Search

The MUltiplexed Survey Telescope (MUST) is a 6.5-meter telescope under development. Dedicated to highly-multiplexed, wide-field spectroscopic surveys, MUST observes over 20,000 targets simultaneously using 6.2-mm pitch positioning robots within a ~5 deg2 field of view. MUST aims to carry out the first Stage-V spectroscopic survey in the 2030s to map the 3D Universe with over 100 million galaxies and quasars, spanning from the nearby Universe to redshift z~5.5, corresponding to around 1 billion years after the Big Bang. To cover this extensive redshift range, we present an initial conceptual target selection algorithm for different types of galaxies, from local bright galaxies, luminous red galaxies, and emission line galaxies to high-redshift (2 < z < 5.5) Lyman-break galaxies. Using Fisher forecasts, we demonstrate that MUST can address fundamental questions in cosmology, including the nature of dark energy, test of gravity theories, and investigations into primordial physics. This is the first paper in the series of science white papers for MUST, with subsequent developments focusing on additional scientific cases such as galaxy and quasar evolution, Milky Way physics, and dynamic phenomena in the time-domain Universe., Comment: To be submitted to SCPMA

ANNZ is a fast and simple algorithm which utilises artificial neural networks (ANNs), it was known as one of the pioneers of machine learning approaches to photometric redshift estimation decades ago. We enhanced the algorithm by introducing new activation functions like tanh, softplus, SiLU, Mish and ReLU variants; its new performance is then vigorously tested on legacy samples like the Luminous Red Galaxy (LRG) and Stripe-82 samples from SDSS, as well as modern galaxy samples like the Physics of the Accelerating Universe Survey (PAUS). This work focuses on testing the robustness of activation functions with respect to the choice of ANN architectures, particularly on its depth and width, in the context of galaxy photometric redshift estimation. Our upgraded algorithm, which we named ANNZ+, shows that the tanh and Leaky ReLU activation functions provide more consistent and stable results across deeper and wider architectures with > 1 per cent improvement in root-mean-square error ($\sigma_{\textrm{RMS}}$) and 68th percentile error ($\sigma_{68}$) when tested on SDSS data sets. While assessing its capabilities in handling high dimensional inputs, we achieved an improvement of 11 per cent in $\sigma_{\textrm{RMS}}$ and 6 per cent in $\sigma_{68}$ with the tanh activation function when tested on the 40-narrowband PAUS dataset; it even outperformed ANNZ2, its supposed successor, by 44 per cent in $\sigma_{\textrm{RMS}}$. This justifies the effort to upgrade the 20-year-old ANNZ, allowing it to remain viable and competitive within the photo-z community today. The updated algorithm ANNZ+ is publicly available at https://github.com/imdadmpt/ANNzPlus., Comment: 38 pages, 9 figures, revised version submitted to JCAP

Recommender systems have become integral to our digital experiences, from online shopping to streaming platforms. Still, the rationale behind their suggestions often remains opaque to users. While some systems employ a graph-based approach, offering inherent explainability through paths associating recommended items and seed items, non-experts could not easily understand these explanations. A popular alternative is to convert graph-based explanations into textual ones using a template and an algorithm, which we denote here as ''template-based'' explanations. Yet, these can sometimes come across as impersonal or uninspiring. A novel method would be to employ large language models (LLMs) for this purpose, which we denote as ''LLM-based''. To assess the effectiveness of LLMs in generating more resonant explanations, we conducted a pilot study with 25 participants. They were presented with three explanations: (1) traditional template-based, (2) LLM-based rephrasing of the template output, and (3) purely LLM-based explanations derived from the graph-based explanations. Although subject to high variance, preliminary findings suggest that LLM-based explanations may provide a richer and more engaging user experience, further aligning with user expectations. This study sheds light on the potential limitations of current explanation methods and offers promising directions for leveraging large language models to improve user satisfaction and trust in recommender systems., Comment: Presented to the Dutch-Belgian Workshop on Recommender Systems 2023 (14-15 December, 2023 - Antwerp, Belgium)

We investigate superconducting gallium in its $\alpha$ phase using scanning tunneling microscopy and spectroscopy at temperatures down to about 100 mK. High-resolution tunneling spectroscopies using both superconducting and normal tips show that superconducting $\alpha$-Ga is accurately described by Bardeen-Cooper-Schrieffer theory, with a gap $\Delta_{\rm Ga}$ = 163 $\mu$eV on the $\alpha-$Ga(112) facet, with highly homogeneous spectra over the surface, including atomic defects and step edges. Using a superconducting Pb tip, we furthermore study the low-bias conductance features of the Josephson junction formed between tip and sample. The features are accurately described by dynamical Coulomb blockade theory, highlighting $\alpha-$Ga as a possible platform for surface science studies of mesoscopic superconductivity.

