Search

Evaluating LLM-generated text has become a key challenge, especially in domain-specific contexts like the medical field. This work introduces a novel evaluation methodology for LLM-generated medical explanatory arguments, relying on Proxy Tasks and rankings to closely align results with human evaluation criteria, overcoming the biases typically seen in LLMs used as judges. We demonstrate that the proposed evaluators are robust against adversarial attacks, including the assessment of non-argumentative text. Additionally, the human-crafted arguments needed to train the evaluators are minimized to just one example per Proxy Task. By examining multiple LLM-generated arguments, we establish a methodology for determining whether a Proxy Task is suitable for evaluating LLM-generated medical explanatory arguments, requiring only five examples and two human experts.

Motile liquid crystal (LC) colloids show peculiar behavior due to the high sensitivity to external stimuli driven by the LC elastic and surface effects. However, few studies focus on harnessing the LC phase transitions to propel colloidal inclusions by the nematic-isotropic (NI) interface. We engineer a quasi-2D active system consisting of solid micron-sized light-absorbent platelets immersed in a thermotropic nematic LC. The platelets self-propel in the presence of light while self-inducing a localized NI phase transition. The sample's temperature, light intensity, and confinement determine three different regimes: a 2D large regime where the platelet-isotropic phase bubble is static and the NI interface remains stable; a compact motile-2D regime where the NI interface lies closer to the platelet's contour; and a motile-3D-confinement regime characterized by the emergence of multipolar configurations of the LC. We perform continuum-theory simulations that predict stationary platelet-LC states when confined in 3D. Our study in an intrinsically far-from-equilibrium landscape is crucial for designing simple synthetic systems that contribute to our understanding of harnessing liquid crystals' phase transitions to propel colloidal inclusions and trigger tunable topological reconfigurations leading to photonic responses., Comment: 29 pages main text. 5 figures main text. Supporting Information included at the end

In this paper, we present several path properties, simulations, inferences, and generalizations of the weighted sub-fractional Brownian motion. A primary focus is on the derivation of the covariance function $R_{f,b}(s,t)$ for the weighted sub-fractional Brownian motion, defined as: \begin{equation*} R_{f,b}(s,t) = \frac{1}{1-b} \int_{0}^{s \wedge t} f(r) \left[(s-r)^{b} + (t-r)^{b} - (t+s-2r)^{b}\right] dr, \end{equation*} where $f:\mathbb{R}_{+} \to \mathbb{R}_{+}$ is a measurable function and $b\in [0,1)\cup(1,2]$. This covariance function $R_{f,b}(s,t)$ is used to define the centered Gaussian process $\zeta_{t,f,b}$, which is the weighted sub-fractional Brownian motion. Furthermore, if there is a positive constant $c$ and $a \in (-1,\infty)$ such that $0 \leq f(u) \leq c u^{a}$ on $[0,T]$ for some $T>0$. Then, for $b \in (0,1)$, $\zeta_{t,f,b}$ exhibits infinite variation and zero quadratic variation, making it a non-semi-martingale. On the other hand, for $b \in (1,2]$, $\zeta_{t,f,b}$ is a continuous process of finite variation and thus a semi-martingale and for $b=0$ the process $\zeta_{t,f,0}$ is a square integrable continuous martingale. We also provide inferential studies using maximum likelihood estimation and perform simulations comparing various numerical methods for their efficiency in computing the finite-dimensional distributions of $\zeta_{t,f,b}$. Additionally, we extend the weighted sub-fractional Brownian motion to $\mathbb{R}^d$ by defining new covariance structures for measurable, bounded sets in $\mathbb{R}^d$. Finally, we define a stochastic integral with respect to $\zeta_{t,f,b}$ and introduce both the weighted sub-fractional Ornstein-Uhlenbeck process and the geometric weighted sub-fractional Brownian motion., Comment: 53 pages, 12 figures, 1 table

In this article we use a covariance function that arises from limit of fluctuations of the rescaled occupation time process of a branching particle system, to introduce a family of weighted long-range dependence Gaussian processes. In particular, we consider two subfamilies for which we show that the process is not a semimartingale, that the processes exhibit long-range dependence and have long-range memory of logarithmic order. Finally, we illustrate that this family of processes is useful for modeling real world data., Comment: 26 pages, 7 Figures, 2 Tables

