Search

Distributional regression aims at estimating the conditional distribution of a targetvariable given explanatory co-variates. It is a crucial tool for forecasting whena precise uncertainty quantification is required. A popular methodology consistsin fitting a parametric model via empirical risk minimization where the risk ismeasured by the Continuous Rank Probability Score (CRPS). For independentand identically distributed observations, we provide a concentration result for theestimation error and an upper bound for its expectation. Furthermore, we considermodel selection performed by minimization of the validation error and provide aconcentration bound for the regret. A similar result is proved for convex aggregationof models. Finally, we show that our results may be applied to various models suchas Ensemble Model Output Statistics (EMOS), distributional regression networks,distributional nearest neighbors or distributional random forests and we illustrateour findings on two data sets (QSAR aquatic toxicity and Airfoil self-noise).

We study a system of drift-diffusion PDEs for a potentially infinite number of incompressible phases, subject to a joint pointwise volume constraint. Our analysis is based on the interpretation as a collection of coupled Wasserstein gradient flows or, equivalently, as a gradient flow in the space of couplings under a `fibered' Wasserstein distance. We prove existence of weak solutions, long-time asymptotics, and stability with respect to the mass distribution of the phases, including the discrete to continuous limit. A key step is to establish convergence of the product of pressure gradient and density, jointly over the infinite number of phases. The underlying energy functional is the objective of entropy regularized optimal transport, which allows us to interpret the model as the relaxation of the classical Angenent-Haker-Tannenbaum (AHT) scheme to the entropic setting. However, in contrast to the AHT scheme's lack of convergence guarantees, the relaxed scheme is unconditionally convergent. We conclude with numerical illustrations of the main results., Comment: 36 pages, 8 figures

Basket trials can efficiently evaluate a single treatment across multiple diseases with a common shared target. Prior methods for randomized basket trials required baskets to have the same sample and effect sizes. To that end, we developed a general randomized basket trial with an interim analysis (RaBIt) that allows for unequal sample sizes and effect sizes per basket. RaBIt is characterized by pruning at an interim stage and then analyzing a pooling of the remaining baskets. We derived the analytical power and type 1 error for the design. We first show that our results are consistent with the prior methods when the sample and effect sizes were the same across baskets. As we adjust the sample allocation between baskets, our threshold for the final test statistic becomes more stringent in order to maintain the same overall type 1 error. Finally, we notice that if we fix a sample size for the baskets proportional to their accrual rate, then at the cost of an almost negligible amount of power, the trial overall is expected to take substantially less time than the non-generalized version., Comment: 23 pages, 3 figures

In the last two years, text-to-image diffusion models have become extremely popular. As their quality and usage increase, a major concern has been the need for better output control. In addition to prompt engineering, one effective method to improve the controllability of diffusion models has been to condition them on additional modalities such as image style, depth map, or keypoints. This forms the basis of ControlNets or Adapters. When attempting to apply these methods to control human poses in outputs of text-to-image diffusion models, two main challenges have arisen. The first challenge is generating poses following a wide range of semantic text descriptions, for which previous methods involved searching for a pose within a dataset of (caption, pose) pairs. The second challenge is conditioning image generation on a specified pose while keeping both high aesthetic and high pose fidelity. In this article, we fix these two main issues by introducing a text-to-pose (T2P) generative model alongside a new sampling algorithm, and a new pose adapter that incorporates more pose keypoints for higher pose fidelity. Together, these two new state-of-the-art models enable, for the first time, a generative text-to-pose-to-image framework for higher pose control in diffusion models. We release all models and the code used for the experiments at https://github.com/clement-bonnet/text-to-pose., Comment: Published at the NeurIPS 2024 Workshop on Compositional Learning: Perspectives, Methods, and Paths Forward

Independent vector analysis (IVA) is an attractive solution to address the problem of joint blind source separation (JBSS), that is, the simultaneous extraction of latent sources from several datasets implicitly sharing some information. Among IVA approaches, we focus here on the celebrated IVA-G model, that describes observed data through the mixing of independent Gaussian source vectors across the datasets. IVA-G algorithms usually seek the values of demixing matrices that maximize the joint likelihood of the datasets, estimating the sources using these demixing matrices. Instead, we write the likelihood of the data with respect to both the demixing matrices and the precision matrices of the source estimate. This allows us to formulate a cost function whose mathematical properties enable the use of a proximal alternating algorithm based on closed form operators with provable convergence to a critical point. After establishing the convergence properties of the new algorithm, we illustrate its desirable performance in separating sources with covariance structures that represent varying degrees of difficulty for JBSS.

We present a finite volume scheme for modeling the diffusion of charged particles, specifically ions, in constrained geometries using a degenerate Poisson-Nernst-Planck system with size exclusion yielding cross-diffusion. Our method utilizes a two-point flux approximation and is part of the exponentially fitted scheme framework. The scheme is shown to be thermodynamically consistent, as it ensures the decay of some discrete version of the free energy. Classical numerical analysis results -- existence of discrete solution, convergence of the scheme as the grid size and the time step go to $0$ -- follow. We also investigate the long-time behavior of the scheme, both from a theoretical and numerical point of view. Numerical simulations confirm our findings, but also point out some possibly very slow convergence towards equilibrium of the system under consideration.

Condensation processes, which are responsible for the main chemical differences between gas and solids in the Galaxy, are the major mechanisms that control the cycle of dust from evolved stars to planetary systems. However, they are still poorly understood, mainly because the thermodynamics and kinetic models of nucleation or grain growth lack experimental data. To bridge this gap, we used a large-volume three-phase alternating-current plasma torch to obtain a full high-temperature condensation sequence at an elevated carbon-to-oxygen ratio from a fluxed chondritic gas composition. We show that the crystallized suites of carbides, silicides, nitrides, sulfides, oxides and silicates and the bulk composition of the condensates are properly modelled by a kinetically inhibited condensation scenario controlled by gas flow. This validates the thermodynamic predictions of the condensation sequence at a high carbon-to-oxygen ratio. On this basis and using appropriate optical properties, we also demonstrate the influence of pressure on dust chemistry as well as the low probability of forming and detecting iron silicides in asymptotic giant branch C-rich circumstellar environments as well as in our chondritic meteorites. By demonstrating the potential of predicting dust mineralogy in these environments, this approach holds high promise for quantitatively characterizing dust composition and formation in diverse astrophysical settings., Comment: 19 pages, 10 figures, paper published in Nature Astronomy