We present photometric redshifts for 1 341 559 galaxies from the Physics of the Accelerating Universe Survey (PAUS) over 50.38 ${\rm deg}^{2}$ of sky to $i_{\rm AB}=23$. Redshift estimation is performed using DEEPz, a deep-learning photometric redshift code. We analyse the photometric redshift precision when varying the photometric and spectroscopic samples. Furthermore, we examine observational and instrumental effects on the precision of the photometric redshifts, and we compare photometric redshift measurements with those obtained using a template method-fitting BCNz2. Finally, we examine the use of photometric redshifts in the identification of close galaxy pairs. We find that the combination of samples from W1+W3 in the training of DEEPz significantly enhances the precision of photometric redshifts. This also occurs when we recover narrow band fluxes using broad bands measurements. We show that DEEPz determines the redshifts of galaxies in the prevailing spectroscopic catalogue used in the training of DEEPz with greater precision. For the faintest galaxies ($i_{\rm AB}=21-23$), we find that DEEPz improves over BCNz2 both in terms of the precision (20-50 per cent smaller scatter) and in returning a smaller outlier fraction in two of the wide fields. The catalogues were tested for the identification of close galaxy pairs, showing that DEEPz is effective for the identification of close galaxy pairs for samples with $i_{\rm AB} < 22.5$ and redshift $0.2 < z < 0.6$. In addition, identifying close galaxy pairs common between DEEPz and BCNz2 is a promising approach to improving the purity of the catalogues of these systems., Comment: 21 pages, 13 figures, submitted to A&A, link to the catalogue

In this paper, we introduce a novel post-processing algorithm that is both model-agnostic and does not require the sensitive attribute at test time. In addition, our algorithm is explicitly designed to enforce minimal changes between biased and debiased predictions; a property that, while highly desirable, is rarely prioritized as an explicit objective in fairness literature. Our approach leverages a multiplicative factor applied to the logit value of probability scores produced by a black-box classifier. We demonstrate the efficacy of our method through empirical evaluations, comparing its performance against other four debiasing algorithms on two widely used datasets in fairness research.

We present the Lyman-$\alpha$ (Ly$\alpha$) and ultraviolet (UV) luminosity function (LF), in bins of redshift, of quasars selected in the Physics of the Accelerating Universe Survey (PAUS). A sample of 915 objects was selected at $2.710^{44}$ ${\rm erg\,s}$$^{-1}$. In order to obtain corrections for the LF estimation, and assess the accuracy of our selection method, we produced mock catalogs of $0

Being the most prominent HI line, Ly$\alpha$ permeates the cosmic web in emission. Despite its potential as a cosmological probe, its detection on large scales remains elusive. We present a new methodology to perform Ly$\alpha$ intensity mapping with broad-band optical images, by cross-correlating them with Ly$\alpha$ forest data using a custom one-parameter estimator. We also develop an analytical large-scale Ly$\alpha$ emission model with two parameters (average luminosity $\langle L_{\rm Ly\alpha} \rangle$ and bias $b_{\rm e}$) that respects observational constraints from QSO luminosity functions. We compute a forecast for DECaLS/BASS $g$-band images cross-correlated with DESI Ly$\alpha$ forest data, setting guidelines for reducing images into Ly$\alpha$ intensity maps. Given the transversal scales of our cross-correlation (26.4 arcmin, $\sim$33 cMpc/h), our study effectively integrates Ly$\alpha$ emission over all the cosmic volume inside the DESI footprint at $2.2 < z < 3.4$ (the $g$-band Ly$\alpha$ redshift range). Over the parameter space ($\langle L_{\rm Ly\alpha} \rangle$, $b_{\rm e}$) sampled by our forecast, we find a 3$\sigma$ of large-scale structure in Ly$\alpha$ likely, with a probability of detection of 23.95\% for DESI-DECaLS/BASS, and 54.93\% for a hypothetical DESI phase II with twice as much Ly$\alpha$ QSOs. Without a detection, we derive upper bounds on $\langle L_{\rm Ly\alpha} \rangle$ competitive with optimistic literature estimates ($2.3 \pm 1 \cdot 10^{\rm 41}$ erg/s/cMpc$^3$ for DESI, and $\sim$35\% lower for its hypothetical phase II). Extrapolation to the DESI-Rubin overlap shows that a detection of large-scale structure with Ly$\alpha$ intensity mapping using next-generation imaging surveys is certain. [abridged], Comment: 26 pages, 24 figures, submitted to MNRAS. Comments welcome!

