Search

Heat transport in highly turbulent convection is not well understood. In this paper, we simulate compressible convection in a box of aspect ratio 4 using computationally-efficient MacCormack-TVD finite difference method on single and multi-GPUs, and reach very high Rayleigh number ($\mathrm{Ra}$) -- $10^{15}$ in two dimensions and $10^{11}$ in three dimensions. We show that the Nusselt number $\mathrm{Nu} \propto \mathrm{Ra}^{0.3}$ (classical scaling) that differs strongly from the ultimate-regime scaling, which is $\mathrm{Nu} \propto \mathrm{Ra}^{1/2}$. The bulk temperature drops adiabatically along the vertical even for high $\mathrm{Ra}$, which is in contrast to the constant bulk temperature in Rayleigh-B\'{e}nard convection (RBC). Unlike RBC, the density decreases with height. In addition, the vertical pressure-gradient ($-dp/dz$) nearly matches the buoyancy term ($\rho g$). But, the difference, $-dp/dz-\rho g$, is equal to the nonlinear term that leads to Reynolds number $\mathrm{Re} \propto \mathrm{Ra}^{1/2}$., Comment: 17 pages, 16 figures

Speech-to-Speech Translation (S2ST) models transform speech from one language to another target language with the same linguistic information. S2ST is important for bridging the communication gap among communities and has diverse applications. In recent years, researchers have introduced direct S2ST models, which have the potential to translate speech without relying on intermediate text generation, have better decoding latency, and the ability to preserve paralinguistic and non-linguistic features. However, direct S2ST has yet to achieve quality performance for seamless communication and still lags behind the cascade models in terms of performance, especially in real-world translation. To the best of our knowledge, no comprehensive survey is available on the direct S2ST system, which beginners and advanced researchers can look upon for a quick survey. The present work provides a comprehensive review of direct S2ST models, data and application issues, and performance metrics. We critically analyze the models' performance over the benchmark datasets and provide research challenges and future directions.

Count data analysis is essential across diverse fields, from ecology and accident analysis to single-cell RNA sequencing (scRNA-seq) and metagenomics. While log transformations are computationally efficient, model-based approaches such as the Poisson-Log-Normal (PLN) model provide robust statistical foundations and are more amenable to extensions. The PLN model, with its latent Gaussian structure, not only captures overdispersion but also enables correlation between variables and inclusion of covariates, making it suitable for multivariate count data analysis. Variational approximations are a golden standard to estimate parameters of complex latent variable models such as PLN, maximizing a surrogate likelihood. However, variational estimators lack theoretical statistical properties such as consistency and asymptotic normality. In this paper, we investigate the consistency and variance estimation of PLN parameters using M-estimation theory. We derive the Sandwich estimator, previously studied in Westling and McCormick (2019), specifically for the PLN model. We compare this approach to the variational Fisher Information method, demonstrating the Sandwich estimator's effectiveness in terms of coverage through simulation studies. Finally, we validate our method on a scRNA-seq dataset.

In this paper, using \textit{hydrodynamic entropy} we quantify the multiscale disorder in Euler and hydrodynamic turbulence. These examples illustrate that the hydrodynamic entropy is not extensive because it is not proportional to the system size. Consequently, we cannot add hydrodynamic and thermodynamic entropies, which measure disorder at macroscopic and microscopic scales, respectively. In this paper, we also discuss the hydrodynamic entropy for the time-dependent Ginzburg-Landau equation and Ising spins., Comment: To appear in Phys. Rev. E

The initial value problem (IVP) for the non-isotropic Schr\"odinger equation posed on the two-dimensional cylinders and $\mathbb{T}^2$ is considered. The IVP is shown to be locally well-posed for small initial data in $H^s(\mathbb{T}\times\mathbb{R})$ if $s\geq0$. For the IVP posed on $\mathbb{R}\times\mathbb{T}$, given data are considered in the anisotropic Sobolev spaces thereby obtaining the local well-posedness result in $H^{s_1, s_2}(\mathbb{R}\times\mathbb{T})$, if $s_1\geq0$ and $s_2>\frac12$. In the purely periodic case, a particular case of the IVP is shown to be locally well-posed for any given initial data in $H^s(\mathbb{T}^2)$ if $s>\frac18$. In some cases, ill-posedness issues are also considered showing that the IVP posed on $\mathbb{T}\times \mathbb{R}$, in the focusing case, is ill-posed in the sense that the application data-solution fails to be uniformly continuous for data in $H^s(\mathbb{T}\times\mathbb{R})$ if $-\frac12\leq s<0$., Comment: 23 pages