Kin selection theory has proven to be a popular and widely accepted account of how altruistic behaviour can evolve under natural selection. Hamilton's rule, first published in 1964, has since been experimentally validated across a range of different species and social behaviours. In contrast to this large body of work in natural populations, however, there has been relatively little study of kin selection \emph{in silico}. In the current work, we offer what is to our knowledge the first demonstration of kin selection emerging naturally within a population of agents undergoing continuous neuroevolution. Specifically, we find that zero-sum transfer of resources from parents to their infant offspring evolves through kin selection in environments where it is hard for offspring to survive alone. In an additional experiment, we show that kin selection in our simulations relies on a combination of kin recognition and population viscosity. We believe that our work may contribute to the understanding of kin selection in minimal evolutionary systems, without explicit notions of genes and fitness maximisation., Comment: 13 pages, 5 figures

In many condensed matter systems, long range order emerges at low temperatures as thermal fluctuations subside. In the presence of competing interactions or quenched disorder, however, some systems can show unusual configurations that become more disordered at low temperature, a rare phenomenon known as "inverse melting". Here, we discover an inverse melting of the polar order in a ferroelectric oxide with quenched chemical disorder (BaTi1-xZrxO3) through direct atomic-scale visualization using in situ scanning transmission electron microscopy. In contrast to the clean BaTiO3 parent system in which long range order tracks lower temperatures, we observe in the doped system BaTi1-xZrxO3 that thermally driven fluctuations at high temperature give way to a more ordered state and then to a re-entrant disordered configuration at even lower temperature. Such an inverse melting of the polar order is likely linked to the random field generated by Zr dopants, which modulates the energy landscape arising from the competition between thermal fluctuations and random field pinning potential. These visualizations highlight a rich landscape of order and disorder in materials with quenched disorder, which may be key to understanding their advanced functionalities.

A central question in multilingual language modeling is whether large language models (LLMs) develop a universal concept representation, disentangled from specific languages. In this paper, we address this question by analyzing latent representations (latents) during a word translation task in transformer-based LLMs. We strategically extract latents from a source translation prompt and insert them into the forward pass on a target translation prompt. By doing so, we find that the output language is encoded in the latent at an earlier layer than the concept to be translated. Building on this insight, we conduct two key experiments. First, we demonstrate that we can change the concept without changing the language and vice versa through activation patching alone. Second, we show that patching with the mean over latents across different languages does not impair and instead improves the models' performance in translating the concept. Our results provide evidence for the existence of language-agnostic concept representations within the investigated models., Comment: 12 pages, 10 figures, previous version published under the title "How Do Llamas Process Multilingual Text? A Latent Exploration through Activation Patching" at the ICML 2024 mechanistic interpretability workshop at https://openreview.net/forum?id=0ku2hIm4BS

Program synthesis methods aim to automatically generate programs restricted to a language that can explain a given specification of input-output pairs. While purely symbolic approaches suffer from a combinatorial search space, recent methods leverage neural networks to learn distributions over program structures to narrow this search space significantly, enabling more efficient search. However, for challenging problems, it remains difficult to train models to perform program synthesis in one shot, making test-time search essential. Most neural methods lack structured search mechanisms during inference, relying instead on stochastic sampling or gradient updates, which can be inefficient. In this work, we propose the Latent Program Network (LPN), a general algorithm for program induction that learns a distribution over latent programs in a continuous space, enabling efficient search and test-time adaptation. We explore how to train these networks to optimize for test-time computation and demonstrate the use of gradient-based search both during training and at test time. We evaluate LPN on ARC-AGI, a program synthesis benchmark that evaluates performance by generalizing programs to new inputs rather than explaining the underlying specification. We show that LPN can generalize beyond its training distribution and adapt to unseen tasks by utilizing test-time computation, outperforming algorithms without test-time adaptation mechanisms., Comment: Code available at https://github.com/clement-bonnet/lpn

In this paper we investigate the tractability of robust Markov Decision Processes (RMDPs) under various structural assumptions on the uncertainty set. Surprisingly, we show that in all generality (i.e. without any assumption on the instantaneous rewards), s-rectangular and sa-rectangular uncertainty sets are the only models of uncertainty that are tractable. Our analysis also shows that existing non-rectangular models, including r-rectangular uncertainty and new generalizations, are only weakly tractable in that they require an additional structural assumption that the instantaneous rewards do not depend on the next state, and in this case they are equivalent to rectangular models, which severely undermines their significance and usefulness. Interestingly, our proof techniques rely on identifying a novel simultaneous solvability property, which we show is at the heart of several important properties of RMDPs, including the existence of stationary optimal policies and dynamic programming-based formulations. The simultaneous solvability property enables a unified approach to studying the tractability of all existing models of uncertainty, rectangular and non-rectangular alike.

We carried out a project involving the systematic analysis of microlensing data from the Korea Microlensing Telescope Network survey. The aim of this project is to identify lensing events with complex anomaly features that are difficult to explain using standard binary-lens or binary-source models. Our investigation reveals that the light curves of microlensing events KMT-2021-BLG-0284, KMT-2022-BLG-2480, and KMT-2024-BLG-0412 display highly complex patterns with three or more anomaly features. These features cannot be adequately explained by a binary-lens (2L1S) model alone. However, the 2L1S model can effectively describe certain segments of the light curve. By incorporating an additional source into the modeling, we identified a comprehensive model that accounts for all the observed anomaly features. Bayesian analysis, based on constraints provided by lensing observables, indicates that the lenses of KMT-2021-BLG-0284 and KMT-2024-BLG-0412 are binary systems composed of M dwarfs. For KMT-2022-BLG-2480, the primary lens is an early K-type main-sequence star with an M dwarf companion. The lenses of KMT-2021-BLG-0284 and KMT-2024-BLG-0412 are likely located in the bulge, whereas the lens of KMT-2022-BLG-2480 is more likely situated in the disk. In all events, the binary stars of the sources have similar magnitudes due to a detection bias favoring binary source events with a relatively bright secondary source star, which increases detection efficiency., Comment: 9 pages, 9 figures

Positive spanning sets (PSSs) are families of vectors that span a given linear space through non-negative linear combinations. Despite certain classes of PSSs being well understood, a complete characterization of PSSs remains elusive. In this paper, we explore a relatively understudied relationship between positive spanning sets and strongly edge-connected digraphs, in that the former can be viewed as a generalization of the latter. We leverage this connection to define a decomposition structure for positive spanning sets inspired by the ear decomposition from digraph theory.