The Physics of the Accelerating Universe (PAU) camera is an optical narrow band and broad band imaging instrument mounted at the prime focus of the William Herschel Telescope. We describe the image calibration procedure of the PAU Survey data. We rely on an external photometric catalogue to calibrate our narrow band data using stars that have been observed by both datasets. We fit stellar templates to the stellar broad band photometry of the Sloan Digital Sky Survey and synthesise narrow band photometry that we compare to the PAUS narrow band data to determine their calibration. Consequently, the PAUS data are in the AB system as inherited from its reference calibrator. We do several tests to check the performance of the calibration. We find it self-consistent when comparing repeated observations of the same objects, with a good overall accuracy to the AB system which we estimate to be at the 2\% precision level and no significant trends as a function of narrow band filter or wavelength. Repeated observations allow us to build a spatial map of the illumination pattern of the system. We also check the wavelength dependence of the calibration comparing to stellar spectra. We find that using only blue stars reduces the effects of variations in the stellar template fitting to broad-band colours, improving the overall precision of the calibration to around 1\% and its wavelength uniformity. The photometric redshift performance obtained with the PAUS data attests to the validity of our calibration to reach the PAUS science goals., Comment: 18 pages, 20 figures, accepted for publication in MNRAS

Narrowband galaxy surveys have recently gained interest as a promising method to achieve the necessary accuracy on the photometric redshift estimate of individual galaxies for stage-IV cosmological surveys. One key advantage is the ability to provide higher spectral resolution information about galaxies that should allow a more accurate and precise estimation of galaxy stellar population properties. However, the impact of adding narrow-band photometry on the stellar population properties estimate is largely unexplored. The scope of this work is two-fold: on one side, leveraging the predictive power of broad-band and narrow-band data to infer galaxy physical properties such as stellar masses, ages, star formation rates and metallicities. On the other hand, evaluating the improvement of performance in estimating galaxy properties when we use narrow-band data instead of broad-band. In this work we measure the stellar population properties of a sample of galaxies in the COSMOS field for which both narrowband and broadband data are available. In particular, we employ narrowband data from PAUS and broad-band data from CFHTLS. We use two different spectral energy distribution fitting codes to measure galaxy properties, namely CIGALE and Prospector. We find that the increased spectral resolution of narrow-band photometry does not yield a substantial improvement on constraining galaxy properties using spectral energy distribution fitting. Still we find that we obtain a more diverse distribution of metallicities and dust optical depths with cigale when employing the narrowband data. The effect is not as prominent as expected, which we relate this to the low narrowband SNR of a majority of the galaxies, the respective drawbacks of both codes as well as the coverage only in the optical regime. The measured properties are afterwards compared to the COSMOS2020 catalogue, showing good agreement., Comment: Published in A&A. 18 pages, 16 figures, 2 tables

In a Josephson junction, the current phase relation relates the phase variation of the superconducting order parameter, $\varphi$, between the two superconducting leads connected through a weak link, to the dissipationless current . This relation is the fingerprint of the junction. It is usually dominated by a $\sin(\varphi)$ harmonic, however its precise knowledge is necessary to design superconducting quantum circuits with tailored properties. Here, we directly measure the current phase relation of a superconducting quantum interference device made with gate-tunable graphene Josephson junctions and we show that it can behave as a $\sin(2\varphi)$ Josephson element, free of the traditionally dominant $\sin(\varphi)$ harmonic. Such element will be instrumental for the development of superconducting quantum bits protected from decoherence., Comment: 15 pages, 3 figures, Supplemental Material in Ancillary files

Two-dimensional and van der Waals ferromagnets are ideal platform to study low dimensional magnetism and proximity effects in van der Waals heterostructures. Their ultimate two dimensional character offers also the opportunity to easily adjust their magnetic properties using strain or electric fields. Among 2D ferromagnets, the Cr$_{1+x}$Te$_2$ compounds with $x$=0-1 are very promising because their magnetic properties depend on the amount of self-intercalated Cr atoms between pure CrTe$_2$ layers and the Curie temperature (T$_C$) can reach room temperature for certain compositions. Here, we investigate the evolution of the composition, structural and magnetic properties of thin Cr$_{1.33}$Te$_2$ (Cr$_2$Te$_3$) films epitaxially grown on graphene upon annealing. We observe a transition above 450{\deg}C from the Cr$_{1.33}$Te$_2$ phase with perpendicular magnetic anisotropy and a T$_C$ of 180 K to a composition close to Cr$_{1.39}$Te$_2$ with in-plane magnetic anisotropy and a T$_C$ of 240-250 K. This phase remains stable up to 650{\deg}C above which a pure Cr film starts to form. This work demonstrates the complex interplay between intercalated Cr, lattice parameters and magnetic properties in Cr$_{1+x}$Te$_2$ compounds., Comment: 5 pages, 5 figures

