Search

This paper provides a comprehensive survey of the latest research on multilingual large language models (MLLMs). MLLMs not only are able to understand and generate language across linguistic boundaries, but also represent an important advancement in artificial intelligence. We first discuss the architecture and pre-training objectives of MLLMs, highlighting the key components and methodologies that contribute to their multilingual capabilities. We then discuss the construction of multilingual pre-training and alignment datasets, underscoring the importance of data quality and diversity in enhancing MLLM performance. An important focus of this survey is on the evaluation of MLLMs. We present a detailed taxonomy and roadmap covering the assessment of MLLMs' cross-lingual knowledge, reasoning, alignment with human values, safety, interpretability and specialized applications. Specifically, we extensively discuss multilingual evaluation benchmarks and datasets, and explore the use of LLMs themselves as multilingual evaluators. To enhance MLLMs from black to white boxes, we also address the interpretability of multilingual capabilities, cross-lingual transfer and language bias within these models. Finally, we provide a comprehensive review of real-world applications of MLLMs across diverse domains, including biology, medicine, computer science, mathematics and law. We showcase how these models have driven innovation and improvements in these specialized fields while also highlighting the challenges and opportunities in deploying MLLMs within diverse language communities and application scenarios. We listed the paper related in this survey and publicly available at https://github.com/tjunlp-lab/Awesome-Multilingual-LLMs-Papers.

Project Lx Conventos aims to study, in a systematic and integrated manner, the impact of the dissolution of religious orders in the dynamics of urban transformation in nineteenth century Lisbon. After the liberal revolution and the civil war, in the 19th century, the dissolution of religious orders led to the alienation, in Lisbon, of nearly 130 religious buildings which were then given profane occupations (mainly public services) or demolished and divided in plots, originating new urban realities. Project Lx Conventos thus aims to show that the extinction of the convents was decisive in the urban development of Lisbon, in the eighteen hundreds. The project stands on a large set of multimedia data which includes historic and contemporary cartography and geo-referenced photos, videos and 3D models, provided by the projects partners, Lisbon Municipality and the Portuguese National Archive, Torre do Tombo. Supported by these materials, the project's team is creating an online system that will implement a spatial and temporal navigation of these resources integrated in an interactive Map of Lisbon. Besides spatially locating and analyzing the data available for each of the religious buildings considered in the project, the tool integrates cutting edge interaction technology for: 1) Enabling a temporal voyage over the available traces of religious buildings; 2) Analyzing the evolution of religious buildings and their surroundings, through available data; 3) Using 3D representations of the buildings for accessing related data, through time. In this paper, the tools under development in the context of Lx Conventos are described, as well as the technologies supporting them. The current status of the system is presented and future developments are proposed.

Sentence encoder encode the semantics of their input, enabling key downstream applications such as classification, clustering, or retrieval. In this paper, we present Serafim PT*, a family of open-source sentence encoders for Portuguese with various sizes, suited to different hardware/compute budgets. Each model exhibits state-of-the-art performance and is made openly available under a permissive license, allowing its use for both commercial and research purposes. Besides the sentence encoders, this paper contributes a systematic study and lessons learned concerning the selection criteria of learning objectives and parameters that support top-performing encoders.

A numerical semigroup $S$ is coated with odd elements (Coe-semigroup), if $\left\{x-1, x+1\right\}\subseteq S$ for all odd element $x$ in $S$. In this note, we will study this kind of numerical semigroups. In particular, we are interested in the study of the Frobenius number, gender and embedding dimension of a numerical semigroup of this type.