Over the past decades, the increase in both frequency and intensity of large-scale wildfires due to climate change has emerged as a significant natural threat. The pressing need to design resilient landscapes capable of withstanding such disasters has become paramount, requiring the development of advanced decision-support tools. Existing methodologies, including Mixed Integer Programming, Stochastic Optimization, and Network Theory, have proven effective but are hindered by computational demands, limiting their applicability. In response to this challenge, we propose using artificial intelligence techniques, specifically Deep Reinforcement Learning, to address the complex problem of firebreak placement in the landscape. We employ value-function based approaches like Deep Q-Learning, Double Deep Q-Learning, and Dueling Double Deep Q-Learning. Utilizing the Cell2Fire fire spread simulator combined with Convolutional Neural Networks, we have successfully implemented a computational agent capable of learning firebreak locations within a forest environment, achieving good results. Furthermore, we incorporate a pre-training loop, initially teaching our agent to mimic a heuristic-based algorithm and observe that it consistently exceeds the performance of these solutions. Our findings underscore the immense potential of Deep Reinforcement Learning for operational research challenges, especially in fire prevention. Our approach demonstrates convergence with highly favorable results in problem instances as large as 40 x 40 cells, marking a significant milestone in applying Reinforcement Learning to this critical issue. To the best of our knowledge, this study represents a pioneering effort in using Reinforcement Learning to address the aforementioned problem, offering promising perspectives in fire prevention and landscape management, Comment: 20 pages, 15 figures

Research on language technology for the development of medical applications is currently a hot topic in Natural Language Understanding and Generation. Thus, a number of large language models (LLMs) have recently been adapted to the medical domain, so that they can be used as a tool for mediating in human-AI interaction. While these LLMs display competitive performance on automated medical texts benchmarks, they have been pre-trained and evaluated with a focus on a single language (English mostly). This is particularly true of text-to-text models, which typically require large amounts of domain-specific pre-training data, often not easily accessible for many languages. In this paper, we address these shortcomings by compiling, to the best of our knowledge, the largest multilingual corpus for the medical domain in four languages, namely English, French, Italian and Spanish. This new corpus has been used to train Medical mT5, the first open-source text-to-text multilingual model for the medical domain. Additionally, we present two new evaluation benchmarks for all four languages with the aim of facilitating multilingual research in this domain. A comprehensive evaluation shows that Medical mT5 outperforms both encoders and similarly sized text-to-text models for the Spanish, French, and Italian benchmarks, while being competitive with current state-of-the-art LLMs in English., Comment: LREC-COLING 2024

We present spot properties on 32 periodic young stellar objects in IC 5070. Long term, $\sim$5 yr, light curves in the $V$, $R$, and $I$-bands are obtained through the HOYS (Hunting Outbursting Young Stars) citizen science project. These are dissected into six months long slices, with 3 months oversampling, to measure 234 sets of amplitudes in all filters. We fit 180 of these with reliable spot solutions. Two thirds of spot solutions are cold spots, the lowest is 2150 K below the stellar temperature. One third are warm spots that are above the stellar temperature by less than $\sim$2000 K. Cold and warm spots have maximum surface coverage values of 40 percent, although only 16 percent of warm spots are above 20 percent surface coverage as opposed to 60 percent of the cold spots. Warm spots are most likely caused by a combination of plages and low density accretion columns, most common on objects without inner disc excess emission in $K-W2$. Five small hot spot solutions have $<3$ percent coverage and are 3000 - 5000 K above the stellar temperature. These are attributed to accretion, and four of them occur on the same object. The majority of our objects are likely to be accreting. However, we observe very few accretion hot spots as either the accretion is not stable on our timescale or the photometry is dominated by other features. We do not identify cyclical spot behaviour on the targets. We additionally identify and discuss a number of objects that have interesting amplitudes, phase changes, or spot properties., Comment: Accepted for publication by MNRAS. 17 + 7 pages, 7 + 23 figures, 1 table

We address the precise determination of the phase diagram of magic angle twisted bilayer graphene under hydrostatic pressure within a self-consistent Hartree-Fock method in real space, including all the remote bands of the system. We further present a novel algorithm that maps the full real-space density matrix to a reduced density matrix based on a $SU(4)$ symmetry of sublattice and valley degrees of freedom. We find a quantum critical point between a nematic and a Kekul\'e phase, and show also that our microscopic approach displays a strong particle-hole asymmetry in the weak coupling regime. We arrive then at the prediction that the superconductivity should be Ising-like in the hole-doped nematic regime, with spin-valley locking, and spin-triplet in the electron-doped regime., Comment: 27 pages, 28 figures