Designing protein nanomaterials of predefined shape and characteristics has the potential to dramatically impact the medical industry. Machine learning (ML) has proven successful in protein design, reducing the need for expensive wet lab experiment rounds. However, challenges persist in efficiently exploring the protein fitness landscapes to identify optimal protein designs. In response, we propose the use of AlphaZero to generate protein backbones, meeting shape and structural scoring requirements. We extend an existing Monte Carlo tree search (MCTS) framework by incorporating a novel threshold-based reward and secondary objectives to improve design precision. This innovation considerably outperforms existing approaches, leading to protein backbones that better respect structural scores. The application of AlphaZero is novel in the context of protein backbone design and demonstrates promising performance. AlphaZero consistently surpasses baseline MCTS by more than 100% in top-down protein design tasks. Additionally, our application of AlphaZero with secondary objectives uncovers further promising outcomes, indicating the potential of model-based reinforcement learning (RL) in navigating the intricate and nuanced aspects of protein design

We use (versions of) the von Neumann inequality for Hilbert space contractions to prove several Schwarz-Pick inequalities. Specifically, we derive an alternate proof for a multi-point Schwarz-Pick inequality by Beardon and Minda, along with a generalized version for operators. Connections with model spaces and Peschl's invariant derivatives are established. Finally, Schwarz-Pick inequalities for analytic functions on polydisks and for higher order derivatives are discussed. An enhanced version of the Schwarz-Pick lemma, using the notion of distinguished variety, is obtained for the bidisk., Comment: 29 pages. Minor modifications. This version is close to the published one

Explainability is a key component in many applications involving deep neural networks (DNNs). However, current explanation methods for DNNs commonly leave it to the human observer to distinguish relevant explanations from spurious noise. This is not feasible anymore when going from easily human-accessible data such as images to more complex data such as genome sequences. To facilitate the accessibility of DNN outputs from such complex data and to increase explainability, we present a modification of the widely used explanation method layer-wise relevance propagation. Our approach enforces sparsity directly by pruning the relevance propagation for the different layers. Thereby, we achieve sparser relevance attributions for the input features as well as for the intermediate layers. As the relevance propagation is input-specific, we aim to prune the relevance propagation rather than the underlying model architecture. This allows to prune different neurons for different inputs and hence, might be more appropriate to the local nature of explanation methods. To demonstrate the efficacy of our method, we evaluate it on two types of data, images and genomic sequences. We show that our modification indeed leads to noise reduction and concentrates relevance on the most important features compared to the baseline., Comment: 15 pages, 5 figures

Superconductivity in ultra-doped $Si_{1-x}Ge_{x}:B$ epilayers is demonstrated by nanosecond laser doping, which allows introducing substitutional B concentrations well above the solubility limit and up to $7\,at.\%$. A Ge fraction $x$ ranging from 0 to 0.21 is incorporated in $Si:B$ : 1) through a precursor gas by Gas Immersion Laser Doping; 2) by ion implantation, followed by nanosecond laser annealing; 3) by UHV-CVD growth of a thin Ge layer, followed by nanosecond laser annealing. The 30 nm and 80 nm thick $Si_{1-x}Ge_{x}:B$ epilayers display superconducting critical temperatures $T_c$ tuned by B and Ge between 0 and 0.6 K. Within BCS weak-coupling theory, $T_c$ evolves exponentially with both the density of states and the electron-phonon potential. While B doping affects both, through the increase of the carrier density and the tensile strain, Ge incorporation allows addressing independently the lattice deformation influence on superconductivity. To estimate the lattice parameter modulation with B and Ge, Vegard's law is validated for the ternary $SiGeB$ bulk alloy by Density Functional Theory calculations. Its validity is furthermore confirmed experimentally by X-Ray Diffraction. We highlight a global linear dependence of $T_c$ vs. lattice parameter, common for both $Si:B$ and $Si_{1-x}Ge_{x}:B$, with $\delta T_c/T_c \sim 50\,\%$ for $\delta a/a \sim 1\,\%$.

Given a dataset of expert demonstrations, inverse reinforcement learning (IRL) aims to recover a reward for which the expert is optimal. This work proposes a model-free algorithm to solve entropy-regularized IRL problem. In particular, we employ a stochastic gradient descent update for the reward and a stochastic soft policy iteration update for the policy. Assuming access to a generative model, we prove that our algorithm is guaranteed to recover a reward for which the expert is $\varepsilon$-optimal using $\mathcal{O}(1/\varepsilon^{2})$ samples of the Markov decision process (MDP). Furthermore, with $\mathcal{O}(1/\varepsilon^{4})$ samples we prove that the optimal policy corresponding to the recovered reward is $\varepsilon$-close to the expert policy in total variation distance.

