Search

Homomorphic encryption is one of the main solutions for building secure and privacy-preserving solutions for Machine Learning as a Service. This motivates the development of homomorphic algorithms for the main building blocks of AI, typically for the components of the various types of neural networks architectures. Among those components, we focus on the Softmax function, defined by $\mathrm{SM}(\mathbf{x}) = \left(\exp(x_i) / \sum_{j=1}^n \exp(x_j) \right)_{1\le i\le n}$. This function is deemed to be one of the most difficult to evaluate homomorphically, because of its multivariate nature and of the very large range of values for $\exp(x_i)$. The available homomorphic algorithms remain restricted, especially in large dimensions, while important applications such as Large Language Models (LLM) require computing Softmax over large dimensional vectors. In terms of multiplicative depth of the computation (a suitable measure of cost for homomorphic algorithms), our algorithm achieves $O(\log n)$ complexity for a fixed range of inputs, where $n$ is the Softmax dimension. Our algorithm is especially adapted to the situation where we must compute many Softmax at the same time, for instance, in the LLM situation. In that case, assuming that all Softmax calls are packed into $m$ ciphtertexts, the asymptotic amortized multiplicative depth cost per ciphertext is, again over a fixed range, $O(1 + m/N)$ for $N$ the homomorphic ring degree. The main ingredient of our algorithms is a normalize-and-square strategy, which interlaces the exponential computation over a large range and normalization, decomposing both in stabler and cheaper smaller steps. Comparing ourselves to the state of the art, our experiments show, in practice, a good accuracy and a gain of a factor 2.5 to 8 compared to state of the art solutions., Comment: ACM Conference on Computer and Communications Security (CCS) 2024

Anomaly detection in distributed systems such as High-Performance Computing (HPC) clusters is vital for early fault detection, performance optimisation, security monitoring, reliability in general but also operational insights. Deep Neural Networks have seen successful use in detecting long-term anomalies in multidimensional data, originating for instance from industrial or medical systems, or weather prediction. A downside of such methods is that they require a static input size, or lose data through cropping, sampling, or other dimensionality reduction methods, making deployment on systems with variability on monitored data channels, such as computing clusters difficult. To address these problems, we present DeepHYDRA (Deep Hybrid DBSCAN/Reduction-Based Anomaly Detection) which combines DBSCAN and learning-based anomaly detection. DBSCAN clustering is used to find point anomalies in time-series data, mitigating the risk of missing outliers through loss of information when reducing input data to a fixed number of channels. A deep learning-based time-series anomaly detection method is then applied to the reduced data in order to identify long-term outliers. This hybrid approach reduces the chances of missing anomalies that might be made indistinguishable from normal data by the reduction process, and likewise enables the algorithm to be scalable and tolerate partial system failures while retaining its detection capabilities. Using a subset of the well-known SMD dataset family, a modified variant of the Eclipse dataset, as well as an in-house dataset with a large variability in active data channels, made publicly available with this work, we furthermore analyse computational intensity, memory footprint, and activation counts. DeepHYDRA is shown to reliably detect different types of anomalies in both large and complex datasets.

The percentage of women receiving bachelors degrees in physics in the U.S. lags well behind that of men, and women leave the major at higher rates. Achieving equity in physics will mean that women stay in physics at the same rates as men, but this will require changes in the culture and support structures. A strong sense of belonging can lead to higher retention rates so interventions meant to increase dimensions of physics identity (interest, recognition, performance, and competence) may increase persistence overall and increase women's retention differentially. We describe our model in which mentorship, an understanding of career options (career conceptualization), and leadership are inputs into the development of these dimensions of physics identity. This paper includes preliminary results from a qualitative study that aims to better understand how career conceptualization, leadership, and mentorship contribute to the development of physics identity and belonging. We report results from a survey of 15 undergraduate physics students which was followed up by interviews with 5 of those students. The students were from a small private liberal arts college in the midwest region of the U.S. and a large public university in the southeast region of the U.S. classified as a Hispanic-serving institution (HSI). With respect to mentorship, we found that it could provide critical support for students' engagement in the physics community. Leadership experiences have not previously been positioned as an important input into identity, yet we found that they helped women in physics feel more confident, contributing to their recognition of themselves as physics people. While the data on how career conceptualization contributed to the building of identity is limited, there are some connections to recognition and competence, and it will be an interesting avenue of future exploration., Comment: 15 pages, 1 figure, Physical Review Physics Education Research, in press