Diverse computing paradigms have emerged to meet the growing needs for intelligent energy-efficient systems. The Margin Propagation (MP) framework, being one such initiative in the analog computing domain, stands out due to its scalability across biasing conditions, temperatures, and diminishing process technology nodes. However, the lack of digital-like automation tools for designing analog systems (including that of MP analog) hinders their adoption for designing large systems. The inherent scalability and modularity of MP systems present a unique opportunity in this regard. This paper introduces KALAM (toolKit for Automating high-Level synthesis of Analog computing systeMs), which leverages factor graphs as the foundational paradigm for synthesizing MP-based analog computing systems. Factor graphs are the basis of various signal processing tasks and, when coupled with MP, can be used to design scalable and energy-efficient analog signal processors. Using Python scripting language, the KALAM automation flow translates an input factor graph to its equivalent SPICE-compatible circuit netlist that can be used to validate the intended functionality. KALAM also allows the integration of design optimization strategies such as precision tuning, variable elimination, and mathematical simplification. We demonstrate KALAM's versatility for tasks such as Bayesian inference, Low-Density Parity Check (LDPC) decoding, and Artificial Neural Networks (ANN). Simulation results of the netlists align closely with software implementations, affirming the efficacy of our proposed automation tool., Comment: 5 Pages, 4 figures

This paper investigates the application of game-theoretic principles combined with advanced Kalman filtering techniques to enhance maritime target tracking systems. Specifically, the paper presents a two-player, imperfect information, non-cooperative, sequential game framework for optimal decision making for a tracker and an evader. The paper also investigates the effectiveness of this game-theoretic decision making framework by comparing it with single-objective optimisation methods based on minimising tracking uncertainty. Rather than modelling a zero-sum game between the tracker and the evader, which presupposes the availability of perfect information, in this paper we model both the tracker and the evader as playing separate zero-sum games at each time step with an internal (and imperfect) model of the other player. The study defines multi-faceted winning criteria for both tracker and evader, and computes winning percentages for both by simulating their interaction for a range of speed ratios. The results indicate that game theoretic decision making improves the win percentage of the tracker compared to traditional covariance minimization procedures in all cases, regardless of the speed ratios and the actions of the evader. In the case of the evader, we find that a simpler linear escape action is most effective for the evader in most scenarios. Overall, the results indicate that the presented sequential-game based decision making framework significantly improves win percentages for a player in scenarios where that player does not have inherent advantages in terms of starting position, speed ratio, or available time (to track / escape), highlighting that game theoretic decision making is particularly useful in scenarios where winning by using more traditional decision making procedures is highly unlikely.

This work analyses the interdependent link creation of patent and shareholding links in interfirm networks, and how this dynamics affects the resilience of such networks in the face of cascading failures. Using the Orbis dataset, we construct very large co-patenting and shareholding networks, globally as well as in terms of individual countries. Besides, we construct smaller overlap networks from those firm pairs which have both types of links between them, for nine years between 2008-2016. We use information theoretic measures, such as mutual information, active information storage, and transfer entropy, to characterise the topological similarities and shared topological information between the relevant co-patenting and shareholding networks. We then construct a cascading failure model, and use it to analyse the resilience of interdependent interfirm networks in terms of multiple failure characteristics. We find that there is relatively high level of mutual information between co-patenting networks and the shareholding networks from later years, suggesting that the formation of shareholding links is influenced by the existence of patent links in previous years. We also show that this influence becomes most apparent after a delay of four years between the formation of co-patenting links and shareholding links. Analysing the resilience of shareholding networks against cascading failures, we show that in terms of both mean downtime, and failure proportion of firms, certain countries have less resilient shareholding networks compared to other countries with significant economies. Based on our results, we postulate that an interfirm network model which considers multiple types of relationships together could be a to highlight important features of economic systems around the world., Comment: 17 pages; 19 figures