One key challenge in Artificial Life is designing systems that display an emergence of complex behaviors. Many such systems depend on a high-dimensional parameter space, only a small subset of which displays interesting dynamics. Focusing on the case of continuous systems, we introduce the 'Phase Transition Finder'(PTF) algorithm, which can be used to efficiently generate parameters lying at the border between two phases. We argue that such points are more likely to display complex behaviors, and confirm this by applying PTF to Lenia showing it can increase the frequency of interesting behaviors more than two-fold, while remaining efficient enough for large-scale searches., Comment: 5 pages

Localizing fast radio bursts (FRBs) to their host galaxies is an essential step to better understanding their origins and using them as cosmic probes. The CHIME/FRB Outrigger program aims to add VLBI-localization capabilities to CHIME, such that FRBs may be localized to tens of milliarcsecond precision at the time of their discovery, more than sufficient for host galaxy identification. The first-built outrigger telescope is KKO, located 66 kilometers west of CHIME. Cross-correlating KKO with CHIME can achieve arcsecond-scale localization in right ascension while avoiding the worst effects of the ionosphere. This paper presents measurements of KKO's performance throughout its commissioning phase, as well as a summary of its design and function. We demonstrate KKO's capabilities as a standalone instrument by producing full-sky images, mapping the angular and frequency structure of the primary beam, and measuring feed positions. To demonstrate the localization capabilities of the CHIME -- KKO baseline, we collected five separate observations each for a set of twenty bright pulsars, and aimed to measure their positions to within 5~arcseconds. All of these pulses were successfully localized to within this specification. The next two outriggers are expected to be commissioned in 2024, and will enable subarcsecond localizations for approximately hundreds of FRBs each year., Comment: 41 pages, 17 figures

We have introduced a q-deformation, i.e., a polynomial in q with natural coefficients, of the binomial coefficient of two finite words u and v counting the number of occurrences of v as a subword of u. In this paper, we examine the q-deformation of Parikh matrices as introduced by E\u{g}ecio\u{g}lu in 2004. Many classical results concerning Parikh matrices generalize to this new framework: Our first important observation is that the elements of such a matrix are in fact q-deformations of binomial coefficients of words. We also study their inverses and as an application, we obtain new identities about q-binomials. For a finite word z and for the sequence $(p_n)_{n\ge 0}$ of prefixes of an infinite word, we show that the polynomial sequence $\binom{p_n}{z}_q$ converges to a formal series. We present links with additive number theory and k-regular sequences. In the case of a periodic word $u^\omega$, we generalize a result of Salomaa: the sequence $\binom{u^n}{z}_q$ satisfies a linear recurrence relation with polynomial coefficients. Related to the theory of integer partition, we describe the growth and the zero set of the coefficients of the series associated with $u^\omega$. Finally, we show that the minors of a q-Parikh matrix are polynomials with natural coefficients and consider a generalization of Cauchy's inequality. We also compare q-Parikh matrices associated with an arbitrary word with those associated with a canonical word $12\cdots k$ made of pairwise distinct symbols., Comment: 26 pages, submitted

Gaussian binomial coefficients are q-analogues of the binomial coefficients of integers. On the other hand, binomial coefficients have been extended to finite words, i.e., elements of the finitely generated free monoids. In this paper we bring together these two notions by introducing q-analogues of binomial coefficients of words. We study their basic properties, e.g., by extending classical formulas such as the q-Vandermonde and Manvel's et al. identities to our setting. As a consequence, we get information about the structure of the considered words: these q-deformations of binomial coefficients of words contain much richer information than the original coefficients. From an algebraic perspective, we introduce a q-shuffle and a family q-infiltration products for non-commutative formal power series. Finally, we apply our results to generalize a theorem of Eilenberg characterizing so-called p-group languages. We show that a language is of this type if and only if it is a Boolean combination of specific languages defined through q-binomial coefficients seen as polynomials over $\mathbb{F}_p$., Comment: 25 pages, submitted

We introduce Mixtral 8x7B, a Sparse Mixture of Experts (SMoE) language model. Mixtral has the same architecture as Mistral 7B, with the difference that each layer is composed of 8 feedforward blocks (i.e. experts). For every token, at each layer, a router network selects two experts to process the current state and combine their outputs. Even though each token only sees two experts, the selected experts can be different at each timestep. As a result, each token has access to 47B parameters, but only uses 13B active parameters during inference. Mixtral was trained with a context size of 32k tokens and it outperforms or matches Llama 2 70B and GPT-3.5 across all evaluated benchmarks. In particular, Mixtral vastly outperforms Llama 2 70B on mathematics, code generation, and multilingual benchmarks. We also provide a model fine-tuned to follow instructions, Mixtral 8x7B - Instruct, that surpasses GPT-3.5 Turbo, Claude-2.1, Gemini Pro, and Llama 2 70B - chat model on human benchmarks. Both the base and instruct models are released under the Apache 2.0 license., Comment: See more details at https://mistral.ai/news/mixtral-of-experts/