Autonomously tailoring content to a set of predetermined affective patterns has long been considered the holy grail of affect-aware human-computer interaction at large. The experience-driven procedural content generation framework realises this vision by searching for content that elicits a certain experience pattern to a user. In this paper, we propose a novel reinforcement learning (RL) framework for generating affect-tailored content, and we test it in the domain of racing games. Specifically, the experience-driven RL (EDRL) framework is given a target arousal trace, and it then generates a racetrack that elicits the desired affective responses for a particular type of player. EDRL leverages a reward function that assesses the affective pattern of any generated racetrack from a corpus of arousal traces. Our findings suggest that EDRL can accurately generate affect-driven racing game levels according to a designer's style and outperforms search-based methods for personalised content generation. The method is not only directly applicable to game content generation tasks but also employable broadly to any domain that uses content for affective adaptation., Comment: 9 pages, 4 figures, 1 table

Retrieval-augmented generation resorts to content retrieved from external sources in order to leverage the performance of large language models in downstream tasks. The excessive volume of retrieved content, the possible dispersion of its parts, or their out of focus range may happen nevertheless to eventually have a detrimental rather than an incremental effect. To mitigate this issue and improve retrieval-augmented generation, we propose a method to refine the retrieved content before it is included in the prompt by resorting to meta-prompting optimization. Put to empirical test with the demanding multi-hop question answering task from the StrategyQA dataset, the evaluation results indicate that this method outperforms a similar retrieval-augmented system but without this method by over 30%.

Severe impairment of the central motor network can result in loss of motor function, clinically recognized as Locked-in Syndrome. Advances in Brain-Computer Interfaces offer a promising avenue for partially restoring compromised communicative abilities by decoding different types of hand movements from the sensorimotor cortex. In this study, we collected ECoG recordings from 8 epilepsy patients and compared the decodability of individual finger flexion and hand gestures with the resting state, as a proxy for a one-dimensional brain-click. The results show that all individual finger flexion and hand gestures are equally decodable across multiple models and subjects (>98.0\%). In particular, hand movements, involving index finger flexion, emerged as promising candidates for brain-clicks. When decoding among multiple hand movements, finger flexion appears to outperform hand gestures (96.2\% and 92.5\% respectively) and exhibit greater robustness against misclassification errors when all hand movements are included. These findings highlight that optimized classical machine learning models with feature engineering are viable decoder designs for communication-assistive systems., Comment: 6 pages, 4 figures

Numerical ``direct'' approaches to time-optimal control often fail to find solutions that are singular in the sense of the Pontryagin Maximum Principle, performing better when searching for saturated (bang-bang) solutions. In previous work by one of the authors, singular solutions were shown to exist for the time-optimal control problem for fully actuated mechanical systems under hard torque constraints. Explicit formulas, based on a Lie theoretic analysis of the problem, were given for singular segments of trajectories, but the global structure of solutions remains unknown. In this work, we review the aforementioned framework, and show how to effectively combine these formulas with the use of general-purpose optimal control software packages. By using the explicit formula given by the theory in the intervals where the numerical solution enters a singular arc, we not only obtain an algebraic expression for the control in that interval but we are also able to remove artifacts present in the numerical solution. In this way, the best features of numerical algorithms and theory complement each other and provide a better picture of the global optimal structure. We illustrate the technique on a two degree of freedom robotic arm example, using two distinct optimal control numerical software packages running on different programming languages.

A common property of globular clusters (GC) is to host multiple populations characterized by peculiar chemical abundances. Recent photometric studies suggest that the He content could vary between the populations of a GC by up to $\Delta$He $\sim$ 0.13, in mass fraction. The initial He content impacts the evolution of low-mass stars by ultimately modifying their lifetimes, luminosity, temperatures, and, more generally, the morphology of post-RGB evolutionary tracks in the Hertzsprung-Russell diagram. We present new physically accurate isochrones with different initial He-enrichments and metallicities, with a focus on the methods implemented to deal with the post-RGB phases. The isochrones are based on tracks computed with the stellar evolution code STAREVOL for different metallicities (Z = 0.0002, 0.0009, 0.002, and 0.008) and with different He-enrichment (from 0.25 to 0.6 in mass fraction). We describe the effect of He-enrichment on the morphology of the isochrones and test these by comparing the predicted number counts of HB and AGB stars with those of selected GCs. Comparing the number ratios, we find that our new theoretical ones agree with the observed values within $1\sigma$ in most cases. The work presented here sets the ground for future studies on stellar populations in globular clusters, in which the abundances of light elements in He-enhanced models will rely on different assumptions for the causes of this enrichment. The developed methodology permits the computation of isochrones from new stellar tracks with non-canonical stellar processes. The checked number counts ensure that, at least in this reference set, the contribution of the luminous late stages of stellar evolution to the integrated light of a GC is represented adequately., Comment: 15 pages, 12 figures, accepted for publication in Astronomy & Astrophysics

