Organization contact [name= K. Desnos (kdesnos at insa-rennes.fr)] :email: Mailing list :mailbox: :calendar: Shared Calendar :calendar: You can subscribe to a dedicated mailing-list, that will exclusively be used to send seminar and Ph.D. defense announcements. To subscribe, send an e-mail to sympa@insa-rennes.fr with the following subject line (with YOUR first and last names):subscribe vaader-seminars FirstName LastName Add all seminars automatically to your agenda with the following: link. If you do so, we advise you to also subscribe to the mailing list Next Seminars

by Antoine Lagrange (B<>com) - 2025.02.27 seminaire Abstract Holography is often considered as the most promising immersive technology because it provides all the depth cues of the human visual system. Still, it relies on diffraction, therefore, to reach a wide field of view, the pixels of computer-generated holograms must be about a micrometer. Holograms of few centimeters contain billion of pixels and this huge amount of data is a major issue. To solve this limitation, we proposed a view-specific layered holographic stereogram approach combined with a view-dependent error diffusion algorithm. This method selects the light waves of the 3D scene that reach a specific viewing area and leverages this particular configuration to apply an error diffusion algorithm. Two additional quality enhancement features are observed: the reduction of the conjugate order perceptibility and the increased brightness of the reconstructions. Digital and optical experiments demonstrate the time savings and quality enhancements of our approach.

by Last First (Affiliation) - 20YY.MM.DD Abstract

by Quentin Vacher (IETR VAADER) - 20YY.MM.DD MemeSiminaireHybrid Abstract This talk presents a hybrid approach to robot arm trajectory planning that combines Tangled Program Graphs (TPGs) with the Soft Actor-Critic (SAC) algorithm. While TPGs alone reduce computational complexity and model size, they struggle with more complex tasks. The hybrid solution leverages the strengths of both methods and achieves superior performance compared to state-of-the-art deep reinforcement learning algorithms, with execution times 6 to 20 times faster.

by Tanya Djavaherpour (McMaster University) - 2025.01.07 TPGbob Abstract This study addresses the challenges of shared temporal memory for evolutionary reinforcement learning agents in partially observable control tasks with short time dependencies. Tangled Program Graphs (TPG) is a genetic programming framework which has been widely studied in memory intensive tasks from video games, time series forecasting, and predictive control domains. In this study, we aim to improve external indexed memory usage in TPG by minimizing the impact of destructive agents during cultural transmission. We test various memory resetting strategies—per agent, per episode, and a no-memory control group—and evaluate their effectiveness in mitigating destructive effects while maintaining performance. Results from Acrobot, Pendulum, and CartPole tasks show that resetting memory more often can significantly boost TPG performance while preserving computational efficiency. These findings highlight the importance of memory management in Reinforcement Learning (RL) and suggest opportunities for further optimization for more complex visual RL environments, including adaptive memory resetting and evolved probabilistic memory operations.

by Last First (Affiliation) - 20YY.MM.DD COCALITE Abstract Time series classification has achieved significant advancements through deep learning models; however, these models often suffer from high complexity and computational costs. To address these challenges while maintaining effectiveness, we introduce COCALITE, an innovative hybrid model that combines the efficient LITE model with an augmented version incorporating Catch22 features during training. COCALITE operates with only 4.7% of the parameters of the state-of-the-art Inception model, significantly reducing computational overhead. By integrating these complementary approaches, COCALITE leverages both effective feature engineering and deep learning techniques to enhance classification accuracy. Our extensive evaluation across 128 datasets from the UCR archive demonstrates that COCALITE achieves competitive performance, offering a compelling solution for resource-constrained environments.

by Ophélie Renaud (Univ. Paris Saclay, IETR VAADER) - 2024.11.15 meme_gpu_720 Abstract The next generation of radio telescopes, such as the Square Kilometre Array (SKA), will require peta-FLOPS processing power to handle the massive amount of data acquired. A new generation of computing pipelines are required to address the SKA challenges leading to the integration of the pipelines on a dedicated heterogeneous High-Performance Computing (HPC) system. The tight real-time and energy constraints are driving the community to study the use of hardware accelerators like GPUs in the computing system. Allocating resources, such as processor times, memory, or communication bandwidth, to support complex algorithms in such systems is known as an NP-complete problem. Existing tools such as Dask and Data Activated 流 Graph Engine (DALiuGE) rely on dataflow Model of Computation (MoC) and have proven to be an efficient solution to specify parallel algorithms and automate their deployment. These models are efficient programming paradigms for expressing the parallelism of an application. However, state-of-the-art dataflow resource allocation only targets CPUs and usually relies on complex graph transformations resulting in a time-consuming process. This paper introduces an automated dataflow resource allocation method and code generation for heterogeneous CPU-GPU systems. Our method efficiently and quickly manages pre-scheduling graph complexity, and optimizes the dataflow model to the target architecture. Experimental results show that the proposed method improves resource allocation and speeds up the process by a factor of 13 compared to the best existing method on a basic architecture. Moreover, the execution times of the obtained implementations are comparable to those of manual implementations.

by Peter SCHELKENS (Vrije Universiteit Brussels) - 2024.09.26 image001 Abstract

by Nicolas Verrier (IRIMAS, Mulhouse)- 2024.09.26 image-seminaire-nicolas-verrier_480 Abstract Les systèmes optiques classiques reposant sur l'acquisition de l'intensité lumineuse transmise, réfléchie, diffusée ou diffractée montrent rapidement leurs limites, notamment dans le contexte de la microscopie. Des alternatives reposants sur l'utilisation de marqueurs fluorescents existent, mais nécessite une préparation parfois lourde et intrusive des échantillons. D'un autre côté, des méthodes d'imagerie non-conventionnelles ont été proposées. Ces dernières reposent sur la connaissance du modèle de formation des images. Ce modèle est alors inversé pour revenir à l'objet à l'origine du signal acquis. Je détaillerai ici le concept d'imagerie non-conventionnelle et discuterai des applications à l'imagerie biologique 3D multifonctionnelle.