Bayesian Neural Networks (BNNs) provide superior estimates of uncertainty by generating an ensemble of predictive distributions. However, inference via ensembling is resource-intensive, requiring additional entropy sources to generate stochasticity which increases resource consumption. We introduce Bayes2IMC, an in-memory computing (IMC) architecture designed for binary Bayesian neural networks that leverage nanoscale device stochasticity to generate desired distributions. Our novel approach utilizes Phase-Change Memory (PCM) to harness inherent noise characteristics, enabling the creation of a binary neural network. This design eliminates the necessity for a pre-neuron Analog-to-Digital Converter (ADC), significantly improving power and area efficiency. We also develop a hardware-software co-optimized correction method applied solely on the logits in the final layer to reduce device-induced accuracy variations across deployments on hardware. Additionally, we devise a simple compensation technique that ensures no drop in classification accuracy despite conductance drift of PCM. We validate the effectiveness of our approach on the CIFAR-10 dataset with a VGGBinaryConnect model, achieving accuracy metrics comparable to ideal software implementations as well as results reported in the literature using other technologies. Finally, we present a complete core architecture and compare its projected power, performance, and area efficiency against an equivalent SRAM baseline, showing a $3.8$ to $9.6 \times$ improvement in total efficiency (in GOPS/W/mm$^2$) and a $2.2 $ to $5.6 \times$ improvement in power efficiency (in GOPS/W). In addition, the projected hardware performance of Bayes2IMC surpasses that of most of the BNN architectures based on memristive devices reported in the literature, and achieves up to $20\%$ higher power efficiency compared to the state-of-the-art.

For many years, image over-segmentation into superpixels has been essential to computer vision pipelines, by creating homogeneous and identifiable regions of similar sizes. Such constrained segmentation problem would require a clear definition and specific evaluation criteria. However, the validation framework for superpixel methods, typically viewed as standard object segmentation, has rarely been thoroughly studied. In this work, we first take a step back to show that superpixel segmentation is fundamentally an ill-posed problem, due to the implicit regularity constraint on the shape and size of superpixels. We also demonstrate through a novel comprehensive study that the literature suffers from only evaluating certain aspects, sometimes incorrectly and with inappropriate metrics. Concurrently, recent deep learning-based superpixel methods mainly focus on the object segmentation task at the expense of regularity. In this ill-posed context, we show that we can achieve competitive results using a recent architecture like the Segment Anything Model (SAM), without dedicated training for the superpixel segmentation task. This leads to rethinking superpixel segmentation and the necessary properties depending on the targeted downstream task.

In this work, we probe the sensitivity of hole-spin properties to hole occupation number in a planar silicon double-quantum dot device fabricated on a 300 mm integrated platform. Using DC transport measurements, we investigate the g-tensor and spin-relaxation induced leakage current within the Pauli spin-blockade regime as a function of magnetic-field orientation at three different hole occupation numbers. We find the g-tensor and spin-leakage current to be highly anisotropic due to light-hole/heavy-hole mixing and spin-orbit mixing, but discover the anisotropies to be relatively insensitive to the dot hole number. Furthermore, we extract the dominant inter-dot spin-orbit coupling mechanism as surface Dresselhaus, with an in-plane orientation parallel to transport and magnitude $\boldsymbol{t_{SO}}$ $\approx$ 300 neV. Finally, we observe a strong correlation between the g-factor difference ($\delta$$\boldsymbol{g}$) between each dot and the spin-leakage current anisotropy, as a result of $\delta$$\boldsymbol{g}$ providing an additional spin-relaxation pathway, and should be considered. ]Our findings indicate that hole-spin devices are not as sensitive to precise operating conditions as anticipated. This has important implications optimizing spin control and readout based on magnetic-field direction, together with tuning large arrays of QDs as spin-qubits.

In recent years,Large Language Models (LLMs) have significantly improved in generating high-quality code, enabling their integration into developers' Integrated Development Environments (IDEs) as code assistants. These assistants, such as GitHub Copilot, deliver real-time code suggestions and can greatly enhance developers' productivity. However, the environmental impact of these tools, in particular their energy consumption, remains a key concern. This paper investigates the energy consumption of LLM-based code assistants by simulating developer interactions with GitHub Copilot and analyzing various configuration factors. We collected a dataset of development traces from 20 developers and conducted extensive software project development simulations to measure energy usage under different scenarios. Our findings reveal that the energy consumption and performance of code assistants are influenced by various factors, such as the number of concurrent developers, model size, quantization methods, and the use of streaming. Notably, a substantial portion of generation requests made by GitHub Copilot is either canceled or rejected by developers, indicating a potential area for reducing wasted computations. Based on these findings, we share actionable insights into optimizing configurations for different use cases, demonstrating that careful adjustments can lead to significant energy savings., Comment: Submitted to JSS

Recent discoveries of gravitational wave sources have advanced our knowledge about the formation of compact object binaries. At present, many questions about the stellar origins of binary neutron stars remain open. We explore the evolution of binary neutron star progenitors with the population synthesis code COSMIC. We identify three dominant evolutionary tracks to form neutron star binaries that merge within the age of the Universe: a scenario that includes a common envelope phase between the first neutron star and its companion, a scenario with almost equal-mass progenitors that evolve quasi-simultaneously and which features a double-core common envelope, and a scenario involving the accretion-induced collapse of an oxygen-neon white dwarf into a neutron star. We show that the distribution of time delays between stellar formation and binary neutron star merger at a given progenitor metallicity does not follow a power-law, but instead features a complex structure that reflects the progenitor properties and the relative contribution of each evolutionary track. We also explore the evolution of the merger rate density with redshift and show that the scenario involving the accretion-induced collapse could be dominant at high redshifts. These results can have important implications for the study of the chemical enrichment of galaxies in r-process elements produced in kilonovae; and of short gamma-ray bursts offsets in their host galaxies., Comment: 14 pages, 10 figures. Accepted for publication in A&A