In the field of autonomous Unmanned Aerial Vehicles (UAVs) landing, conventional approaches fall short in delivering not only the required precision but also the resilience against environmental disturbances. Yet, learning-based algorithms can offer promising solutions by leveraging their ability to learn the intelligent behaviour from data. On one hand, this paper introduces a novel multimodal transformer-based Deep Learning detector, that can provide reliable positioning for precise autonomous landing. It surpasses standard approaches by addressing individual sensor limitations, achieving high reliability even in diverse weather and sensor failure conditions. It was rigorously validated across varying environments, achieving optimal true positive rates and average precisions of up to 90%. On the other hand, it is proposed a Reinforcement Learning (RL) decision-making model, based on a Deep Q-Network (DQN) rationale. Initially trained in sumlation, its adaptive behaviour is successfully transferred and validated in a real outdoor scenario. Furthermore, this approach demonstrates rapid inference times of approximately 5ms, validating its applicability on edge devices.

We propose and demonstrate a modulation transfer protocol to increase the detection sensitivity of a Rydberg RF receiver to fields out of resonance from the transition between Rydberg levels. This protocol is based on a phase modulation of the control field used to create the Electromagnetically Induced Transparency (EIT) signal. The nonlinear wave-mixing of the multi-component coupling laser and the probe laser transfers the modulation to the probe laser, which is used for RF-field detection. The measurements compare well with semi-classical simulations of atom-light interaction and show an improvement in the RF bandwidth of the sensor and an improved sensitivity of the response to weak fields.

Most Galactic Globular Clusters (GCs) harbour multiple populations of stars (MPs), composed of at least two generations: the first characterized by a "standard" $\alpha$-enhanced metal mixture, as observed in field halo stars of the Milky Way, and the second displaying anti-correlated CN--ONa chemical abundance pattern in combination with an enhanced helium fraction. Adequate collections of stellar spectra are needed to characterize the effect of such stellar abundance changes on the integrated light of GCs. We present a grid of synthetic stellar spectra covering the atmospheric parameters relevant to old stellar populations at four subsolar metallicities and two abundance patterns, representative of first- and second-generations of stars in GCs. Integrated spectra of populations were computed using our stellar grid and empirical stellar populations, namely, colour-magnitude diagrams from literature for Galactic GCs. The spectra range from 290 to 1000nm, where we measured the effect on several spectrophotometric indices due to the surface abundance variations attributed to MPs. We find non-negligible effects of the MPs on spectroscopic indices sensitive to C, N, Ca, or Na, and on Balmer indices; we also describe how MPs modify specific regions in the near-UV and near-IR that can be measured with narrow or medium photometric passbands. The effects vary with metallicity. A number of these changes remain detectable even when accounting for the stochastic fluctuations due to the finite nature of the stellar population cluster.

Leveraging research on the neural modelling of Portuguese, we contribute a collection of datasets for an array of language processing tasks and a corresponding collection of fine-tuned neural language models on these downstream tasks. To align with mainstream benchmarks in the literature, originally developed in English, and to kick start their Portuguese counterparts, the datasets were machine-translated from English with a state-of-the-art translation engine. The resulting PORTULAN ExtraGLUE benchmark is a basis for research on Portuguese whose improvement can be pursued in future work. Similarly, the respective fine-tuned neural language models, developed with a low-rank adaptation approach, are made available as baselines that can stimulate future work on the neural processing of Portuguese. All datasets and models have been developed and are made available for two variants of Portuguese: European and Brazilian., Comment: Preprint - Paper accepted for BUCC 2024