The Hunting Outbursting Young Stars (HOYS) project performs long-term, optical, multi-filter, high cadence monitoring of 25 nearby young clusters and star forming regions. Utilising Gaia DR3 data we have identified about 17000 potential young stellar members in 45 coherent astrometric groups in these fields. Twenty one of them are clear young groups or clusters of stars within one kiloparsec and they contain 9143 Gaia selected potential members. The cluster distances, proper motions and membership numbers are determined. We analyse long term (about 7yr) V, R, and I-band light curves from HOYS for 1687 of the potential cluster members. One quarter of the stars are variable in all three optical filters, and two thirds of these have light curves that are symmetric around the mean. Light curves affected by obscuration from circumstellar materials are more common than those affected by accretion bursts, by a factor of 2-4. The variability fraction in the clusters ranges from 10 to almost 100 percent, and correlates positively with the fraction of stars with detectable inner disks, indicating that a lot of variability is driven by the disk. About one in six variables shows detectable periodicity, mostly caused by magnetic spots. Two thirds of the periodic variables with disk excess emission are slow rotators, and amongst the stars without disk excess two thirds are fast rotators - in agreement with rotation being slowed down by the presence of a disk., Comment: accepted for publication in MNRAS, 1 table, 9 figures

Recent advancements in Large Language Models (LLMs) have led to high-quality Machine-Generated Text (MGT), giving rise to countless new use cases and applications. However, easy access to LLMs is posing new challenges due to misuse. To address malicious usage, researchers have released datasets to effectively train models on MGT-related tasks. Similar strategies are used to compile these datasets, but no tool currently unifies them. In this scenario, we introduce TextMachina, a modular and extensible Python framework, designed to aid in the creation of high-quality, unbiased datasets to build robust models for MGT-related tasks such as detection, attribution, mixcase, or boundary detection. It provides a user-friendly pipeline that abstracts away the inherent intricacies of building MGT datasets, such as LLM integrations, prompt templating, and bias mitigation. The quality of the datasets generated by TextMachina has been assessed in previous works, including shared tasks where more than one hundred teams trained robust MGT detectors., Comment: 14 pages, 10 figures

This paper presents a non-zoned discrete dielectric lens comprising two or three matching layers to reduce the 50-110 GHz frequency range reflections. Based on Chebyshev and binomial multi-section transformers, the designed models use matching layers at the top and bottom. In addition, the presented designs use pins instead of the conventional slots for the matching layers, thus easing the manufacturing process. The results show that the broadband realized gain obtained using the proposed design is higher for both the two- and three-layer design than the commonly used quarter-wave transformer. A Binomial lens with two matchings layers using 38 unit cells is fabricated and illuminated by an open-ended waveguide to validate the simulation results obtained using CST Microwave Studio. The fabrication process uses stereolithography additive manufacturing.

In 2020, the HAWC Collaboration presented the first catalog of gamma-ray sources emitting above 56 TeV and 100 TeV. With nine sources detected, this was the highest-energy source catalog to date. Here, we present the results of re-analysis of the old data, along with additional data acquired since then. We use a new version of the reconstruction software with better pointing accuracy and improved gamma/hadron separation. We now see more than 25 sources above 56 TeV, with most sources being located in the Galactic plane. The vast majority of these seem to be leptonic pulsar wind nebulae, but some have been shown to have hadronic emission. We will show spectra and discuss possible emission mechanisms of some of the most interesting sources, including the ones the HAWC Collaboration considers PeVatron candidates., Comment: ICRC 2023 proceeding

The detection of gravitational waves from a binary neutron star merger by Advanced LIGO and Advanced Virgo (GW170817), along with the discovery of the electromagnetic counterparts of this gravitational wave event, ushered in a new era of multimessenger astronomy, providing the first direct evidence that BNS mergers are progenitors of short gamma-ray bursts (GRBs). Such events may also produce very-high-energy (VHE, > 100GeV) photons which have yet to be detected in coincidence with a gravitational wave signal. The Cherenkov Telescope Array (CTA) is a next-generation VHE observatory which aims to be indispensable in this search, with an unparalleled sensitivity and ability to slew anywhere on the sky within a few tens of seconds. New observing modes and follow-up strategies are being developed for CTA to rapidly cover localization areas of gravitational wave events that are typically larger than the CTA field of view. This work will evaluate and provide estimations on the expected number of of gravitational wave events that will be observable with CTA, considering both on- and off-axis emission. In addition, we will present and discuss the prospects of potential follow-up strategies with CTA., Comment: Presented at the 38th International Cosmic Ray Conference (ICRC 2023), 2023 (arXiv:2309.08219)