The Learning With Errors ($\mathsf{LWE}$) problem asks to find $\mathbf{s}$ from an input of the form $(\mathbf{A}, \mathbf{b} = \mathbf{A}\mathbf{s}+\mathbf{e}) \in (\mathbb{Z}/q\mathbb{Z})^{m \times n} \times (\mathbb{Z}/q\mathbb{Z})^{m}$, for a vector $\mathbf{e}$ that has small-magnitude entries. In this work, we do not focus on solving $\mathsf{LWE}$ but on the task of sampling instances. As these are extremely sparse in their range, it may seem plausible that the only way to proceed is to first create $\mathbf{s}$ and $\mathbf{e}$ and then set $\mathbf{b} = \mathbf{A}\mathbf{s}+\mathbf{e}$. In particular, such an instance sampler knows the solution. This raises the question whether it is possible to obliviously sample $(\mathbf{A}, \mathbf{A}\mathbf{s}+\mathbf{e})$, namely, without knowing the underlying $\mathbf{s}$. A variant of the assumption that oblivious $\mathsf{LWE}$ sampling is hard has been used in a series of works to analyze the security of candidate constructions of Succinct Non interactive Arguments of Knowledge (SNARKs). As the assumption is related to $\mathsf{LWE}$, these SNARKs have been conjectured to be secure in the presence of quantum adversaries. Our main result is a quantum polynomial-time algorithm that samples well-distributed $\mathsf{LWE}$ instances while provably not knowing the solution, under the assumption that $\mathsf{LWE}$ is hard. Moreover, the approach works for a vast range of $\mathsf{LWE}$ parametrizations, including those used in the above-mentioned SNARKs. This invalidates the assumptions used in their security analyses, although it does not yield attacks against the constructions themselves.

Crossword puzzles are one of the most popular word games, played in different languages all across the world, where riddle style can vary significantly from one country to another. Automated crossword resolution is challenging, and typical solvers rely on large databases of previously solved crosswords. In this work, we extend WebCrow 2.0, an automatic crossword solver, to French, making it the first program for crossword solving in the French language. To cope with the lack of a large repository of clue-answer crossword data, WebCrow 2.0 exploits multiple modules, called experts, that retrieve candidate answers from heterogeneous resources, such as the web, knowledge graphs, and linguistic rules. We compared WebCrow's performance against humans in two different challenges. Despite the limited amount of past crosswords, French WebCrow was competitive, actually outperforming humans in terms of speed and accuracy, thus proving its capabilities to generalize to new languages., Comment: Accepted Paper for EAI Intetain 2023 - 14th EAI International Conference on Intelligent Technologies for Interactive Entertainment

Radiative shocks (RS) are complex phenomena which are ubiquitous in astrophysical environments. The study of such hypersonic shocks in the laboratory, under controlled conditions, is of primary interest to understand the physics at play and also to check the ability of numerical simulations to reproduce the experimental results. In this context, we conducted, at the Prague Asterix Laser System facility (PALS), the first experiments dedicated to the study of two counter-propagating radiative shocks propagating at non-equal speeds up to 25-50 km/s in noble gases at pressures ranging between 0.1 and 0.6 bar. These experiments highlighted the interaction between the two radiative precursors. This interaction is qualitatively but not quantitatively described by 1D simulations. Preliminary results obtained with XUV spectroscopy leading to the estimation of shock temperature and ion charge of the plasma are also presented.