Dependencies among attributes are a common aspect of tabular data. However, whether existing tabular data generation algorithms preserve these dependencies while generating synthetic data is yet to be explored. In addition to the existing notion of functional dependencies, we introduce the notion of logical dependencies among the attributes in this article. Moreover, we provide a measure to quantify logical dependencies among attributes in tabular data. Utilizing this measure, we compare several state-of-the-art synthetic data generation algorithms and test their capability to preserve logical and functional dependencies on several publicly available datasets. We demonstrate that currently available synthetic tabular data generation algorithms do not fully preserve functional dependencies when they generate synthetic datasets. In addition, we also showed that some tabular synthetic data generation models can preserve inter-attribute logical dependencies. Our review and comparison of the state-of-the-art reveal research needs and opportunities to develop task-specific synthetic tabular data generation models., Comment: Submitted to Pattern Recognition Journal

Community health workers (CHWs) provide last-mile healthcare services but face challenges due to limited medical knowledge and training. This paper describes the design, deployment, and evaluation of ASHABot, an LLM-powered, experts-in-the-loop, WhatsApp-based chatbot to address the information needs of CHWs in India. Through interviews with CHWs and their supervisors and log analysis, we examine factors affecting their engagement with ASHABot, and ASHABot's role in addressing CHWs' informational needs. We found that ASHABot provided a private channel for CHWs to ask rudimentary and sensitive questions they hesitated to ask supervisors. CHWs trusted the information they received on ASHABot and treated it as an authoritative resource. CHWs' supervisors expanded their knowledge by contributing answers to questions ASHABot failed to answer, but were concerned about demands on their workload and increased accountability. We emphasize positioning LLMs as supplemental fallible resources within the community healthcare ecosystem, instead of as replacements for supervisor support.

This paper introduces a novel approach to design of functional H_\infty filters for a class of nonlinear descriptor systems subjected to disturbances. Departing from conventional assumptions regarding system regularity, we adopt a more inclusive approach by considering general descriptor systems that satisfy a rank condition on their coefficient matrices. Under this rank condition, we establish a linear matrix inequality (LMI) as a sufficient criterion ensuring the stability of the error system and constraining the L 2 gain of the mapping from disturbances to errors to a predetermined level. The efficacy of the proposed approach is demonstrated through a practical example involving a simple constrained mechanical system., Comment: 13 pages, 4 figures

Considered in this work is the initial value problem (IVP) associated to a higher order water wave model \begin{equation*} \begin{cases} \eta_t+\eta_x-\gamma_1 \eta_{xxt}+\gamma_2\eta_{xxx}+\delta_1 \eta_{xxxxt}+\delta_2\eta_{xxxxx}+\frac{3}{2}\eta \eta_x+\gamma (\eta^2)_{xxx}-\frac{7}{48}(\eta_x^2)_x-\frac{1}{8}(\eta^3)_x=0,\\ \eta(x,0) = \eta_0(x). \end{cases} \end{equation*} The main interest is in addressing the well-posedness issues of the IVP when the given initial data are considered in the modulation space $M_s^{2,p}(\mathbb{R})$ or the $L^p$-based Sobolev spaces $H^{s,p}(\mathbb{R})$, $1\leq p<\infty$. We derive some multilinear estimates in these spaces and prove that the above IVP is locally well-posed for data in $M_s^{2,p}(\mathbb{R})$ whenever $s>1$ and $p\geq 1$, and in $H^{s,p}(\mathbb{R})$ whenever $p\in [1,\infty)$ and $s\geq \max\left\{ \frac1{p}+\frac12, 1 \right\}$. We also use a combination of high-low frequency technique and an {\em a priori estimate}, and prove that the local solution with data in the modulation spaces $M_s^{2,p}(\mathbb{R})$ can be extended globally to the time interval $[0, T]$ for any given $T\gg1$ if $1\leq \frac32-\frac1p

Hydrodynamic turbulence exhibits nonequilibrium behaviour with $k^{-5/3}$ energy spectrum, and equilibrium behaviour with $k^{d-1}$ energy spectrum and zero viscosity, where $d$ is the space dimension. Using recursive renormalization group {in Craya-Herring basis}, we show that the nonequilibrium solution is valid only for $d < 6$, whereas equilibrium solution with zero viscosity is the only solution for $d>6$. Thus, $d=6$ is the critical dimension for hydrodynamic turbulence. In addition, we show that the energy flux changes sign from positive to negative near $d=2.15$. We also compute the energy flux and Kolmogorov's constants for various $d$'s, and observe that our results are in good agreement with past numerical results., Comment: Accepted in Phys. Rev. E. 15 pages. arXiv admin note: text overlap with arXiv:2309.05207