The "100-year flood" is commonly used, for instance in newspapers, but flood hazard assessment is more complex than it seems. We first describe an animation entitled "bag of floods" to make flood quantiles more concrete, using marbles whose colour corresponds to a class of return period. Discussing the analogies and differences between drawing a marble from the bag and the next annual flood make it easier to explain that flood hazard assessment (i) must not be focussed on the "100-yr flood", (ii) is often expressed as a probability over one given year, but for planning it should be estimated over a much longer duration (like successive draws from the bag) and (iii) variability is significant and matters. Scripts allowing to simulate long series of draws confirm that empirical probabilities get close to theoretical probabilities, but also illustrate less intuitive results : on average one quarter of 100 successive draws, contains two floods or more with a discharge exceeding the "100-yr discharge". To go further, Sample2Gumbel is a teaching tool drawing annual maxima discharges. It compares on a graph (i) a "real distribution", coded in the script and used to draw a sample, (ii) the sample expressed with respect to "plotting position", expressed as a return period but which is in fact a crude estimation to allow plotting, (iii) the distribution fitted on the sample. This demo tool illustrates the variability of different tries, with samples of the same length, and shows how uncertainty evolves with the sample size. To improve it, more distributions could be included, and damage estimation could be added., Comment: in French language: Novatech 2023, 11e conf{\'e}rence internationale ''L'eau dans la ville'', Novatech, Jul 2023, Lyon, France

Explainability and uncertainty quantification are two pillars of trustable artificial intelligence. However, the reasoning behind uncertainty estimates is generally left unexplained. Identifying the drivers of uncertainty complements explanations of point predictions in recognizing model limitations and enhances trust in decisions and their communication. So far, explanations of uncertainties have been rarely studied. The few exceptions rely on Bayesian neural networks or technically intricate approaches, such as auxiliary generative models, thereby hindering their broad adoption. We present a simple approach to explain predictive aleatoric uncertainties. We estimate uncertainty as predictive variance by adapting a neural network with a Gaussian output distribution. Subsequently, we apply out-of-the-box explainers to the model's variance output. This approach can explain uncertainty influences more reliably than literature baselines, which we evaluate in a synthetic setting with a known data-generating process. We further adapt multiple metrics from conventional XAI research to uncertainty explanations. We quantify our findings with a nuanced benchmark analysis that includes real-world datasets. Finally, we apply our approach to an age regression model and discover reasonable sources of uncertainty. Overall, we explain uncertainty estimates with little modifications to the model architecture and demonstrate that our approach competes effectively with more intricate methods., Comment: Simon Witzke and Pascal Iversen contributed equally

We present photometric redshifts (photo-$z$) for the deep wide fields of the Physics of the Accelerating Universe Survey (PAUS), covering an area of $\sim$50 deg$^{2}$, for $\sim$1.8 million objects up to $i_{\textrm{AB}}<23$. The PAUS deep wide fields overlap with the W1 and W3 fields from CFHTLenS and the G09 field from KiDS/GAMA. Photo-$z$ are estimated using the 40 narrow bands (NB) of PAUS and the broad bands (BB) of CFHTLenS and KiDS. We compute the redshifts with the SED template-fitting code BCNZ, with a modification in the calibration technique of the zero-point between the observed and the modelled fluxes, that removes any dependence on spectroscopic redshift samples. We enhance the redshift accuracy by introducing an additional photo-$z$ estimate ($z_{\textrm{b}}$), obtained through the combination of the BCNZ and the BB-only photo-$z$. Comparing with spectroscopic redshifts estimates ($z_{\textrm{s}}$), we obtain a $\sigma_{68} \simeq 0.019$ for all galaxies with $i_{\textrm{AB}}<23$ and a typical bias $|z_{\textrm{b}}-z_{\textrm{s}}|$ smaller than 0.01. For $z_{\textrm{b}} \sim (0.10-0.75)$ we find $\sigma_{68} \simeq (0.003-0.02)$, this is a factor of $10-2$ higher accuracy than the corresponding BB-only results. We obtain similar performance when we split the samples into red (passive) and blue (active) galaxies. We validate the redshift probability $p(z)$ obtained by BCNZ and compare its performance with that of $z_{\textrm{b}}$. These photo-$z$ catalogues will facilitate important science cases, such as the study of galaxy clustering and intrinsic alignment at high redshifts ($z \lesssim 1$) and faint magnitudes., Comment: 24 pages, 26 figures, submitted to MNRAS

