Portail HAL Ensta
Not a member yet
11080 research outputs found
Sort by
Codeplay: Autotelic Learning through Collaborative Self-Play in Programming Environments
No full textInternational audienceAutotelic learning is the training setup where agents learn by setting their own goals and trying to achieve them. However, creatively generating freeform goals is challenging for autotelic agents. We present Codeplay, an algorithm casting autotelic learning as a game between a Setter agent and a Solver agent, where the Setter generates programming puzzles of appropriate difficulty and novelty for the solver and the Solver learns to achieve them. Early experiments with the Setter demonstrates one can effectively control the tradeoff between difficulty of a puzzle and its novelty by tuning the reward of the Setter, a code language model finetuned with deep reinforcement learning
On evolution of irreversible systems: computing analysis results
No full textInternational audienceWhat happens to the stability of critical equilibrium states, as time goes by? In evolutionaryproblems, structural effects play a key role. We study evolving systems, singularly perturbed andunder constraints. Picture fracture as a process: as a phenomenon of natural evolution, it offers aninstance of a particularly complex natural system showing pattern formation and jumps, unknownboth in time and space
In vitro construction and long read sequencing analysis of a 24 kb long artificial DNA sequence encoding the Universal Declaration of the Rights of Man and of the Citizen
No full textAbstract In absence of DNA template, the ab initio production of long double-stranded DNA molecules of predefined sequences is particularly challenging. The DNA synthesis step remains a bottleneck for many applications such as functional assessment of ancestral genes, analysis of alternative splicing or DNA-based data storage. We propose in this report a fully in vitro protocol to generate very long double-stranded DNA molecule starting from commercially available short DNA blocks in less than 3 days. This innovative application of Golden Gate assembly allowed us to streamline the assembly process to produce a 24 kb long DNA molecule storing part of the Universal Declaration of Human rights and citizens. The DNA molecule produced can be readily cloned into suitable host/vector system for amplification and selection
Riemannian SPD learning to represent and characterize fixational oculomotor Parkinsonian abnormalities
No full textInternational audienceParkinson's disease (PD) is the second most common neurodegenerative disorder, mainly characterized by motor alterations. Despite multiple efforts, there is no definitive biomarker to diagnose, quantify, and characterize the disease early. Recently, abnormal fixational oculomotor patterns have emerged as a promising disease biomarker with high sensitivity, even at early stages. Nonetheless, the complex patterns and potential correlations with the disease remain largely unexplored, among others, because of the limitations of standard setups that only analyze coarse measures and poorly exploit the associated PD alterations. This work introduces a new strategy to represent, analyze and characterize fixational patterns from non-invasive video analysis, adjusting a geometric learning strategy. A deep Riemannian framework is proposed to discover potential oculomotor patterns aimed at withstanding data scarcity and geometrically interpreting the latent space. A convolutional representation is first built, then aggregated onto a symmetric positive definite matrix (SPD). The latter encodes second-order statistics of deep convolutional features and feeds a non-linear hierarchical architecture that processes SPD data by maintaining them into their Riemannian manifold. The complete representation discriminates between Parkinson and Healthy (Control) fixational observations, even at PD stages 2.5 and 3. Besides, the proposed geometrical representation exhibit capabilities to statistically differentiate observations among Parkinson's stages. The developed tool demonstrates coherent results from explainability maps back-propagated from output probabilities
Accelerating Random Forest on Memory-Constrained Devices through Data Storage Optimization
No full textInternational audienc
Personalized incentives as feedback design in generalized Nash equilibrium problems
No full textInternational audienceWe investigate both stationary and time-varying, nonmonotone generalized Nash equilibrium problems that exhibit symmetric interactions among the agents, which are known to be potential. As may happen in practical cases, however, we envision a scenario in which the formal expression of the underlying potential function is not available, and we design a semi-decentralized Nash equilibrium seeking algorithm. In the proposed two-layer scheme, a coordinator iteratively integrates the (possibly noisy and sporadic) agents' feedback to learn the pseudo-gradients of the agents, and then design personalized incentives for them. On their side, the agents receive those personalized incentives, compute a solution to an extended game, and then return feedback measurements to the coordinator. In the stationary setting, our algorithm returns a Nash equilibrium in case the coordinator is endowed with standard learning policies, while it returns a Nash equilibrium up to a constant, yet adjustable, error in the time-varying case. As a motivating application, we consider the ridehailing service provided by several companies with mobility as a service orchestration, necessary to both handle competition among firms and avoid traffic congestion, which is also adopted to run numerical experiments verifying our results
