Portail HAL des publications du LIRMM
Not a member yet
13279 research outputs found
Sort by
Energy Efficient Time Series Anomaly Detection
No full textInternational audienceAnomalous events are commonly observed in realworld temporal data, known as time series. Time series anomaly detection is pervasive for process monitoring in almost every scientific application. The area presents extensive literature and several state-of-the-art methods. Traditionally, choosing a method for a given application is mainly driven by detection accuracy and runtime. However, with the rapid evolution of hardware and connected devices, massive amounts of time series data are produced, and the real-time analysis of such time series brings new demands not only for accurate and scalable solutions, but also for energy consumption management. In this scenario, any improvement in energy efficiency can have a considerable impact on both the environmental footprint and the monetary expenses. However, to the best of our knowledge, there is no existing work on energy efficient time series anomaly detection. This paper fills this gap by addressing for the first time the problem of benchmarking time series anomaly detection methods based on the trade-off between accuracy, runtime, and energy consumption. We introduce a new metric for evaluating relative energy efficiency performance, called saveUp, and provide a novel methodology, inspired by skyline queries, for benchmarking methods based on a more comprehensive set of metrics, including peak power usage and total energy consumption. Experimental results based on large datasets show that our methodology is useful for selecting the methods that provide the best performance with the lowest energy impacts. Moreover, results indicate that speedup and saveUp are not always directly correlated as believed a priori, and sometimes it is best to "take it slow" in favor of green applications
A Digital SRAM Modeling for Cell-Aware Testing and Test Algorithms Evaluation
No full textInternational audienceModern integrated circuits such as System-On-Chip (SOC) require a large amount of memory. The integration density of these memories is based on the extreme miniaturization of the technological nodes. However, the miniaturization process contributes to an increase in the occurrence of physical defects, and in the complexity of the memory faults. In order to verify the quality of memories, new test methods, based on structural testing, have been proposed recently. However, these methods rely on a digital test environment and require an accurate modeling of the memory. This work proposes a digital SRAM modeling compatible with digital simulation and test environments. The digital SRAM modeling presented in this paper is functionally equivalent to an analog SRAM model (i.e., memorization, Write and Read operations). The memorization capability of the model enables it to consider the data-background of the memory array during testing, facilitating the evaluation of complex test sequences, such as March algorithms, in covering structural fault models (i.e., Cell-Aware fault models)
Explainable Evidential Clustering Generating decision-maker–aligned explanations : Presenting the Iterative Evidential Mistakeness Minimization algorithm
No full textInternational audienc
Module Learning with Errors with Truncated Matrices
No full textInternational audienceThe Module Learning with Errors (MLWE) problem is one of the most commonly used hardness assumption in lattice-based cryptography. In its standard version, a matrix A is sampled uniformly at random over a quotient ring R q , as well as noisy linear equations in the form of As + e mod q, where s is the secret, sampled uniformly at random over R q , and e is the error, coming from a Gaussian distribution. Many previous works have focused on variants of MLWE, where the secret and/or the error are sampled from different distributions. Only few works have focused on different distributions for the matrix A. One variant proposed in the literature is to consider matrix distributions, where the low-order bits of a uniform A are deleted. This seems a natural approach in order to save in bandwidth. We call it truncated MLWE. In this work, we show that the hardness of standard MLWE implies the hardness of truncated MLWE, both for search and decision versions. Prior works only covered the search variant and relied on the (module) NTRU assumption, limitations which we are able to overcome. Overall, we provide two approaches, offering different advantages. The first uses a general R´enyi divergence argument, applicable to a wide range of secret/error distributions, but which only works for the search variants of (truncated) MLWE. The second applies to the decision versions, by going through an intermediate variant of MLWE, where additional hints on the secret are given to the adversary. However, the reduction makes use of discrete Gaussian distributions
Audiocarnet - Deep Representation Learning from Unlabeled Bioacoustic Data
No full textContrastive learning requires creating distinct views of the same input while preserving important details. Standard audio augmentations, such as random resized crops, can remove species-specific characteristics. We propose mixing vocalizations as a domain-agnostic data augmentation, which preserves the unique features of the species of interest while forming a distinct view. This simple strategy allows contrastive learning to capture species-specific features in bird vocalizations from unlabeled data
Natural efficient gaits from Nonholonomic Locomotion Nonlinear Normal Mode (NL-NNM): The Pendrivencar case
No full textInternational audienceBio-inspired robots remain far less energy-efficient than animals because conventional controllers impose trajectories that fight passive dynamics, whereas animals exploit resonance through natural nonlinear normal modes (NNM), whose periodic internal motions form a smooth 2D invariant surface; We ask how to define and compute the natural motions of a conservative locomotion system: propulsion arises only from no-slip constraints, and once initiated, a gait persists without actuation-like a frictionless pendulum. We tackle non-holonomic constraints on the Pendrivencar, a vehicle driven by a motorised pendulum with a cubic torsional spring ; We introduce the Nonholonomic Locomotion -NNM (NL-NNM): extract a high-speed spectral seed -where chassis oscillations vanish and the pendulum is neutrally stable -refine the periodic orbit, and continue the resulting 2D invariant manifold via pseudo-arclength across three slow centre manifolds (stable for positive speed, neutral at zero, unstable for negative) from non-isolated rectilinear equilibria; We demonstrate the first NL-NNM for a moving non-holonomic robot: internal orbits produce a pendulum-chassis choreography whose energy-dependent frequency shifts and harmonic richness exceed linear predictions. Via geometric phase, each orbit yields undulatory straight-line motion. A dual-loop control simulation confirms autonomous path tracking with only the pendulum; Extending to dissipative regimes via non-linear resonant modes offers a path to high-efficiency locomotion in aquatic, aerial, legged, soft-bodied, and other robots