This paper presents the overview of the AuTexTification shared task as part of the IberLEF 2023 Workshop in Iberian Languages Evaluation Forum, within the framework of the SEPLN 2023 conference. AuTexTification consists of two subtasks: for Subtask 1, participants had to determine whether a text is human-authored or has been generated by a large language model. For Subtask 2, participants had to attribute a machine-generated text to one of six different text generation models. Our AuTexTification 2023 dataset contains more than 160.000 texts across two languages (English and Spanish) and five domains (tweets, reviews, news, legal, and how-to articles). A total of 114 teams signed up to participate, of which 36 sent 175 runs, and 20 of them sent their working notes. In this overview, we present the AuTexTification dataset and task, the submitted participating systems, and the results., Comment: Accepted at SEPLN 2023

The classical Losev-Manin space can be interpreted as a toric compactification of the moduli space of $n$ points in the affine line modulo translation and scaling. Motivated by this, we study its higher-dimensional toric counterparts, which compactify the moduli space of $n$ distinct labeled points in affine space modulo translation and scaling. We show that these moduli spaces are a fibration over a product of projective spaces -- with fibers isomorphic to the Losev-Manin space -- and that they are isomorphic to the normalization of a Chow quotient. Moreover, we present a criterion to decide whether the blow-up of a toric variety along the closure of a subtorus is a Mori dream space. As an application, we demonstrate that a related generalization of the moduli space of pointed rational curves proposed by Chen, Gibney, and Krashen is not a Mori dream space when the number of points is at least nine, regardless of the dimension., Comment: 31 pages, 3 figures

When developing tools for automated cortical segmentation, the ability to produce topologically correct segmentations is important in order to compute geometrically valid morphometry measures. In practice, accurate cortical segmentation is challenged by image artifacts and the highly convoluted anatomy of the cortex itself. To address this, we propose a novel deep learning-based cortical segmentation method in which prior knowledge about the geometry of the cortex is incorporated into the network during the training process. We design a loss function which uses the theory of Laplace's equation applied to the cortex to locally penalize unresolved boundaries between tightly folded sulci. Using an ex vivo MRI dataset of human medial temporal lobe specimens, we demonstrate that our approach outperforms baseline segmentation networks, both quantitatively and qualitatively., Comment: Accepted at the 28th biennial international conference on Information Processing in Medical Imaging (IPMI 2023)

We study the limit fluctuations of the rescaled occupation time process of a branching particle system in $\mathbb{R}^d$, where the particles are subject to symmetric $\alpha$-stable migration ($0<\alpha\leq2$), critical binary branching, and general non-lattice lifetime distribution. We focus on two different regimes: lifetime distributions having finite expectation, and Pareto-type lifetime distributions, i.e. distributions belonging to the normal domain of attraction of a $\gamma$-stable law with $\gamma\in(0,1)$. In the latter case we show that, for dimensions $\alpha\gamma

In this expository note we discuss a class of graded algebras named Cox rings, which are naturally associated to algebraic varieties generalizing the homogeneous coordinate rings of projective spaces. Whenever the Cox ring is finitely generated, the variety admits a quotient presentation by a quasitorus, which resembles the quotient construction of the projective space. We discuss the problem of the finite generation of Cox rings from a geometric perspective and provide examples of both the finitely and non-finitely generated cases., Comment: 17 pages, 6 figures

We investigate the combinatorics of max-pooling layers, which are functions that downsample input arrays by taking the maximum over shifted windows of input coordinates, and which are commonly used in convolutional neural networks. We obtain results on the number of linearity regions of these functions by equivalently counting the number of vertices of certain Minkowski sums of simplices. We characterize the faces of such polytopes and obtain generating functions and closed formulas for the number of vertices and facets in a 1D max-pooling layer depending on the size of the pooling windows and stride, and for the number of vertices in a special case of 2D max-pooling., Comment: 35 pages, 11 figures, 4 tables. V2: Improved exposition, added computations in Section 4, and expanded analysis of data