This article investigates how AI-generated content can disrupt central revenue streams of the creative industries, in particular the collection of dividends from intellectual property (IP) rights. It reviews the IP and copyright questions related to the input and output of generative AI systems. A systematic method is proposed to assess whether AI-generated outputs, especially images, infringe previous copyrights, using a similarity metric (CLIP) between images against historical copyright rulings. An examination (economic and technical feasibility) of previously proposed compensation frameworks reveals their financial implications for creatives and IP holders. Lastly, we propose a novel IP framework for compensation of artists and IP holders based on their published "licensed AIs" as a new medium and asset from which to collect AI royalties., Comment: 7 pages, 2 figures, submitted to AAAI

This paper introduces kDGLM, an R package designed for Bayesian analysis of Generalized Dynamic Linear Models (GDLM), with a primary focus on both uni- and multivariate exponential families. Emphasizing sequential inference for time series data, the kDGLM package provides comprehensive support for fitting, smoothing, monitoring, and feed-forward interventions. The methodology employed by kDGLM, as proposed in Alves et al. (2024), seamlessly integrates with well-established techniques from the literature, particularly those used in (Gaussian) Dynamic Models. These include discount strategies, autoregressive components, transfer functions, and more. Leveraging key properties of the Kalman filter and smoothing, kDGLM exhibits remarkable computational efficiency, enabling virtually instantaneous fitting times that scale linearly with the length of the time series. This characteristic makes it an exceptionally powerful tool for the analysis of extended time series. For example, when modeling monthly hospital admissions in Brazil due to gastroenteritis from 2010 to 2022, the fitting process took a mere 0.11s. Even in a spatial-time variant of the model (27 outcomes, 110 latent states, and 156 months, yielding 17,160 parameters), the fitting time was only 4.24s. Currently, the kDGLM package supports a range of distributions, including univariate Normal (unknown mean and observational variance), bivariate Normal (unknown means, observational variances, and correlation), Poisson, Gamma (known shape and unknown mean), and Multinomial (known number of trials and unknown event probabilities). Additionally, kDGLM allows the joint modeling of multiple time series, provided each series follows one of the supported distributions. Ongoing efforts aim to continuously expand the supported distributions.

The combination of language processing and image processing keeps attracting increased interest given recent impressive advances that leverage the combined strengths of both domains of research. Among these advances, the task of editing an image on the basis solely of a natural language instruction stands out as a most challenging endeavour. While recent approaches for this task resort, in one way or other, to some form of preliminary preparation, training or fine-tuning, this paper explores a novel approach: We propose a preparation-free method that permits instruction-guided image editing on the fly. This approach is organized along three steps properly orchestrated that resort to image captioning and DDIM inversion, followed by obtaining the edit direction embedding, followed by image editing proper. While dispensing with preliminary preparation, our approach demonstrates to be effective and competitive, outperforming recent, state of the art models for this task when evaluated on the MAGICBRUSH dataset.

To foster the neural encoding of Portuguese, this paper contributes foundation encoder models that represent an expansion of the still very scarce ecosystem of large language models specifically developed for this language that are fully open, in the sense that they are open source and openly distributed for free under an open license for any purpose, thus including research and commercial usages. Like most languages other than English, Portuguese is low-resourced in terms of these foundational language resources, there being the inaugural 900 million parameter Albertina and 335 million Bertimbau. Taking this couple of models as an inaugural set, we present the extension of the ecosystem of state-of-the-art open encoders for Portuguese with a larger, top performance-driven model with 1.5 billion parameters, and a smaller, efficiency-driven model with 100 million parameters. While achieving this primary goal, further results that are relevant for this ecosystem were obtained as well, namely new datasets for Portuguese based on the SuperGLUE benchmark, which we also distribute openly.