Monitoring water stage and discharge at hydrometric stations is essential for flood characterization and prediction. Continuous measurement is feasible for stage records, whereas discharge must be calculated, typically using a rating curve. Several methods have been developed, such as surface velocity measurement or rating curve methods. Nevertheless, hydrometric stations may be influenced by tidal fluctuations, leading to transient flow conditions that can disrupt the stage-discharge relation, complicating the accurate estimation of discharge. In the case of a quasi-unsteady flow, the dynamic can be managed by a relationship between water level, elevation, and discharge based on the Manning-Strickler formula and measurements of water level and slope of the waterline. However, in unsteady flow, when tidal effects are pronounced, these types of relationships and their variations have proven to be less effective. To capture the complex flow dynamics, including flow reversal, an approach via a 1D hydrodynamic model is proposed. Here, model estimation refers to estimating the posterior distribution of parameters and the structural error model.To set up the model, the cross-sectional geometry, friction coefficient, upstream discharge(s), and downstream water level are necessary. In hydrodynamic modeling, the friction coefficients, often represented through a set of Strickler coefficients, are the main calibration parameter, but manual calibration is difficult due to spatial variations in roughness combined with unsteady flow. Furthermore, understanding and quantifying uncertainties associated with data and the model are an important step in the calibration process. Therefore, automatic calibration of friction coefficients is proposed through Bayesian inference.In terms of numerical tools, the 1D hydrodynamic code used here is Mage, developed by INRAE, which solves the 1D Saint-Venant equations for river flows and transients. However, the proposed method is not specific to a given simulation code: it can be applied to any usual 1D hydrodynamic code. Bayesian calibration is performed using the software BaM! (Bayesian modeling: https://github.com/BaM-tools), which allows specifying prior information on model parameters (in this case, friction coefficients) to then estimate them with associated uncertainties, using observations themselves with their uncertainties (not only water levels but also gauging campaigns)., Comment: in French language, Colloque SHF "Pr{\'e}vision des crues et des inondations -- Avanc{\'e}es, valorisation et perspectives'', Soci{\'e}t{\'e} Hydrotechnique de France (SHF), Nov 2023, Toulouse, France

Extreme adaptive optics instruments have revealed exquisite details on debris discs, allowing to extract the optical properties of the dust particles such as the phase function, the degree of polarisation and the spectral reflectance. These are three powerful diagnostic tools to understand the physical properties of the dust : the size, shape and composition of the dust particles. This can inform us on the population of parent bodies, also called planetesimals, which generate those particles through collisions. It is however very rare to be able to combine all those three observables for the same system, as this requires different high-contrast imaging techniques to suppress the starlight and reveal the faint scattered light emission from the dust. Due to its brightness, the ring detected around the A-type star HR 4796 is a notable exception, with both unpolarised and polarised images covering near-infrared wavelengths. Here, we show how measurements of dust particles in the laboratory can reproduce the observed near-infrared photo-polarimetric properties of the HR 4796 disc. Experimental characterisation of dust allows to bypass the current limitations of dust models to reproduce simultaneously the phase function, the degree of polarisation and the spectral reflectance., Comment: 4 pages, 4 figures, Proceedings of the Annual meeting of the French Society of Astronomy and Astrophysics

Polarisation is a powerful remote-sensing tool to study the nature of particles scattering the starlight. It is widely used to characterise interplanetary dust particles in the Solar System and increasingly employed to investigate extrasolar dust in debris discs' systems. We aim to measure the scattering properties of the dust from the debris ring around HD 181327 at near-infrared wavelengths. We obtained high-contrast polarimetric images of HD 181327 in the H band with the SPHERE / IRDIS instrument on the Very Large Telescope (ESO). We complemented them with archival data from HST / NICMOS in the F110W filter reprocessed in the context of the Archival Legacy Investigations of Circumstellar Environments (ALICE) project. We developed a combined forward-modelling framework to simultaneously retrieve the scattering phase function in polarisation and intensity. We detected the debris disc around HD 181327 in polarised light and total intensity. We measured the scattering phase function and the degree of linear polarisation of the dust at 1.6 micron in the birth ring. The maximum polarisation is 23.6% +/- 2.6% and occurs between a scattering angle of 70 deg and 82 deg. We show that compact spherical particles made of a highly refractive and relatively absorbing material in a differential power-law size distribution of exponent $-3.5$ can simultaneously reproduce the polarimetric and total intensity scattering properties of the dust. This type of material cannot be obtained with a mixture of silicates, amorphous carbon, water ice, and porosity, and requires a more refracting component such as iron-bearing minerals. We reveal a striking analogy between the near-infrared polarisation of comets and that of HD 181327. The methodology developed here combining VLT/SPHERE and HST/NICMOS may be applicable in the future to combine the polarimetric capabilities of SPHERE with the sensitivity of JWST., Comment: 21 pages, 22 figures, accepted in Astronomy & Astrophysics

Event detection in time series is a challenging task due to the prevalence of imbalanced datasets, rare events, and time interval-defined events. Traditional supervised deep learning methods primarily employ binary classification, where each time step is assigned a binary label indicating the presence or absence of an event. However, these methods struggle to handle these specific scenarios effectively. To address these limitations, we propose a novel supervised regression-based deep learning approach that offers several advantages over classification-based methods. Our approach, with a limited number of parameters, can effectively handle various types of events within a unified framework, including rare events and imbalanced datasets. We provide theoretical justifications for its universality and precision and demonstrate its superior performance across diverse domains, particularly for rare events and imbalanced datasets.