Abstract layer for leakyReLU for neural network verification based on abstract interpretation
No full textInternational audienceDeep neural networks have been widely used in several complex tasks such as robotics, self-driving cars, medicine, etc. However, they have recently shown to be vulnerable in uncertain environments where inputs are noisy. As a consequence, the robustness of neural networks has become an essential property for their application in critical systems. Robustness is the capacity to take the same decision even when inputs are disturbed under different types of perturbations, including adversarial attacks. The great difficulty today is providing a formal guarantee of robustness, which is the context of this paper. To do so, abstract interpretation, a popular state-of-the-art method, consisting of converting the layers of the neural network into abstract layers, has been recently proposed. An abstract layer can act on a geometric abstract object or shape comprising implicitly an infinite number of inputs rather than an individual input. In this paper, we propose a new mathematical formulation of an abstract transformer to convert a LeakyReLU activation layer to an abstract layer. Moreover, we implement and integrate our transformer into the ERAN tool. For validation, we assess the performance of our transformer according to the LeakyReLU hyperparameter, and we study the robustness of the neural network according to the input perturbation intensity. Our approach is evaluated on three different datasets: MNIST, Fashion and a robotic dataset. The obtained results demonstrate the efficacy of our abstract transformer in terms of mathematical formulation and implementation
Pourquoi et comment enseigner l'écriture créative aux ingénieurs ?
No full textInternational audienceHow and why should we teach creative writing workshops to engineers? The basic claim, defended here with both reference to scientific studies on creativity and anecdotal evidence drawn from teaching experience, is that creative writing workshops can be taught in such a way as to help future innovators to both better understand the creative process and to think deeply about the ethical and social impacts involved in acts of creation. The somewhat paradoxical claim animating this article is the idea that the people who may really need courses in creative writing are not writers but engineers, those whose creations may bring us towards science fiction futures, and who, for this very reason, need to be capable of creating fictions as a guide for imagining the ethical consequences of their innovations
Ultrasonic imaging in highly heterogeneous backgrounds
No full textInternational audienceThis work formally investigates the differential evolution indicators as a tool for ultrasonic tracking of elastic transformation and fracturing in randomly heterogeneous solids. Within the framework of periodic sensing, it is assumed that the background at time t◦ contains (i) a multiply connected set ofviscoelastic, anisotropic, and piece-wise homogeneous inclusions, and (ii) a union of possibly disjoint fractures and pores. The support, material properties, and interfacial condition of scatterers in (i) and (ii) are unknown, while elastic constants of the matrix are provided. The domain undergoes progressive variations of arbitrary chemo-mechanical origins such that its geometric configuration and elastic properties at future times are distinct. At every sensing step t◦, t1, . . ., multi-modal incidents are generated by a set of boundary excitations, and the resulting scattered fields are captured over the observation surface. The test data are then used to construct a sequence of wavefront densities by solving the spectral scatteringequation. The incident fields affiliated with distinct pairs of obtained wavefronts are analyzed over the stationary and evolving scatterers for a suit ofgeometric and elastic evolution scenarios entailing both interfacial and volumetric transformations. The main theorem establishes the invariance of pertinent incident fields at the loci of static fractures and inclusions between a given pair of time steps, while certifying variation of the same fields over the modified regions. These results furnish a basis for theoretical justification of differential evolution indicators for imaging in complex composites which, in turn, enable the exclusive tomography of evolution in a background endowed with many unknown features
The Claim of Reason in a Planetary Age
No full textInternational audienceThis essay is a creative inheritance destined for a volume celebrating the ongoing relevance of Thomas Kuhn and Stanley Cavell. But if it is inspired by, and converses with them, it is neither a reconstruction of their conversations nor a textual exegesis, but an attempt to reflect critically on the rationality of Earthlings in the Anthropocene while drawing orientation from Kuhn and Cavell. Arguably, such philosophical modernism is in spirit intensely Cavellian. Pursuing Emersonian self-reliance, this paper aims to make “philosophy yet another kind of problem for itself.” Therefore, this text is not Kuhnian. It couldn’t be — Kuhn claimed that his “vocation” was to be a “historian of science,” a member of the “American Historical, not the American Philosophical, Association.” But in its concern with science and history, and above all in its acceptance that our current historical context, the Anthropocene, cannot be thought outside of paradigmatic shifts within the history of science, notably the development of planetary science as a comparative and thus inter- planetary model for understanding our own terrestrial condition, what follows is Kuhnian