We report on the design and first results from experiments looking at the formation of radiative shocks on the Shenguang-II (SG-II) laser at the Shanghai Institute of Optics and Fine Mechanics in China. Laser-heating of a two-layer CH/CH-Br foil drives a $\sim$40 km/s shock inside a gas-cell filled with argon at an initial pressure of 1 bar. The use of gas-cell targets with large (several mm) lateral and axial extent allows the shock to propagate freely without any wall interactions, and permits a large field of view to image single and colliding counter-propagating shocks with time resolved, point-projection X-ray backlighting ($\sim20$ $\mu$m source size, 4.3 keV photon energy). Single shocks were imaged up to 100 ns after the onset of the laser drive allowing to probe the growth of spatial non-uniformities in the shock apex. These results are compared with experiments looking at counter-propagating shocks, showing a symmetric drive which leads to a collision and stagnation from $\sim$40 ns onward. We present a preliminary comparison with numerical simulations with the radiation hydrodynamics code ARWEN, which provides expected plasma parameters for the design of future experiments in this facility., Comment: 16 pages, 5 figures, accepted for publication in High Power Laser Science and Engineering (25 March 2021)

Radiative shocks are strong shocks characterized by plasma at a high temperature emitting an important fraction of its energy as radiation. Radiative shocks are commonly found in many astrophysical systems and are templates of radiative hydrodynamic flows, which can be studied experimentally using high-power lasers. This is not only important in the context of laboratory astrophysics but also to benchmark numerical studies. We present details on the design of experiments on radiative shocks in xenon gas performed at the kJ scale PALS laser facility. It includes technical specifications for the tube targets design and numerical studies with the 1-D radiative hydrodynamics code MULTI. Emphasis is given to the technical feasibility of an XUV imaging diagnostic with a 21 nm (~58 eV) probing beam, which allows to probe simultaneously the post-shock and the precursor region ahead of the shock. The novel design of the target together with the improved X-ray optics and XUV source allow to show both the dense post-shock structure and the precursor of the radiative shock., Comment: 12 pages, 4 figures

Systolic arrays are a promising computing concept which is in particular inline with CMOS technology trends and linear algebra operations found in the processing of artificial neural networks. The recent success of such deep learning methods in a wide set of applications has led to a variety of models, which albeit conceptual similar as based on convolutions and fully-connected layers, in detail show a huge diversity in operations due to a large design space: An operand's dimension varies substantially since it depends on design principles such as receptive field size, number of features, striding, dilating and grouping of features. Last, recent networks extent previously plain feedforward models by various connectivity, such as in ResNet or DenseNet. The problem of choosing an optimal systolic array configuration cannot be solved analytically, thus instead methods and tools are required that facilitate a fast and accurate reasoning about optimality in terms of total cycles, utilization, and amount of data movements. In this work we introduce Camuy, a lightweight model of a weight-stationary systolic array for linear algebra operations that allows quick explorations of different configurations, such as systolic array dimensions and input/output bitwidths. Camuy aids accelerator designers in either finding optimal configurations for a particular network architecture or for robust performance across a variety of network architectures. It offers simple integration into existing machine learning tool stacks (e.g TensorFlow) through custom operators. We present an analysis of popular DNN models to illustrate how it can estimate required cycles, data movement costs, as well as systolic array utilization, and show how the progress in network architecture design impacts the efficiency of inference on accelerators based on systolic arrays., Comment: 12 pages, 6 figures