We study quantum algorithms for verifying properties of the output probability distribution of a classical or quantum circuit, given access to the source code that generates the distribution. We consider the basic task of uniformity testing, which is to decide if the output distribution is uniform on $[d]$ or $\epsilon$-far from uniform in total variation distance. More generally, we consider identity testing, which is the task of deciding if the output distribution equals a known hypothesis distribution, or is $\epsilon$-far from it. For both problems, the previous best known upper bound was $O(\min\{d^{1/3}/\epsilon^{2},d^{1/2}/\epsilon\})$. Here we improve the upper bound to $O(\min\{d^{1/3}/\epsilon^{4/3}, d^{1/2}/\epsilon\})$, which we conjecture is optimal., Comment: 21 pages

A long-lived civilization will inevitably have to migrate towards a nearby star as its home star runs out of nuclear fuel. One way to achieve such a migration is by transforming its star into a stellar engine, and to control its motion in the galaxy. We first provide a brief overview of stellar engines and conclude that looking for technosignatures of stellar engines has taken two roads: on the observational side, hypervelocity stars have been the target of such searches, but without good candidates. On the theoretical side, stellar engine concepts have been proposed but are poorly linked to observable technosignatures. Since about half the stars in our galaxy are in binary systems where life might develop too, we introduce a model of a binary stellar engine. We propose mechanisms for acceleration, deceleration, steering in the orbital plane and outside of the orbital plane. We apply the model to candidate systems, spider pulsars, which are binary stars composed of one millisecond pulsar and a very low-mass companion star that is heavily irradiated by the pulsar wind. We discuss potential signatures of acceleration, deceleration, steering, as well as maneuvers such as gravitational assists or captures. Keywords: Pulsars: Spiders, Technosignatures, Stellar engine, Interstellar travel, Comment: 11 pages, 5 figures

The automatic generation of hints by Large Language Models (LLMs) within Intelligent Tutoring Systems (ITSs) has shown potential to enhance student learning. However, generating pedagogically sound hints that address student misconceptions and adhere to specific educational objectives remains challenging. This work explores using LLMs (GPT-4o and Llama-3-8B-instruct) as teachers to generate effective hints for students simulated through LLMs (GPT-3.5-turbo, Llama-3-8B-Instruct, or Mistral-7B-instruct-v0.3) tackling math exercises designed for human high-school students, and designed using cognitive science principles. We present here the study of several dimensions: 1) identifying error patterns made by simulated students on secondary-level math exercises; 2) developing various prompts for GPT-4o as a teacher and evaluating their effectiveness in generating hints that enable simulated students to self-correct; and 3) testing the best-performing prompts, based on their ability to produce relevant hints and facilitate error correction, with Llama-3-8B-Instruct as the teacher, allowing for a performance comparison with GPT-4o. The results show that model errors increase with higher temperature settings. Notably, when hints are generated by GPT-4o, the most effective prompts include prompts tailored to specific errors as well as prompts providing general hints based on common mathematical errors. Interestingly, Llama-3-8B-Instruct as a teacher showed better overall performance than GPT-4o. Also the problem-solving and response revision capabilities of the LLMs as students, particularly GPT-3.5-turbo, improved significantly after receiving hints, especially at lower temperature settings. However, models like Mistral-7B-Instruct demonstrated a decline in performance as the temperature increased., Comment: Accepted at NeurIPS 2024 Workshop on Large Foundation Models for Educational Assessment (FM-Assess)

Our understanding of the process of terrestrial planet formation has grown markedly over the past 20 years, yet key questions remain. This review begins by first addressing the critical, earliest stage of dust coagulation and concentration. While classic studies revealed how objects that grow to $\sim$meter sizes are rapidly removed from protoplanetary disks via orbital decay (seemingly precluding growth to larger sizes), this chapter addresses how this is resolved in contemporary, streaming instability models that favor rapid planetesimal formation via gravitational collapse of solids in over-dense regions. Once formed, planetesimals grow into Mars-Earth-sized planetary embryos by a combination of pebble- and planetesimal accretion within the lifetime of the nebular disk. After the disk dissipates, these embryos typically experience a series of late giant impacts en route to attaining their final architectures. This review also highlights three different inner Solar System formation models that can match a number of empirical constraints, and also reviews ways that one or more might be ruled out in favor of another in the near future. These include (1) the Grand Tack, (2) the Early Instability and (3) Planet Formation from Rings. Additionally, this chapter discusses formation models for the closest known analogs to our own terrestrial planets: super-Earths and terrestrial exoplanets in systems also hosting gas giants. Finally, this review lays out a chain of events that may explain why the Solar System looks different than more than 99% of exoplanet systems., Comment: To be published in: Handbook of Exoplanets, 2nd Edition, Hans Deeg and Juan Antonio Belmonte (Eds. in Chief), Springer International Publishing AG, part of Springer Nature. 75 pages, 9 figures. This is an update of arXiv:1803.08830

I consider treatment-effect estimation with a two-periods panel, using a first-difference regression of the outcome evolution $\Delta Y_g$ on the treatment evolution $\Delta D_g$. To justify this regression, one may assume that $\Delta D_g$ is as good as randomly assigned, namely uncorrelated to the residual of the first-differenced model and to the treatment's effect. This note shows that if one posits a causal model in levels between the treatment and the outcome, then the residual of the first-differenced model is a function of $D_{g,1}$, so $\Delta D_g$ uncorrelated to that residual essentially implies that $\Delta D_g$ is uncorrelated to $D_{g,1}$. This is a strong, testable condition. If $\Delta D_g$ is correlated to $D_{g,1}$, assuming that $\Delta D_g$ is uncorrelated to the treatment effect and to the remaining terms of the residual may not be sufficient to have that the first-difference regression identifies a convex combination of treatment effects. I use these results to revisit Acemoglu et al (2016).