We consider the initial value problem (IVP) for the 2D generalized Zakharov-Kuznetsov (ZK) equation \begin{equation} \begin{cases} \partial_{t}u+\partial_{x}\Delta u+\mu \partial_{x}u^{k+1}=0, \,\;\; (x, y) \in \mathbb{R}^2, \, t \in \mathbb{R},\\ u(x,y,0)=u_0(x,y), \end{cases} \end{equation} where $\Delta=\partial_x^2+\partial_y^2$, $\mu=\pm 1$, $k=1,2$ and the initial data $u_0$ is real analytic in a strip around the $x$-axis of the complex plane and have radius of spatial analyticity $\sigma_0$. For both $k=1$ and $k=2$ we prove that there exists $T_0>0$ such that the radius of spatial analyticity of the solution remains the same in the time interval $[-T_0, T_0]$. We also consider the evolution of the radius of spatial analyticity when the local solution extends globally in time. For the Zakharov-Kuznetsov equation ($k=1$), we prove that, in both focusing ($\mu=1$) and defocusing ($\mu=-1$) cases, and for any $T> T_0$, the radius of analyticity cannot decay faster than $cT^{-4+\epsilon}$, $\epsilon>0$, $c>0$. For the modified Zakharov-Kuznetsov equation ($k=2)$ in the defocusing case ($\mu=-1$), we prove that the radius of spatial analyticity cannot decay faster than $cT^{-\frac{4}{3}}$, $c>0$, for any $T>T_0$. These results on the algebraic lower bounds for the evolution of the radius of analyticity improve the ones obtained by Shan and Zhang in [J. Math. Anal. Appl., 501 (2021) 125218] and by Quian and Shan in [Nonlinear Analysis, 235 (2023) 113344] where the authors have obtained lower bounds involving exponential decay., Comment: 34 pages. arXiv admin note: text overlap with arXiv:2308.08541

Generating synthetic samples from the convex space of the minority class is a popular oversampling approach for imbalanced classification problems. Recently, deep-learning approaches have been successfully applied to modeling the convex space of minority samples. Beyond oversampling, learning the convex space of neighborhoods in training data has not been used to generate entire tabular datasets. In this paper, we introduce a deep learning architecture (NextConvGeN) with a generator and discriminator component that can generate synthetic samples by learning to model the convex space of tabular data. The generator takes data neighborhoods as input and creates synthetic samples within the convex space of that neighborhood. Thereafter, the discriminator tries to classify these synthetic samples against a randomly sampled batch of data from the rest of the data space. We compared our proposed model with five state-of-the-art tabular generative models across ten publicly available datasets from the biomedical domain. Our analysis reveals that synthetic samples generated by NextConvGeN can better preserve classification and clustering performance across real and synthetic data than other synthetic data generation models. Synthetic data generation by deep learning of the convex space produces high scores for popular utility measures. We further compared how diverse synthetic data generation strategies perform in the privacy-utility spectrum and produced critical arguments on the necessity of high utility models. Our research on deep learning of the convex space of tabular data opens up opportunities in clinical research, machine learning model development, decision support systems, and clinical data sharing., Comment: 30 pages, 10 figures, submitted to Pattern Recognition journal

We introduce Meta 3D Gen (3DGen), a new state-of-the-art, fast pipeline for text-to-3D asset generation. 3DGen offers 3D asset creation with high prompt fidelity and high-quality 3D shapes and textures in under a minute. It supports physically-based rendering (PBR), necessary for 3D asset relighting in real-world applications. Additionally, 3DGen supports generative retexturing of previously generated (or artist-created) 3D shapes using additional textual inputs provided by the user. 3DGen integrates key technical components, Meta 3D AssetGen and Meta 3D TextureGen, that we developed for text-to-3D and text-to-texture generation, respectively. By combining their strengths, 3DGen represents 3D objects simultaneously in three ways: in view space, in volumetric space, and in UV (or texture) space. The integration of these two techniques achieves a win rate of 68% with respect to the single-stage model. We compare 3DGen to numerous industry baselines, and show that it outperforms them in terms of prompt fidelity and visual quality for complex textual prompts, while being significantly faster.