Models and observations indicate that the impact of matter accreting onto the surface of young stars produces regions at the base of accretion columns, in which optically thin and thick plasma components coexist. Thus an accurate description of these impacts requires to account for the effects of absorption and emission of radiation. We study the effects of radiation emerging from shock-heated plasma in impact regions on the structure of the pre-shock downfalling material. We investigate if a significant absorption of radiation occurs and if it leads to a pre-shock heating of the accreting gas. We developed a radiation hydrodynamics model describing an accretion column impacting onto the surface of a Classical T Tauri Star. The model takes into account the stellar gravity, the thermal conduction, and the effects of both radiative losses and absorption of radiation by matter in the non local thermodynamic equilibrium regime. After the impact, a hot slab of post-shock plasma develops at the base of the accretion column. Part of radiation emerging from the slab is absorbed by the pre-shock accreting material. As a result, the pre-shock accretion column gradually heats up to temperatures of $10^5$ K, forming a radiative precursor of the shock. The precursor has a thermal structure with the hottest part at $T \approx 10^5$ K, with size comparable to that of the hot slab, above the post-shock region. At larger distances the temperature gradually decreases to $T \approx 10^4$ K., Comment: Letter Accepted for publication in A&A

Modeling the dynamics of most astrophysical structures requires an adequate description of the radiation-matter interaction. Several numerical (magneto)hydrodynamics codes were upgraded with a radiation module to fulfill this request. However, those among them that use either the flux-limited diffusion (FLD) or the M1 radiation moment approaches are restricted to the local thermodynamic equilibrium (LTE). This assumption may be not valid in some astrophysical cases. We present an upgraded version of the LTE radiation-hydrodynamics module implemented in the PLUTO code, originally developed by Kolb et al. (2013), which we have extended to handle non-LTE regimes. Starting from the general frequency-integrated comoving-frame equations of radiation hydrodynamics (RHD), we have justified all the assumptions made to obtain the non-LTE equations actually implemented in the module, under the FLD approximation. An operator-split method is employed, with two substeps: the hydrodynamic part is solved with an explicit method by the solvers already available in PLUTO, the non-LTE radiation diffusion and energy exchange part is solved with an implicit method. The module is implemented in the PLUTO environment. It uses databases of radiative quantities that can be provided independently by the user: the radiative power loss, the Planck and Rosseland mean opacities. Our implementation has been validated through different tests, in particular radiative shock tests. The agreement with the semi-analytical solutions (when available) is good, with a maximum error of 7%. Moreover, we have proved that non-LTE approach is of paramount importance to properly model accretion shock structures. Our radiation FLD module represents a step toward the general non-LTE RHD modeling. The module is available, under request, for the community., Comment: 15 pages, 4 Figures, Accepted for publication in Section 15: Numerical methods and codes of Astronomy and Astrophysics. The official date of acceptance is 05/07/2019

Parsing is essential for a wide range of use cases, such as stream processing, bulk loading, and in-situ querying of raw data. Yet, the compute-intense step often constitutes a major bottleneck in the data ingestion pipeline, since parsing of inputs that require more involved parsing rules is challenging to parallelise. This work proposes a massively parallel algorithm for parsing delimiter-separated data formats on GPUs. Other than the state-of-the-art, the proposed approach does not require an initial sequential pass over the input to determine a thread's parsing context. That is, how a thread, beginning somewhere in the middle of the input, should interpret a certain symbol (e.g., whether to interpret a comma as a delimiter or as part of a larger string enclosed in double-quotes). Instead of tailoring the approach to a single format, we are able to perform a massively parallel FSM simulation, which is more flexible and powerful, supporting more expressive parsing rules with general applicability. Achieving a parsing rate of as much as 14.2 GB/s, our experimental evaluation on a GPU with 3584 cores shows that the presented approach is able to scale to thousands of cores and beyond. With an end-to-end streaming approach, we are able to exploit the full-duplex capabilities of the PCIe bus and hide latency from data transfers. Considering the end-to-end performance, the algorithm parses 4.8 GB in as little as 0.44 seconds, including data transfers.

