Search

While demographic factors like age and gender change the way people talk, and in particular, the way people talk to machines, there is little investigation into how large pre-trained language models (LMs) can adapt to these changes. To remedy this gap, we consider how demographic factors in LM language skills can be measured to determine compatibility with a target demographic. We suggest clinical techniques from Speech Language Pathology, which has norms for acquisition of language skills in humans. We conduct evaluation with a domain expert (i.e., a clinically licensed speech language pathologist), and also propose automated techniques to complement clinical evaluation at scale. Empirically, we focus on age, finding LM capability varies widely depending on task: GPT-3.5 mimics the ability of humans ranging from age 6-15 at tasks requiring inference, and simultaneously, outperforms a typical 21 year old at memorization. GPT-3.5 also has trouble with social language use, exhibiting less than 50% of the tested pragmatic skills. Findings affirm the importance of considering demographic alignment and conversational goals when using LMs as public-facing tools. Code, data, and a package will be available., Comment: 17 pages, 9 figures, 5 tables

The synergy of metasurface-based holographic surfaces (HoloS) and reconfigurable intelligent surfaces (RIS) is considered a key aspect for future communication networks. However, the optimization of dynamic metasurfaces requires the use of numerical algorithms, for example, based on the singular value decomposition (SVD) and gradient descent methods, which are usually computationally intensive, especially when the number of elements is large. In this paper, we analyze low complexity designs for RIS-aided HoloS communication systems, in which the configurations of the HoloS transmitter and the RIS are given in a closed-form expression. We consider implementations based on diagonal and non-diagonal RISs. Over line-of-sight channels, we show that the proposed schemes provide performance that is close to that offered by complex numerical methods., Comment: Submitted to EUSIPCO 2023

Hot giant planets like MASCARA-1 b are expected to have thermally inverted atmospheres, that makes them perfect laboratory for the atmospheric characterization through high-resolution spectroscopy. Nonetheless, previous attempts of detecting the atmosphere of MASCARA-1 b in transmission have led to negative results. In this paper we aim at the detection of the optical emission spectrum of MASCARA-1 b. We used the high-resolution spectrograph PEPSI to observe MASCARA-1 (spectral type A8) near the secondary eclipse of the planet. We cross-correlated the spectra with synthetic templates computed for several atomic and molecular species. We obtained the detection of FeI, CrI and TiI in the atmosphere of MASCARA-1 b with a S/N ~7, 4 and 5 respectively, and confirmed the expected systemic velocity of ~13 km/s and the radial velocity semi-amplitude of MASCARA-1 b of ~200 km/s. The detection of Ti is of particular importance in the context of the recently proposed Ti cold-trapping below a certain planetary equilibrium temperature. We confirm the presence of an the atmosphere around MASCARA-1 b through emission spectroscopy. We conclude that the atmospheric non detection in transmission spectroscopy is due to the high gravity of the planet and/or to the overlap between the planetary track and its Doppler shadow.

Absorption features in stellar atmospheres are often used to calibrate photocentric velocities for kinematic analysis of further spectral lines. The Li feature at $\sim$ 6708 {\AA} is commonly used, especially in the case of young stellar objects for which it is one of the strongest absorption lines. However, this is a complex line comprising two isotope fine-structure doublets. We empirically measure the wavelength of this Li feature in a sample of young stars from the PENELLOPE/VLT programme (using X-Shooter, UVES and ESPRESSO data) as well as HARPS data. For 51 targets, we fit 314 individual spectra using the STAR-MELT package, resulting in 241 accurately fitted Li features, given the automated goodness-of-fit threshold. We find the mean air wavelength to be 6707.856 {\AA}, with a standard error of 0.002 {\AA} (0.09 km/s) and a weighted standard deviation of 0.026 {\AA} (1.16 km/s). The observed spread in measured positions spans 0.145 {\AA}, or 6.5 km/s, which is up to a factor of six higher than typically reported velocity errors for high-resolution studies. We also find a correlation between the effective temperature of the star and the wavelength of the central absorption. We discuss how exclusively using this Li feature as a reference for photocentric velocity in young stars could potentially be introducing a systematic positive offset in wavelength to measurements of further spectral lines. If outflow tracing forbidden lines, such as [O i] 6300 {\AA}, are actually more blueshifted than previously thought, this then favours a disk wind as the origin for such emission in young stars., Comment: Accepted for publication in A&A

In this paper, we study surface-based communication systems based on different levels of channel state information for system optimization. We analyze the system performance in terms of rate and degrees of freedom (DoF). We show that the deployment of a reconfigurable intelligent surface (RIS) results in increasing the number of DoF, by extending the near-field region. Over Rician fading channels, we show that an RIS can be efficiently optimized only based on the positions of the transmitting and receiving surfaces, while providing good performance if the Rician fading factor is not too small.