We present Meta 3D AssetGen (AssetGen), a significant advancement in text-to-3D generation which produces faithful, high-quality meshes with texture and material control. Compared to works that bake shading in the 3D object's appearance, AssetGen outputs physically-based rendering (PBR) materials, supporting realistic relighting. AssetGen generates first several views of the object with factored shaded and albedo appearance channels, and then reconstructs colours, metalness and roughness in 3D, using a deferred shading loss for efficient supervision. It also uses a sign-distance function to represent 3D shape more reliably and introduces a corresponding loss for direct shape supervision. This is implemented using fused kernels for high memory efficiency. After mesh extraction, a texture refinement transformer operating in UV space significantly improves sharpness and details. AssetGen achieves 17% improvement in Chamfer Distance and 40% in LPIPS over the best concurrent work for few-view reconstruction, and a human preference of 72% over the best industry competitors of comparable speed, including those that support PBR. Project page with generated assets: https://assetgen.github.io, Comment: Project Page: https://assetgen.github.io

In the domain growth process, small structures gradually vanish, leaving behind larger ones. We investigate spectral energy transfers in two standard models for domain growth: (a) the {\it Cahn-Hilliard} (CH) equation with conserved dynamics, and (b) the {\it time-dependent Ginzburg-Landau} (TDGL) equation with non-conserved dynamics. The nonlinear terms in these equations dissipate fluctuations and facilitate energy transfers among Fourier modes. In the TDGL equation, only the $\phi(\mathbf{k} = 0, t)$ mode survives, and the order parameter $\phi(\mathbf{r},t)$ approaches a uniform state with $\phi = +1$ or $-1$. On the other hand, there is no dynamics of the $\phi(\mathbf{k} = 0, t)$ mode in the CH equation due to the conservation law, highlighting the different dynamics of these equations.

Southeast Asia (SEA) is a region rich in linguistic diversity and cultural variety, with over 1,300 indigenous languages and a population of 671 million people. However, prevailing AI models suffer from a significant lack of representation of texts, images, and audio datasets from SEA, compromising the quality of AI models for SEA languages. Evaluating models for SEA languages is challenging due to the scarcity of high-quality datasets, compounded by the dominance of English training data, raising concerns about potential cultural misrepresentation. To address these challenges, we introduce SEACrowd, a collaborative initiative that consolidates a comprehensive resource hub that fills the resource gap by providing standardized corpora in nearly 1,000 SEA languages across three modalities. Through our SEACrowd benchmarks, we assess the quality of AI models on 36 indigenous languages across 13 tasks, offering valuable insights into the current AI landscape in SEA. Furthermore, we propose strategies to facilitate greater AI advancements, maximizing potential utility and resource equity for the future of AI in SEA., Comment: https://seacrowd.github.io/ Accepted in EMNLP 2024

We investigate the initial value problem (IVP) associated to the modified Korteweg-de Vries equation (mKdV) in the defocusing scenario: \begin{equation*} \left\{\begin{array}{l} \partial_t u+ \partial_x^3u-u^2\partial_x(u) = 0, \quad x,t\in\mathbb{R}, \\ u(x,0) = u_0(x), \end{array}\right. \end{equation*} where $u$ is a real valued function and the initial data $u_0$ is analytic on $\mathbb{R}$ and has uniform radius of analyticity $\sigma_0$ in the spatial variable. It is well-known that the solution $u$ preserves its analyticity with the same radius $\sigma_0$ for at least some time span $0

Analyzing high-dimensional count data is a challenge and statistical model-based approaches provide an adequate and efficient framework that preserves explainability. The (multivariate) Poisson-Log-Normal (PLN) model is one such model: it assumes count data are driven by an underlying structured latent Gaussian variable, so that the dependencies between counts solely stems from the latent dependencies. However PLN doesn't account for zero-inflation, a feature frequently observed in real-world datasets. Here we introduce the Zero-Inflated PLN (ZIPLN) model, adding a multivariate zero-inflated component to the model, as an additional Bernoulli latent variable. The Zero-Inflation can be fixed, site-specific, feature-specific or depends on covariates. We estimate model parameters using variational inference that scales up to datasets with a few thousands variables and compare two approximations: (i) independent Gaussian and Bernoulli variational distributions or (ii) Gaussian variational distribution conditioned on the Bernoulli one. The method is assessed on synthetic data and the efficiency of ZIPLN is established even when zero-inflation concerns up to $90\%$ of the observed counts. We then apply both ZIPLN and PLN to a cow microbiome dataset, containing $90.6\%$ of zeroes. Accounting for zero-inflation significantly increases log-likelihood and reduces dispersion in the latent space, thus leading to improved group discrimination., Comment: 27 pages including appendices. 8 figures, 1 table