Interpretable DNFs
No full textInternational audienceA classifier is considered interpretable if each of its decisions has an explanation which is small enough to be easily understood by a human user. A DNF can be seen as a binary classifier kappa over boolean domains. The size of an explanation of a positive decision taken by a DNF kappa is bounded by the size of the terms in kappa, since we can explain a positive decision by giving a term of kappa that evaluates to true. Since both positive and negative decisions must be explained, we consider that interpretable DNFs are those kappa for which both kappa and its complement can be expressed as DNFs composed of terms of bounded size. In this paper, we investigate the family of k-DNFs whose complements can also be expressed as k-DNFs. We compare two such families, namely depth-k decision trees and nested k-DNFs, a novel family of models. Experimental evidence indicates that nested k-DNFs are an interesting alternative to decision trees in terms of interpretability and accuracy
IP Masking with Generic Security Guarantees under Minimum Assumptions, and Applications
No full textInternational audienceLeakage-resilient secret sharing is a fundamental building block for securing implementations against side-channel attacks. In general, such schemes correspond to a tradeoff between the complexity of the resulting masked implementations, their security guarantees and the physical assumptions they require to be effective. In this work, we revisit the Inner-Product (IP) framework, where a secret y is encoded by two vectors (), such that their inner product is equal to y. So far, the state of the art is split in two. On the one hand, the most efficient IP masking schemes (in which is public but random) are provably secure with the same security notions (i.e., in the abstract probing model) as Boolean masking, yet at the cost of a slightly more expensive implementation. Hence, their theoretical interest and practical relevance remain unclear. On the other hand, the most secure IP masking schemes (in which is secret) lead to expensive implementations. We improve this state of the art by investigating the leakage resilience of IP masking with public coefficients in the bounded leakage model, which depicts well implementation contexts where the physical noise is negligible. Furthermore, we do that without assuming independent leakage from the shares, which may be challenging to enforce in practice. In this model, we show that if m bits are leaked from the d shares of the encoding over an n-bit field, then, with probability at least over the choice of , the scheme is leakage resilient. We additionally show that in large Mersenne-prime fields, a wise choice of the public coefficients can yield leakage resilience up to in the case where one physical bit from each share is revealed to the adversary. The exponential rate of the leakage resilience we put forward significantly improves upon previous bounds in additive masking, where the past literature exhibited a constant exponential rate only. We additionally discuss the applications of our results, and the new research challenges they raise
OntoPortal-Astro, a semantic artefact catalogue for astronomy
No full textInternational audienceThe astronomy communities are widely recognised as mature communities for their open science practices. However, while their data ecosystems are rather advanced and permit efficient data interoperability, there are still gaps between these ecosystems. Semantic artefacts (SAs) – e.g., ontologies, thesauri, vocabularies or metadata schemas – are a means to bridge that gap as they allow to semantically described the data and map the underlying concepts. The increasing use of SAs in astronomy presents challenges in description, selection, evaluation, trust, and mappings. The landscape remains fragmented, with SAs scattered across various registries in diverse formats and structures – not yet fully developed or encoded with rich semantic web standards like OWL or SKOS – and often with overlapping scopes. Enhancing data semantic interoperability requires common platforms to catalogue, align, and facilitate the sharing of FAIR (Findable, Accessible, Interoperable and Reusable) SAs. In the frame of the FAIR-IMPACT project, we prototyped a SA catalogue for astronomy, heliophysics and planetary sciences. This exercise resulted in improved vocabulary and ontology management in the communities, and is now paving the way for better interdisciplinary data discovery and reuse. This article presents current practices in our discipline, reviews candidate SAs for such a catalogue, presents driving use cases and the perspective of a real production service for the astronomy community based on the OntoPortal technology, that will be called OntoPortal-Astro
Integrating Environmental Regulations Into Autonomous Agricultural Robotics: A Case for Waterbody-Aware Fertilization
No full textInternational audienceThe operation of autonomous robots in the agricultural domain requires compliance with the regulatory aspects of the process. For instance, the improper spraying of chemicals near water bodies (e.g., pesticides or fertilizers) may cause significant environmental damage and, therefore, is strictly regulated at the legislation level. In this paper, we introduce a reasoning-enhanced framework to operate autonomous robots spreading chemicals near water bodies. Our framework leverages and extends semantic web vocabularies to integrate regulatory constraints and environmental conditions where the robot is operating. Then, it uses rules and reasoning to detect violations on real-time data generated by the autonomous robot. The inference of violations subsequently triggers actions controlling the robot behavior, but can also be transparently explained in our framework, thereby avoiding the robot to behave as a black-box to the supervising technician. Our approach has been implemented, and its feasibility showcased in a simulation environment