by Richard Green (Univ. of Canterbury, NZ) - 2024.09.26 d3rb0ar5 Abstract We cannot automate what we cannot see – so our recent breakthroughs with NeRF and Gaussian splatting is helping to solve leaf occlusion to enable a rapid uptake of agricultural automation. I will describe our contributions across these research areas, including recent autonomous systems research into drones pruning forests, robots pruning vineyards, autonomous underwater vehicles (AUVs) inspecting mussel lines to detect invasive biofouling species and AUVs mapping the seabed to locate scallops. I will also discuss the challenges that remain and so propose potential directions for future work.

by Pascal Picart (Univ Le Mans) - 2024.09.26 IllutrationSeminairePPicart Abstract Anurans are considered as the most vocal of vertebrates. Indeed, they use acoustic communication in many behavioral contexts, even in their reproduction process [1]. In the process of acoustic communication, sound localization is an important part and was studied for mammal species. In the duplex theory of Lord Rayleigh, in the horizontal plane sound localization relies on binaural cues that are, the inter-aural time difference (ITD), and the interaural level difference (ILD) [2]. In the vertical plane, sound localization is made possible by monaural cues derived from the diffraction of sound by the individual's external anatomy. From the point of view of frequency and size, most species have dominant frequencies of a few thousand hertz [3] and small size from around 10 mm to 320 mm. So, it follows that sound localization should be impossible for animals with small heads and hearing frequency range below 10 kHz. However, anurans break this physical prerequisite because they localize sound with great accuracy in the audible human range. It turns out that understanding the directionality of anuran amphibian hearing is a formidable scientific challenge. Some fundamental mechanisms of hearing remain partially or totally uncertain. This may be due to the fact that our understanding of vertebrate sound localization is based largely, 1) on studies of mammals (mainly humans) and archosaurs, and 2) on the multiple ways in which sound can reach the inner ear, including tympanic and extra-tympanic pathways (e.g. lungs, oral cavity, nostrils). The duplex theory of Rayleigh reaches its limits in frogs (broadly defined), due to the small size of these animals in relation to the wavelengths of the sounds produced for communication. Unfortunately, the ear structures involved are difficult to access, and direct observation of the ear from the outside or inside is impossible. In this paper, we aim at reporting our contribution to this topic. In general, most of the methods currently used to observe the ear are invasive and destructive, and in some cases may even alter the mechano-acoustic response of these structures. The authors decided to follow a methodological breakthrough for the study of anuran hearing by considering, on one hand, innovative imaging methods combined with numerical modeling to provide a non-invasive understanding of these hearing structures. In this paper, we focus on holographic imaging and we present an achiral digital Fresnel holography set-up able to measure the tympanic membrane vibration on both sides of the head of in-vivo anurans submitted to acoustic excitation in the range of 1000Hz-5000Hz. A dedicated acoustic antenna was built with 24 loudspeakers organized around 360° and three angles of elevation, 0°, 30° and 45°. The in vivo specimen is located in the central part of an achiral holographic set-up with two high-speed cameras and a microphone record the excitation emission of the animal. Electronic synchronization is provided to launch the acoustic emission sequence whereas recording on both cameras the sequences of digital holograms. Once processed the acoustic emission yields the exact excitation frequency and amplitude The time-sequences of digital holograms are processed to yield the optical phase variations [4]. The frequencies are considered as input of an inverse problem leading to the accurate estimation of the amplitude and phase of the tympanic vibration of the in-vivo anuran. We present the first experimental results with specimen Pelphylax lessonae and Xenophrys aceras and related interpretations of observations ved.

by Hasna Essakali (IETR VAADER) - 2024.09.03 PPG Abstract

by Jiayue Yin (IETR VAADER) - 2024.09.03 YJY_image Abstract

by Gwendal Bourdet (IETR VAADER) - 2024.09.03 deeptec_img Abstract

by Anne-Lise Eberlin - 2024.08.29 Capture d’écran 2024-08-28 142149 Abstract

by Marcus Pinare - 2024.08.29 image-fun Abstract

by Kuilong Li - 2024.08.29 ACV_memes Abstract

by Mathieu Lefort - 2024.06.20 Abstract Mes travaux visent à obtenir des systèmes apprenants plus autonomes et robustes s'appuyant principalement sur des principes bio inspirés. Ils s'articulent autour de 3 axes de recherche principaux: la fusion de données multi-sensorielles (audio-visuelles) topologiquement organisées avec des modèles neuro-inspirés permettant la modélisation de données comportementales chez l'humain, avec des transferts éventuels en robotique l'apprentissage de représentations (visuelles) structurées, entre autres par l'action, en utilisant des approches auto-supervisées appliquées à des réseaux de neurones profonds l'apprentissage actif pour l'amélioration des performances (ou de la vitesse) d'apprentissage qui pose la question de l'apprentissage incrémental

by Loïc Sylvestre (Lip6) - 2024.05.24 SmileDimensions Abstract Eclat est un langage de programmation généraliste, compilé au niveau transfert de registres, pour concevoir et implanter des applications matérielles réactives sur des circuits FPGA. Le langage unifie plusieurs styles de programmation (synchrone, parallèle, fonctionnel-impératif), tout en offrant un contrôle

by Manuel Clergue (ESIEA) - 2024.04.18 Sans titre Abstract