We investigate the star formation history of the IC1396 region by studying its kinematics and completing the population census. We use multiwavelength data, combining optical spectroscopy (to identify and classify new members), near-infrared photometry (to trace shocks, jets, and outflows and the interactions between the cluster members and the cloud), along with Gaia EDR3 to identify new potential members in the multidimensional proper motion/parallax space. The revised Gaia EDR3 distance is 925$\pm$73 pc, slightly closer than previously obtained with DR2. The Gaia data reveal four distinct subclusters in the region. These subclusters are consistent in distance but display differences in proper motion. This, with their age differences, hints towards a complex and varied star formation history. Gaia data also unveil the intermediate-mass objects that tend to evade spectroscopic and disk surveys. Our analysis allows us to identify 334 new members. We estimate an average age of $\sim$4 Myr, confirming previous age estimates. With the new members added to our study, we estimate a disk fraction of 28\%, lower than previous values, due to our method detecting mainly new, diskless intermediate-mass stars. We find age differences between the subclusters, which evidences a complex star formation history with different episodes of star formation., Comment: A&A in press

Algorithms for text-generation in dialogue can be misguided. For example, in task-oriented settings, reinforcement learning that optimizes only task-success can lead to abysmal lexical diversity. We hypothesize this is due to poor theoretical understanding of the objectives in text-generation and their relation to the learning process (i.e., model training). To this end, we propose a new theoretical framework for learning to generate text in dialogue. Compared to existing theories of learning, our framework allows for analysis of the multi-faceted goals inherent to text-generation. We use our framework to develop theoretical guarantees for learners that adapt to unseen data. As an example, we apply our theory to study data-shift within a cooperative learning algorithm proposed for the GuessWhat?! visual dialogue game. From this insight, we propose a new algorithm, and empirically, we demonstrate our proposal improves both task-success and human-likeness of the generated text. Finally, we show statistics from our theory are empirically predictive of multiple qualities of the generated dialogue, suggesting our theory is useful for model-selection when human evaluations are not available.

[Abridged] We present a novel method to study the thermal emission of exoplanets as a function of orbital phase at very high spectral resolution, and apply it to investigate the climate of the ultra-hot Jupiter KELT-9b. We combine 3 nights of HARPS-N and 2 nights of CARMENES optical spectra, covering orbital phases between quadratures (0.25 < phi < 0.75), when the planet shows its day-side hemisphere with different geometries. We co-add the signal of thousands of FeI lines through cross-correlation, which we map to a likelihood function. We investigate the phase-dependence of: (i) the line depths of FeI, and (ii) their Doppler shifts, by introducing a new method that exploits the very high spectral resolution of our observations. We confirm a previous detection of FeI emission and demonstrate a combined precision of 0.5 km s-1 on the orbital properties of KELT-9b. By studying the phase-resolved Doppler shift of FeI lines, we detect an anomaly in the planet's orbital radial velocity well-fitted with a slightly eccentric orbit (e = 0.016$\pm$0.003, w = 150$^{+13\circ}_{-11},~5\sigma$ preference). However, we argue that such anomaly can be explained by a day-night wind of a few km s-1 blowing neutral iron gas. Additionally, we find that the FeI emission line depths are symmetric around the substellar point within 10 deg ($2\sigma$). We show that these results are qualitatively compatible with predictions from general circulation models for ultra-hot Jupiter planets. Very high-resolution spectroscopy phase curves have the sensitivity to reveal a phase dependence in both the line depths and their Doppler shifts throughout the orbit. They are highly complementary to space-based phase curves obtained with HST and JWST, and open a new window into the still poorly understood climate and atmospheric structure of the hottest planets known., Comment: Recommended for publication on A&A after referee report, awaiting acceptance. 25 pages, 19 figures, 3 tables