The discovery of seven ~Earth-mass planets, orbiting the 0.09 solar mass M-Dwarf TRAPPIST-1 captivated the public and sparked a proliferation of investigations into the system's origins. Among other properties, the resonant architecture of the planets has been interpreted to imply that orbital migration played a dominant role in the system's early formation. If correct, this hypothesis could imply that all of the seven worlds formed far from the star, and might harbor enhanced inventories of volatile elements. However, multiple factors also contradict this interpretation. In particular, the planets' apparent rocky compositions and non-hierarchical mass distribution might evidence them having formed closer to their current orbital locations. In this paper, we investigate the latter possibility with over 600 accretion simulations that model the effects of collisional fragmentation. In addition to producing multiple TRAPPIST-like configurations, we experiment with a number of different models for tracking the evolution of the planets' volatile contents and bulk iron-to-silicate ratios. We conclude that a trend in bulk iron contents is the more likely explanation for the observed radial trend of decreasing uncompressed densities in the real system. Given the degree of radial mixing that occurs in our simulations, in most cases we find that all seven planets finish with similar volatile contents. Another confounding quality of the TRAPPIST-1 system is the fact that the innermost planets are not in first-order resonances with one-another. By applying a tidal migration model to our most promising accretion model results, we demonstrate cases where higher-order resonances are populated., Comment: 22 pages, 9 figures, 3 tables, accepted for publication in AJ

Semiconducting transition metal dichalcogenides of group VI are well-known for their prominent excitonic effects and the transition from an indirect to a direct band gap when reduced to monolayers. While considerable efforts have elucidated the Rydberg series of excitons in monolayers, understanding their properties in multilayers remains incomplete. In these structures, despite an indirect band gap, momentum-direct excitons largely shape the optical response. In this work, we combine magneto-reflectance experiments with theoretical modeling based on the $\mathbf {k\cdot p}$ approach to investigate the origin of excitonic resonances in WSe$_2$ bi-, tri-, and quadlayers. For all investigated thicknesses, we observe a series of excitonic resonances in the reflectance spectra, initiated by a ground state with an amplitude comparable to the ground state of the 1$s$ exciton in the monolayer. Higher energy states exhibit a decrease in intensity with increasing energy, as expected for the excited states of the Rydberg series, although a significant increase in the diamagnetic shift is missing in tri- and quadlayers. By comparing the experimental observations with theoretical predictions, we discover that the excitonic resonances observed in trilayers originate from two Rydberg series, while quadlayers exhibit four such series, and bilayers host a single Rydberg series similar to that found in monolayers., Comment: 31 pages, 16 figures, published in Phys. Rev. B

A vertex coloring $\varphi$ of a graph $G$ is $p$-centered if for every connected subgraph $H$ of $G$, either $\varphi$ uses more than $p$ colors on $H$, or there is a color that appears exactly once on $H$. We prove that for every fixed positive integer $t$, every $K_t$-minor-free graph admits a $p$-centered coloring using $\mathcal{O}(p^{t-1})$ colors., Comment: 23 pages, 10 figures

Statistical learning in high-dimensional spaces is challenging without a strong underlying data structure. Recent advances with foundational models suggest that text and image data contain such hidden structures, which help mitigate the curse of dimensionality. Inspired by results from nonparametric statistics, we hypothesize that this phenomenon can be partially explained in terms of decomposition of complex tasks into simpler subtasks. In this paper, we present a controlled experimental framework to test whether neural networks can indeed exploit such ``hidden factorial structures.'' We find that they do leverage these latent patterns to learn discrete distributions more efficiently, and derive scaling laws linking model sizes, hidden factorizations, and accuracy. We also study the interplay between our structural assumptions and the models' capacity for generalization.

In this paper, we introduce WaKA (Wasserstein K-nearest neighbors Attribution), a novel attribution method that leverages principles from the LiRA (Likelihood Ratio Attack) framework and applies them to \( k \)-nearest neighbors classifiers (\( k \)-NN). WaKA efficiently measures the contribution of individual data points to the model's loss distribution, analyzing every possible \( k \)-NN that can be constructed using the training set, without requiring sampling or shadow model training. WaKA can be used \emph{a posteriori} as a membership inference attack (MIA) to assess privacy risks, and \emph{a priori} for data minimization and privacy influence measurement. Thus, WaKA can be seen as bridging the gap between data attribution and membership inference attack (MIA) literature by distinguishing between the value of a data point and its privacy risk. For instance, we show that self-attribution values are more strongly correlated with the attack success rate than the contribution of a point to model generalization. WaKA's different usages were also evaluated across diverse real-world datasets, demonstrating performance very close to LiRA when used as an MIA on \( k \)-NN classifiers, but with greater computational efficiency.

This article establishes a method to answer a finite set of linear queries on a given dataset while ensuring differential privacy. To achieve this, we formulate the corresponding task as a saddle-point problem, i.e. an optimization problem whose solution corresponds to a distribution minimizing the difference between answers to the linear queries based on the true distribution and answers from a differentially private distribution. Against this background, we establish two new algorithms for corresponding differentially private data release: the first is based on the differentially private Frank-Wolfe method, the second combines randomized smoothing with stochastic convex optimization techniques for a solution to the saddle-point problem. While previous works assess the accuracy of differentially private algorithms with reference to the empirical data distribution, a key contribution of our work is a more natural evaluation of the proposed algorithms' accuracy with reference to the true data-generating distribution.

A class of acyclic digraphs $\mathscr{C}$ is linearly unavoidable if there exists a constant $c$ such that every digraph $D\in \mathscr{C}$ is contained in all tournaments of order $c\cdot |V(D)|$. The class of all acyclic digraphs is not linearly avoidable, and Fox, He, and Widgerson recently showed that this is not even the case for acyclic digraphs with bounded maximum degree. On the positive side, Thomason and H\"aggkvist proved that the class of oriented trees is linearly unavoidable. In this work, we generalize this result to acyclic digraphs obtained from an oriented tree by adding at most $k$ vertices, and $k$-blow-ups of oriented trees, for every fixed integer $k$. More precisely, we show that if $D$ is obtained from an oriented tree $F$ of order $n$ by adding $k$ universal vertices, then $D$ is contained in every tournament of order $2\cdot 3^{(k+1)(2k+1)} \cdot n$; and if $D$ is obtained from $F$ by replacing each vertex $u$ by an independent set $X_u$ of size $k$ and every arc $uv$ by all possible arcs from $X_u$ to $X_v$, then $D$ is contained in every tournament of order $2^{10+18k}k \cdot n$.