To advance the neural decoding of Portuguese, in this paper we present a fully open Transformer-based, instruction-tuned decoder model that sets a new state of the art in this respect. To develop this decoder, which we named Gerv\'asio PT*, a strong LLaMA~2 7B model was used as a starting point, and its further improvement through additional training was done over language resources that include new instruction data sets of Portuguese prepared for this purpose, which are also contributed in this paper. All versions of Gerv\'asio are open source and distributed for free under an open license, including for either research or commercial usage, and can be run on consumer-grade hardware, thus seeking to contribute to the advancement of research and innovation in language technology for Portuguese.

We explore the relationship between sodium (Na$^+$) and potassium (K$^+$) gating variables in the 4-dimensional (4D) Hodgkin-Huxley (HH) electrophysiology model and reduce its complexity by deriving new 3D and 2D models that retain the original model's dynamic properties. The new 3D and 2D models are based on the relationship $h \simeq c - n$ between the gating variables $h$ and $n$ of the 4D HH model, where $c$ is a constant, which suggests an interdependence between the dynamics of Na$^+$ and K$^+$ transmembrane voltage-gated channels. We derive the corresponding cable equations for the three HH-type models and demonstrate that the action potential propagates along the axon at a speed described by $v(R, C_m) = \alpha / (C_m R^{\beta}) = \gamma D^{\beta}$, where $\alpha > 0$, $0 < \beta < 1$, and $\gamma$ are constants independent of the local stimulus intensity, $D$ is the diffusion coefficient of the electric signal along the axon, $C_m$ is the axon transmembrane capacitance, and $R$ is the axon conducting resistivity. The width $w$ of the action potential spikes is inversely related to the resistivity of the axon, with $w = \alpha_2 / R^{\beta_2}$, where $\alpha_2 > 0$ and $\beta_2 > 0$., Comment: 8 pages, 12 figures

There are several tools available to infer phylogenetic trees, which depict the evolutionary relationships among biological entities such as viral and bacterial strains in infectious outbreaks, or cancerous cells in tumor progression trees. These tools rely on several inference methods available to produce phylogenetic trees, with resulting trees not being unique. Thus, methods for comparing phylogenies that are capable of revealing where two phylogenetic trees agree or differ are required. An approach is then to compute a similarity or dissimilarity measure between trees, with the Robinson- Foulds distance being one of the most used, and which can be computed in linear time and space. Nevertheless, given the large and increasing volume of phylogenetic data, phylogenetic trees are becoming very large with hundreds of thousands of leafs. In this context, space requirements become an issue both while computing tree distances and while storing trees. We propose then an efficient implementation of the Robinson-Foulds distance over trees succinct representations. Our implementation generalizes also the Robinson-Foulds distances to labelled phylogenetic trees, i.e., trees containing labels on all nodes, instead of only on leaves. Experimental results show that we are able to still achieve linear time while requiring less space. Our implementation is available as an open-source tool at https://github.com/pedroparedesbranco/TreeDiff.

This work proposes a two-dimensional electrophysiological model for describing neuronal responses to external electric stimuli under patch-clamped conditions. Our proposed model successfully captures the key features of the Hodgkin-Huxley model, while offering improved informativeness and numerical efficiency. By analysing bifurcation diagrams from various reduced models, we describe the contributions of different ionic channels to the stability or instability of the neurones' response. We also explore the influence of various cell parameters in the neurone's spiking frequency, including the membrane capacitance and the ionic channels' conductivities and Nerst potentials. Additionally, we suggest a spatial model for the axon that effectively simulates unidirectional signal transmission along neurons unaffected by membrane potential perturbations. After analysing the characteristics of this compartmental model, including the impact of intracellular resistance and membrane capacitance on the signal's velocity, we observed considerable discrepancies between our findings and those predicted by the cable equation model for signal propagation. Moreover, our investigation reveals that when the transmission of neuronal signals occurs in multiple peaks, there is a transient variation in their wavelength and frequency, resulting in different peak velocities. Furthermore, our findings consistently demonstrate that the first peak - in some cases, the only peak observed - maintains a relatively constant velocity along the axon regardless of the stimulus intensity and is always faster than the subsequent peaks., Comment: 40 pages, 48 figures