Transmission spectroscopy is among the most fruitful techniques to infer the main opacity sources present in the upper atmosphere of a transiting planet and to constrain the composition of the thermosphere and of the unbound exosphere. Not having a public tool able to automatically extract a high-resolution transmission spectrum creates a problem of reproducibility for scientific results. As a consequence, it is very difficult to compare the results obtained by different research groups and to carry out a homogeneous characterization of the exoplanetary atmospheres. In this work, we present a standard, publicly available, user-friendly tool, named SLOPpy (Spectral Lines Of Planets with python), to automatically extract and analyze the optical transmission spectrum of exoplanets as accurately as possible. Several data reduction steps are first performed by SLOPpy to correct the input spectra for sky emission, atmospheric dispersion, the presence of telluric features and interstellar lines, center-to-limb variation, and Rossiter-McLaughlin effect, thus making it a state-of-the-art tool. The pipeline has successfully been applied to HARPS and HARPS-N data of ideal targets for atmospheric characterization. To first assess the code's performance and to validate its suitability, here we present a comparison with the results obtained from the previous analyses of other works on HD 189733 b, WASP-76 b, WASP-127 b, and KELT-20 b. Comparing our results with other works that have analyzed the same datasets, we conclude that this tool gives results in agreement with the published results within 1$\sigma$ most of the time, while extracting, with SLOPpy, the planetary signal with a similar or higher statistical significance., Comment: 31 pages, 28 figures, accepted for publication in Astronomy & Astrophysics

Investigating cooperativity of interlocutors is central in studying pragmatics of dialogue. Models of conversation that only assume cooperative agents fail to explain the dynamics of strategic conversations. Thus, we investigate the ability of agents to identify non-cooperative interlocutors while completing a concurrent visual-dialogue task. Within this novel setting, we study the optimality of communication strategies for achieving this multi-task objective. We use the tools of learning theory to develop a theoretical model for identifying non-cooperative interlocutors and apply this theory to analyze different communication strategies. We also introduce a corpus of non-cooperative conversations about images in the GuessWhat?! dataset proposed by De Vries et al. (2017). We use reinforcement learning to implement multiple communication strategies in this context and find empirical results validate our theory.

Multiclass neural networks are a common tool in modern unsupervised domain adaptation, yet an appropriate theoretical description for their non-uniform sample complexity is lacking in the adaptation literature. To fill this gap, we propose the first PAC-Bayesian adaptation bounds for multiclass learners. We facilitate practical use of our bounds by also proposing the first approximation techniques for the multiclass distribution divergences we consider. For divergences dependent on a Gibbs predictor, we propose additional PAC-Bayesian adaptation bounds which remove the need for inefficient Monte-Carlo estimation. Empirically, we test the efficacy of our proposed approximation techniques as well as some novel design-concepts which we include in our bounds. Finally, we apply our bounds to analyze a common adaptation algorithm that uses neural networks.

A new approach to solving eigenvalue optimization problems for large structured matrices is proposed and studied. The class of optimization problems considered is related to computing structured pseudospectra and their extremal points, and to structured matrix nearness problems such as computing the structured distance to instability or to singularity. The structure can be a general linear structure and includes, for example, large matrices with a given sparsity pattern, matrices with given range and co-range, and Hamiltonian matrices. Remarkably, the eigenvalue optimization can be performed on the manifold of complex (or real) rank-1 matrices, which yields a significant reduction of storage and in some cases of the computational cost. The method relies on a constrained gradient system and the projection of the gradient onto the tangent space of the manifold of complex rank-$1$ matrices. It is shown that near a local minimizer this projection is very close to the identity map, and so the computationally favorable rank-1 projected system behaves locally like the %computationally expensive gradient system.

This is the second paper in a series aimed at modeling the black hole (BH) mass function, from the stellar to the (super)massive regime. In the present work we focus on (super)massive BHs and provide an ab-initio computation of their mass function across cosmic times. We consider two main mechanisms to grow the central BH, that are expected to cooperate in the high-redshift star-forming progenitors of local massive galaxies. The first is the gaseous dynamical friction process, that can cause the migration toward the nuclear regions of stellar-mass BHs originated during the intense bursts of star formation in the gas-rich host progenitor galaxy, and the buildup of a central heavy BH seed $M_\bullet\sim 10^{3-5}\, M_\odot$ within short timescales $\lesssim$ some $10^7$ yr. The second mechanism is the standard Eddington-type gas disk accretion onto the heavy BH seed, through which the central BH can become (super)massive $M_\bullet\sim 10^{6-10}\, M_\odot$ within the typical star-formation duration $\lesssim 1$ Gyr of the host. We validate our semi-empirical approach by reproducing the observed redshift-dependent bolometric AGN luminosity functions and Eddington ratio distributions, and the relationship between the star-formation and the bolometric luminosity of the accreting central BH. We then derive the relic (super)massive BH mass function at different redshifts via a generalized continuity equation approach, and compare it with present observational estimates. Finally, we reconstruct the overall BH mass function from the stellar to the (super)massive regime, over more than ten orders of magnitudes in BH mass., Comment: 29 pages, 11 figures. Accepted by ApJ