Context. Accreted material onto CTTSs is expected to form a hot quasi-periodic plasma structure that radiates in X-rays. Simulations of this phenomenon only partly match with observations. They all rely on a static model for the chromosphere model and on the assumption that radiation and matter are decoupled. Aims. We explore the effects on the structure and on the dynamics of the accretion flow of both a shock-heated chromosphere and of the coupling between radiation and hydrodynamics. Methods. We simulate accretion columns falling onto a stellar chromosphere using the 1D ALE code AstroLabE. This code solves the hydrodynamics equations along with the two first momenta equations for radiation transfer, with the help of a dedicated opacity table for the coupling between matter and radiation. We derive the total electron and ions densities from collisional-radiative NLTE ionisation equilibrium. Results. The chromospheric acoustic heating has an impact on the duration of the cycle and on the structure of the heated slab. In addition, the coupling between radiation and hydrodynamics leads to a heating of the accretion flow and the chromosphere, inducing a possible unburial of the whole column. These two last conclusions are in agreement with the computed monochromatic intensity. Both effects (acoustic heating and radiation coupling) have an influence on the amplitude and temporal variations of the net X-ray luminosity, which varies between 30 and 94% of the incoming mechanical energy flux, depending on the model considered., Comment: Re-submitted to A&A, revised version to reflects referee's comments

Classical T Tauri stars (CTTSs) are young low-mass stellar objects accreting mass from their circumstellar disks. They are characterized by high levels of coronal activity as revealed by X-ray observations. This activity may affect the disk stability and the circumstellar environment. Here we investigate if an intense coronal activity due to flares occurring close to the accretion disk may perturb the inner disk stability, disrupt the inner part of the disk and, possibly, trigger accretion phenomena with rates comparable with those observed. We model a magnetized protostar surrounded by an accretion disk through 3D magnetohydrodinamic simulations. We explore cases characterized by a dipole plus an octupole stellar magnetic field configuration and different density of the disk or by different levels of flaring activity. As a result of the simulated intense flaring activity, we observe the formation of several loops that link the star to the disk; all these loops build up a hot extended corona with an X-ray luminosity comparable with typical values observed in CTTSs. The intense flaring activity close to the disk can strongly perturb the disk stability. The flares trigger overpressure waves which travel through the disk and modify its configuration. Accretion funnels may be triggered by the flaring activity, thus contributing to the mass accretion rate of the star. Accretion rates synthesized from the simulations are in a range between 10^-10 and 10^-9M_sun yr^-1 The accretion columns can be perturbed by the flares and they can interact with each other, possibly merging together in larger streams. As a result, the accretion pattern can be rather complex: the streams are highly inhomogeneous, with a complex density structure, and clumped, Comment: 12 pages, 9 figures

Past high-energy density laboratory experiments provided insights into the physics of supernovae, supernova remnants, and the destruction of interstellar clouds. In a typical experimental setting, a laser-driven planar blast wave interacts with a compositionally-homogeneous spherical or cylindrical target. In this work we propose a new laboratory platform that accounts for curvature of the impacting shock and density stratification of the target. Both characteristics reflect the conditions expected to exist shortly after a supernova explosion in a close binary system. We provide details of a proposed experimental design (laser drive, target configuration, diagnostic system), optimized to capture the key properties of recent ejecta-companion interaction models. Good qualitative agreement found between our experimental models and their astrophysical counterparts highlights strong potential of the proposed design to probe details of the ejecta-companion interaction for broad classes of objects by means of high energy density laboratory experiments., Comment: 13 pages, 14 figures, accepted for publication in ApJ

As a typical application, the Lenstra-Lenstra-Lovasz lattice basis reduction algorithm (LLL) is used to compute a reduced basis of the orthogonal lattice for a given integer matrix, via reducing a special kind of lattice bases. With such bases in input, we propose a new technique for bounding from above the number of iterations required by the LLL algorithm. The main technical ingredient is a variant of the classical LLL potential, which could prove useful to understand the behavior of LLL for other families of input bases., Comment: ISSAC'18