We use the GALFORM semi-analytical galaxy formation model implemented in the Planck Millennium N-body simulation to build a mock galaxy catalogue on an observer's past lightcone. The mass resolution of this N-body simulation is almost an order of magnitude better than in previous simulations used for this purpose, allowing us to probe fainter galaxies and hence build a more complete mock catalogue at low redshifts. The high time cadence of the simulation outputs allows us to make improved calculations of galaxy properties and positions in the mock. We test the predictions of the mock against the Physics of the Accelerating Universe Survey, a narrow band imaging survey with highly accurate and precise photometric redshifts, which probes the galaxy population over a lookback time of 8 billion years. We compare the model against the observed number counts, redshift distribution and evolution of the observed colours and find good agreement; these statistics avoid the need for model-dependent processing of the observations. The model produces red and blue populations that have similar median colours to the observations. However, the bimodality of galaxy colours in the model is stronger than in the observations. This bimodality is reduced on including a simple model for errors in the GALFORM photometry. We examine how the model predictions for the observed galaxy colours change when perturbing key model parameters. This exercise shows that the median colours and relative abundance of red and blue galaxies provide constraints on the strength of the feedback driven by supernovae used in the model.

Event detection in time series data is crucial in various domains, including finance, healthcare, cybersecurity, and science. Accurately identifying events in time series data is vital for making informed decisions, detecting anomalies, and predicting future trends. Despite extensive research exploring diverse methods for event detection in time series, with deep learning approaches being among the most advanced, there is still room for improvement and innovation in this field. In this paper, we present a new deep learning supervised method for detecting events in multivariate time series data. Our method combines four distinct novelties compared to existing deep-learning supervised methods. Firstly, it is based on regression instead of binary classification. Secondly, it does not require labeled datasets where each point is labeled; instead, it only requires reference events defined as time points or intervals of time. Thirdly, it is designed to be robust by using a stacked ensemble learning meta-model that combines deep learning models, ranging from classic feed-forward neural networks (FFNs) to state-of-the-art architectures like transformers. This ensemble approach can mitigate individual model weaknesses and biases, resulting in more robust predictions. Finally, to facilitate practical implementation, we have developed a Python package to accompany our proposed method. The package, called eventdetector-ts, can be installed through the Python Package Index (PyPI). In this paper, we present our method and provide a comprehensive guide on the usage of the package. We showcase its versatility and effectiveness through different real-world use cases from natural language processing (NLP) to financial security domains., Comment: 2023 International Conference on Machine Learning and Applications (ICMLA)

Let $G$ be a real reductive linear group in the Harish-Chandra class. Suppose that $P$ is a parabolic subgroup of $G$ with Langlands decomposition $P=MAN$. Let $\pi$ be an irreducible representation of the Levi factor $L=MA$. We give sufficient conditions on the infinitesimal character of $\pi$ for the induced representation $i_P^G(\pi)$ to be irreducible. In particular, we prove that if $\pi_M$ is an irreducible representation of $M$, then for a generic character $\chi_\nu$ of $A$, the induced representation $i_P^G(\pi_M\boxtimes \chi_\nu)$ is irreducible. Here the parameter $\nu$ is in $\mathfrak{a}^*=(\mathrm{Lie}(A)\otimes_\mathbb R \mathbb C)^*$ and generic means outside a countable, locally finite union of hyperplanes which depends only on the infinitesimal character of $\pi$. Notice that there is no other assumption on $\pi$ or $\pi_M$ than being irreducible, so the result is not limited to generalised principal series or standard representations, for which the result is already well known., Comment: 18 pages, version 3, typos, mistakes and omissions corrected, references added, statements of the main results reformulated

We introduce Mistral 7B v0.1, a 7-billion-parameter language model engineered for superior performance and efficiency. Mistral 7B outperforms Llama 2 13B across all evaluated benchmarks, and Llama 1 34B in reasoning, mathematics, and code generation. Our model leverages grouped-query attention (GQA) for faster inference, coupled with sliding window attention (SWA) to effectively handle sequences of arbitrary length with a reduced inference cost. We also provide a model fine-tuned to follow instructions, Mistral 7B -- Instruct, that surpasses the Llama 2 13B -- Chat model both on human and automated benchmarks. Our models are released under the Apache 2.0 license., Comment: Models and code are available at https://mistral.ai/news/announcing-mistral-7b/