ainsi l'interprétabilité et la pertinence des comportements découverts. Testé sur Flow Lenia, une CA continue connue pour ses comportements riches et émergents, l'E&E révèle systématiquement des solutions plus diversifiées que les méthodes d'exploration existantes. Une analyse généalogique révèle également que les solutions issues des expéditions influencent de manière disproportionnée l'exploration à long terme, ouvrant de nouvelles niches comportementales qui servent de tremplins pour les recherches ultérieures. Ces résultats soulignent la capacité de l'E&E à dépasser les limites de la nouveauté locale et à explorer des paysages comportementaux de manières interprétables et adaptées aux humains, offrant un modèle prometteur pour une exploration ouverte dans la vie artificielle et au-delà.</div
RibbonSculpt: Voronoi Ball based 3D Sculpting from Sparse VR Ribbons
No full textInternational audienceWe introduce RibbonSculpt, the first method for interactive freeform shape design in VR through progressive sketching of sparse, oriented ribbons. Instead of reconstructing a surface from a fully drawn VR sketch, our method allows the real-time creation and progressive refinement of a closed surface of any topological genus, thanks to the continuous update of a volumetricproxy. The latter corresponds to a filtered subset of the Voronoi balls defined by the user-sketched ribbons. At each visualization step, a mesh extracted from the proxy is beautified through Laplacian-based energy minimization, yielding a smooth surface that interpolates the ribbons. Guided by this surface, users can easily refine their design by adding or removing ribbons, which sculpts, in return, the set of Voronoi balls forming the proxy. Our results, supported by user studies, show that RibbonSculpt allows VR users to easily and quickly draft the 3D shapes they have in mind