We present measurements of the microlensing optical depth and event rate toward the Galactic bulge using the dataset from the 2006--2014 MOA-II survey, which covers 22 bulge fields spanning ~42 deg^2 between -5 deg < l < 10 deg and -7 deg < b < -1 deg. In the central region with |l|<5 deg, we estimate an optical depth of {\tau} = [1.75+-0.04]*10^-6exp[(0.34+-0.02)(3 deg-|b|)] and an event rate of {\Gamma} = [16.08+-0.28]*10^-6exp[(0.44+-0.02)(3 deg-|b|)] star^-1 year^-1 using a sample consisting of 3525 microlensing events, with Einstein radius crossing times of tE < 760 days and source star magnitude of IsWe confirm our results are consistent with the latest measurements from OGLE-IV 8 year dataset (Mr\'oz et al. 2019). We find our result is inconsistent with a prediction based on Galactic models, especially in the central region with |b|<3 deg. These results can be used to improve the Galactic bulge model, and more central regions can be further elucidated by future microlensing experiments, such as The PRime-focus Infrared Microlensing Experiment (PRIME) and Nancy Grace Roman Space Telescope.

Earth remains the only known example of a planet with technology, and future projections of Earth's trajectory provide a basis and motivation for approaching the search for extraterrestrial technospheres. Conventional approaches toward projecting Earth's technosphere include applications of the Kardashev scale, which suggest the possibility that energy-intensive civilizations may expand to harness the entire energy output available to their planet, host star, or even the entire galaxy. In this study, we argue that the Kardashev scale is better understood as a "luminosity limit" that describes the maximum capacity for a civilization to harvest luminous stellar energy across a given spatial domain, and we note that thermodynamic efficiency will always keep a luminosity-limited technosphere from actually reaching this theoretical limit. We suggest the possibility that an advanced technosphere might evolve beyond this luminosity limit to draw its energy directly from harvesting stellar mass, and we also discuss possible trajectories that could exist between Earth today and such hypothetical "stellivores." We develop a framework to describe trajectories for long-lived technospheres that optimize their growth strategies between exploration and exploitation, unlike Earth today. We note that analyses of compact accreting stars could provide ways to test the stellivore hypothesis, and we more broadly suggest an expansion of technosignature search strategies beyond those that reside exactly at the luminosity limit., Comment: Submitted to Acta Astronautica

Uncloneable encryption is a cryptographic primitive which encrypts a classical message into a quantum ciphertext, such that two quantum adversaries are limited in their capacity of being able to simultaneously decrypt, given the key and quantum side-information produced from the ciphertext. Since its initial proposal and scheme in the random oracle model by Broadbent and Lord [TQC 2020], uncloneable encryption has developed into an important primitive at the foundation of quantum uncloneability for cryptographic primitives. Despite sustained efforts, however, the question of unconditional uncloneable encryption (and in particular of the simplest case, called an uncloneable bit) has remained elusive. Here, we propose a candidate for the unconditional uncloneable bit problem, and provide strong evidence that the adversary's success probability in the related security game converges quadratically as ${1}/{2}+{1}/{(2\sqrt{K})}$, where $K$ represents the number of keys and ${1}/{2}$ is trivially achievable. We prove this bound's validity for $K$ ranging from $2$ to $7$ and demonstrate the validity up to $K = 17$ using computations based on the NPA hierarchy. We furthemore provide compelling heuristic evidence towards the general case. In addition, we prove an asymptotic upper bound of ${5}/{8}$ and give a numerical upper bound of $\sim 0.5980$, which to our knowledge is the best-known value in the unconditional model.

Air pollution is a pressing environmental risk to public health, particularly in cities where population density and pollution levels are high. Traditional methods for exposure analysis often rely on census data, but recent studies highlight the impact of daily mobility on individuals' exposure. Here, we develop a methodology to determine unprecedented pedestrian exposure estimates at the city scale by combining sidewalk pedestrian flows with high-resolution (25 m x 25 m) NO2 data from bias-corrected predictions of the air quality system CALIOPE-Urban. Applied to Barcelona (Spain) for the year 2019, we show that pedestrian flow and NO2 levels exhibit negligible temporal correlation. While short-term (hourly) exposure is driven by pedestrian mobility, long-term (monthly) exposure is dominated by NO2 patterns. We identify strong spatial gradients of exposure, highlighting the importance for high-resolution solutions at the sidewalks scale. Finally, we determine that exposure mitigation strategies should consider different citizen subgroups based on their mobility and preferred routes, as significant differences were found between residential and pedestrian exposure. Our results provide exposure indicators designed for city planners and policymakers, helping to prioritize mitigation measures where and when they are most needed.

Reinforcement learning (RL) is a promising approach for aligning large language models (LLMs) knowledge with sequential decision-making tasks. However, few studies have thoroughly investigated the impact on LLM agents capabilities of fine-tuning them with RL in a specific environment. In this paper, we propose a novel framework to analyze the sensitivity of LLMs to prompt formulations following RL training in a textual environment. Our findings reveal that the performance of LLMs degrades when faced with prompt formulations different from those used during the RL training phase. Besides, we analyze the source of this sensitivity by examining the model's internal representations and salient tokens. Finally, we propose to use a contrastive loss to mitigate this sensitivity and improve the robustness and generalization capabilities of LLMs.

The spatial distribution of spurious dots in SiMOS single-electron transistors (SETs), fabricated on an industrial 300 mm process line, has been statistically analyzed. To have a deeper understanding of the origin of these spurious dots, we analyzed SETs with three different oxide thicknesses: 8 nm, 12 nm and 20 nm. By combining spurious dot triangulation cryo-measurement with simulations of strain, gate bias, and location of the electron wave function, we demonstrate that most spurious dots are formed through the combined effects of strain and gate bias, leading to variations in the conduction band energy. Despite the similar thermal expansion coefficients of polycrystalline silicon gates and single-crystalline silicon substrates, strain remains a crucial factor in spurious dots formation. This learning can be use to optimize the device design and the oxide thickness, to reduce the density of spurious dot while keeping quantum dot tunability., Comment: 10 pages, 9 figures

The Zirconium (Z = 40) isotopic chain has attracted interest for more than four decades. The abrupt lowering of the energy of the first $2^+$ state and the increase in the transition strength B(E2; $2_1^\rightarrow 0_1^+$ going from $^{98}$Zr to $^{100}$Zr has been the first example of "quantum phase transition" in nuclear shapes, which has few equivalents in the nuclear chart. Although a multitude of experiments have been performed to measure nuclear properties related to nuclear shapes and collectivity in the region, none of the measured lifetimes were obtained using the Recoil Distance Doppler Shift method in the $\gamma\gamma$-coincidence mode where a gate on the direct feeding transition of the state of interest allows a strict control of systematical errors. This work reports the results of lifetime measurements for the first yrast excited states in $^{98-104}$Zr carried out to extract reduced transition probabilities. The new lifetime values in $\gamma\gamma$-coincidence and $\gamma$-single mode are compared with the results of former experiments. Recent predictions of the Interacting Boson Model with Configuration Mixing, the Symmetry Conserving Configuration Mixing model based on the Hartree-Fock-Bogoliubov approach and the Monte Carlo Shell Model are presented and compared with the experimental data.

Fabrication of quantum processors in advanced 300 mm wafer-scale complementary metal-oxide-semiconductor (CMOS) foundries provides a unique scaling pathway towards commercially viable quantum computing with potentially millions of qubits on a single chip. Here, we show precise qubit operation of a silicon two-qubit device made in a 300 mm semiconductor processing line. The key metrics including single- and two-qubit control fidelities exceed 99% and state preparation and measurement fidelity exceeds 99.9%, as evidenced by gate set tomography (GST). We report coherence and lifetimes up to $T_\mathrm{2}^{\mathrm{*}} = 30.4$ $\mu$s, $T_\mathrm{2}^{\mathrm{Hahn}} = 803$ $\mu$s, and $T_1 = 6.3$ s. Crucially, the dominant operational errors originate from residual nuclear spin carrying isotopes, solvable with further isotopic purification, rather than charge noise arising from the dielectric environment. Our results answer the longstanding question whether the favourable properties including high-fidelity operation and long coherence times can be preserved when transitioning from a tailored academic to an industrial semiconductor fabrication technology., Comment: 10 pages, 4 figures, 4 extended data figures

Privacy and regulatory constraints make data generation vital to advancing machine learning without relying on real-world datasets. A leading approach for tabular data generation is the Forest Flow (FF) method, which combines Flow Matching with XGBoost. Despite its good performance, FF is slow and makes errors when treating categorical variables as one-hot continuous features. It is also highly sensitive to small changes in the initial conditions of the ordinary differential equation (ODE). To overcome these limitations, we develop Heterogeneous Sequential Feature Forest Flow (HS3F). Our method generates data sequentially (feature-by-feature), reducing the dependency on noisy initial conditions through the additional information from previously generated features. Furthermore, it generates categorical variables using multinomial sampling (from an XGBoost classifier) instead of flow matching, improving generation speed. We also use a Runge-Kutta 4th order (Rg4) ODE solver for improved performance over the Euler solver used in FF. Our experiments with 25 datasets reveal that HS3F produces higher quality and more diverse synthetic data than FF, especially for categorical variables. It also generates data 21-27 times faster for datasets with $\geq20%$ categorical variables. HS3F further demonstrates enhanced robustness to affine transformation in flow ODE initial conditions compared to FF. This study not only validates the HS3F but also unveils promising new strategies to advance generative models.

We report the existence of hydrostatic equilibrium states for a composite body made of two rigidly rotating, homogeneous layers bounded by spheroidal surfaces, where the core has a prolate shape. These new configurations require an oblate envelope that spins faster than the core. No solution with a prolate envelope is found. For some parameters, the prolate core can even be at rest. Numerical experiments based on the self-consistent field method support this result in the case of heterogeneous layers with polytropic equations of state. The possible cancellation of the first gravitational moment, $J_2$, is discussed., Comment: Accepted for publication in A&A. 7 pages, 4 figures, 4 tables

Labor plays a major, albeit largely unrecognized role in the development of artificial intelligence. Machine learning algorithms are predicated on data-intensive processes that rely on humans to execute repetitive and difficult-to-automate, but no less essential, tasks such as labeling images, sorting items in lists, recording voice samples, and transcribing audio files. Online platforms and networks of subcontractors recruit data workers to execute such tasks in the shadow of AI production, often in lower-income countries with long-standing traditions of informality and lessregulated labor markets. This study unveils the resulting complexities by comparing the working conditions and the profiles of data workers in Venezuela, Brazil, Madagascar, and as an example of a richer country, France. By leveraging original data collected over the years 2018-2023 via a mixed-method design, we highlight how the cross-country supply chains that link data workers to core AI production sites are reminiscent of colonial relationships, maintain historical economic dependencies, and generate inequalities that compound with those inherited from the past. The results also point to the importance of less-researched, non-English speaking countries to understand key features of the production of AI solutions at planetary scale., Comment: Jack Qiu, Shinjoung Yeo, Richard Maxwell. The Handbook of Digital Labor, Wiley Blackwell, In press, ISBN10: 1119981808

The diagnosis of diabetic retinopathy, which relies on fundus images, faces challenges in achieving transparency and interpretability when using a global classification approach. However, segmentation-based databases are significantly more expensive to acquire and combining them is often problematic. This paper introduces a novel method, termed adversarial style conversion, to address the lack of standardization in annotation styles across diverse databases. By training a single architecture on combined databases, the model spontaneously modifies its segmentation style depending on the input, demonstrating the ability to convert among different labeling styles. The proposed methodology adds a linear probe to detect dataset origin based on encoder features and employs adversarial attacks to condition the model's segmentation style. Results indicate significant qualitative and quantitative through dataset combination, offering avenues for improved model generalization, uncertainty estimation and continuous interpolation between annotation styles. Our approach enables training a segmentation model with diverse databases while controlling and leveraging annotation styles for improved retinopathy diagnosis., Comment: preprint

The past years have seen Large Language Models (LLMs) strive not only as generative models but also as agents solving textual sequential decision-making tasks. When facing complex environments where their zero-shot abilities are insufficient, recent work showed online Reinforcement Learning (RL) could be used for the LLM agent to discover and learn efficient strategies interactively. However, most prior work sticks to on-policy algorithms, which greatly reduces the scope of methods such agents could use for both exploration and exploitation, such as experience replay and hindsight relabeling. Yet, such methods may be key for LLM learning agents, and in particular when designing autonomous intrinsically motivated agents sampling and pursuing their own goals (i.e. autotelic agents). This paper presents and studies an adaptation of Soft Actor-Critic and hindsight relabeling to LLM agents. Our method not only paves the path towards autotelic LLM agents that learn online but can also outperform on-policy methods in more classic multi-goal RL environments.

We study numerically the nonintegrable dynamics of coherent, solitonic, nonlinear waves, in a spatially nonlocal nonlinear Schrodinger equation relevant to realistic modelling of optical systems: the Schrodinger-Helmholtz equation. We observe a single oscillating, coherent solitary wave emerging from a variety of initial conditions. Using the direct scattering transform of the (integrable) cubic nonlinear Schrodinger equation, we find that this structure is a bound state, comprising of a primary and secondary soliton whose amplitudes oscillate out of phase. We interpret this as the solitons periodically exchanging mass. We observe this bound state self-organising from a state of incoherent turbulence, and from solitonic structures launched into the system. When a single (primary) solitonic structure is launched, a resonance process between it and waves in the system generates the secondary soliton, resulting in the bound state. Further, when two solitons are initially launched, we show that they can merge if their phases are synchronised when they collide. When the system is launched from a turbulent state comprised of many initial solitons, we propose that the bound state formation is preceded a sequence of binary collisions, in which the mass is transferred on average from the weak soliton to the strong one, with occasional soliton mergers. Both processes lead to increasingly stronger and fewer dominant solitons. The final state -- a solitary double-solitonic bound state surrounded by weakly nonlinear waves -- is robust and ubiquitous. We propose that for nonlocal media, it is a more typical statistical attractor than a single-soliton attractor suggested in previous literature., Comment: 13 figures

Temporal logics are a powerful tool to specify properties of computational systems. For concurrent programs, Higher Dimensional Automata (HDA) are a very expressive model of non-interleaving concurrency. HDA recognize languages of partially ordered multisets, or pomsets. Recent work has shown that Monadic Second Order (MSO) logic is as expressive as HDA for pomset languages. In this paper, we investigate the class of pomset languages that are definable in First Order (FO) logic. As expected, this is a strict subclass of MSO-definable languages. In the case of words, Kamp's theorem states that FO is as expressive as Linear Temporal Logic (LTL). Our aim is to prove a variant of Kamp's theorem for pomset languages. Thus, we define a temporal logic called Sparse Pomset Temporal Logic (SPTL), and show that it is equivalent to FO, when there is no autoconcurrency., Comment: 18 pages + references + 2 pages appendix, submitted to FoSSaCS 2025

The Leaf Area Index (LAI) is a critical parameter to understand ecosystem health and vegetation dynamics. In this paper, we propose a novel method for pixel-wise LAI prediction by leveraging the complementary information from Sentinel 1 radar data and Sentinel 2 multi-spectral data at multiple timestamps. Our approach uses a deep neural network based on multiple U-nets tailored specifically to this task. To handle the complexity of the different input modalities, it is comprised of several modules that are pre-trained separately to represent all input data in a common latent space. Then, we fine-tune them end-to-end with a common decoder that also takes into account seasonality, which we find to play an important role. Our method achieved 0.06 RMSE and 0.93 R2 score on publicly available data. We make our contributions available at https://github.com/valentingol/LeafNothingBehind for future works to further improve on our current progress.

We investigate the theoretical performances of the Partial Least Square (PLS) algorithm in a high dimensional context. We provide upper bounds on the risk in prediction for the statistical linear model when considering the PLS estimator. Our bounds are non-asymptotic and are expressed in terms of the number of observations, the noise level, the properties of the design matrix, and the number of considered PLS components. In particular, we exhibit some scenarios where the variability of the PLS may explode and prove that we can get round of these situations by introducing a Ridge regularization step. These theoretical findings are illustrated by some numerical simulations.

With the large amount of data generated every day, public sentiment is a key factor for various fields, including marketing, politics, and social research. Understanding the public sentiment about different topics can provide valuable insights. However, most traditional approaches for sentiment analysis often depend on supervised learning, which requires a significant amount of labeled data. This makes it both expensive and time-consuming to implement. This project introduces an approach using unsupervised learning techniques, particularly Affinity Propagation (AP) clustering, to analyze sentiment. AP clustering groups text data based on natural patterns, without needing predefined cluster numbers. The paper compares AP with K-means clustering, using TF-IDF Vectorization for text representation and Principal Component Analysis (PCA) for dimensionality reduction. To enhance performance, AP is combined with Agglomerative Hierarchical Clustering. This hybrid method refines clusters further, capturing both global and local sentiment structures more effectively. The effectiveness of these methods is evaluated using the Silhouette Score, Calinski-Harabasz Score, and Davies-Bouldin Index. Results show that AP with Agglomerative Hierarchical Clustering significantly outperforms K-means. This research contributes to Natural Language Processing (NLP) by proposing a scalable and efficient unsupervised learning framework for sentiment analysis, highlighting the significant societal impact of advanced AI techniques in analyzing public sentiment without the need for extensive labeled data.

The development of Electronic Health Records summarization systems has revolutionized patient data management. Previous research advanced this field by adapting Large Language Models for clinical tasks, using diverse datasets to generate general EHR summaries. However, clinicians often require specific, focused summaries for quicker insights. This project builds on prior work by creating a system that generates clinician-preferred, focused summaries, improving EHR summarization for more efficient patient care. The proposed system leverages the Google Flan-T5 model to generate tailored EHR summaries based on clinician-specified topics. The approach involved fine-tuning the Flan-T5 model on an EHR question-answering dataset formatted in the Stanford Question Answering Dataset (SQuAD) style, which is a large-scale reading comprehension dataset with questions and answers. Fine-tuning utilized the Seq2SeqTrainer from the Hugging Face Transformers library with optimized hyperparameters. Key evaluation metrics demonstrated promising results: the system achieved an Exact Match (EM) score of 81.81%. ROUGE (Recall-Oriented Understudy for Gisting Evaluation) metrics showed strong performance, with ROUGE-1 at 96.03%, ROUGE-2 at 86.67%, and ROUGE-L at 96.10%. Additionally, the Bilingual Evaluation Understudy (BLEU) score was 63%, reflecting the model's coherence in generating summaries. By enhancing EHR summarization through LLMs, this project supports digital transformation efforts in healthcare, streamlining workflows, and enabling more personalized patient care.