Let $\F$ be a non-Archimedean local field.~Consider $\G_{n}:= \Sp_{2n}(\F)$ and let $\M:= \GL_l \times \G_{n-l}$ be a maximal Levi subgroup of $\G_{n}$.~This paper undertakes the computation of the Jacquet module of representations of $\G_{n}$ with respect to the maximal Levi subgroup, belonging to a particular class. Finally, we conclude that for a subclass of representations of $\G_{n},$ multiplicity of the Jacquet module does not exceed 2., Comment: 12 pages; comments are welcome. arXiv admin note: text overlap with arXiv:2405.05719

Let F be a non-Archimedean local field. Consider G_n:= GL_n(F) and let M:= G_l * G_{n-l} be a maximal Levi subgroup of G_n. In this article, we compute the semisimplified Jacquet module of representations of G_n with respect to the maximal Levi subgroup M, belonging to a particular category of representations. Utilizing our results, we prove that the Jacquet module is multiplicity-free for a specific subcategory of representations. Our findings are based on the Zelevinsky classification., Comment: 10 pages, comments are welcome

Let $D$ denote a quaternion division algebra over a non-archimedean local field $F$ with characteristic zero. Let $Sp_n(D)$ be the unique non-split inner form of the symplectic group $Sp_{2n}(F)$. An irreducible admissible representation $(\pi, V)$ of $GL_{n}(D)$ is said to have a symplectic model (or said to be $Sp_n(D)$-distinguished) if there exists a linear functional $\phi$ on $V$ such that $\phi(\pi(h)v) = \phi(v)$ for all $v \in V$ and $h \in Sp_n(D)$. This article classifies those ladder representations of $GL_n(D)$ that possess a symplectic model (i.e., those representations that are $Sp_n(D)$-distinguished). Recently, Prasad conjectured that non-supercuspidal discrete series representations of $GL_n(D)$ do not admit a symplectic model. We confirm this for the Steinberg representations, which serve as canonical examples of discrete series representations. Furthermore, we demonstrate the hereditary nature of the symplectic model for induced representations derived from finite-length representations. In addition, we prove a part of Prasad's conjecture, which provides a family of irreducible unitary representations, all equipped with a symplectic model., Comment: 22 pages, comments are welcome

Human papillomavirus (HPV) infection is the most common sexually transmitted infection in the world. Persistent oncogenic Human papillomavirus infection has been a leading threat to global health and can lead to serious complications such as cervical cancer. Prevention interventions including vaccination and screening have been proved effective in reducing the risk of HPV-related diseases. In recent decades, computational epidemiology has been serving as a very useful tool to study HPV transmission dynamics and evaluation of prevention strategies. In this paper, we conduct a comprehensive literature review on state-of-the-art computational epidemic models for HPV disease dynamics, transmission dynamics, as well as prevention efforts. We summarise current research trends, identify gaps in the present literature, and identify future research directions with potential in accelerating the containment and/or elimination of HPV infection.

Large language models (LLMs) show remarkable human-like capability in various domains and languages. However, a notable quality gap arises in low-resource languages, e.g., Indonesian indigenous languages, rendering them ineffective and inefficient in such linguistic contexts. To bridge this quality gap, we introduce Cendol, a collection of Indonesian LLMs encompassing both decoder-only and encoder-decoder architectures across a range of model sizes. We highlight Cendol's effectiveness across a diverse array of tasks, attaining 20% improvement, and demonstrate its capability to generalize to unseen tasks and indigenous languages of Indonesia. Furthermore, Cendol models showcase improved human favorability despite their limitations in capturing indigenous knowledge and cultural values in Indonesia. In addition, we discuss the shortcomings of parameter-efficient tunings, such as LoRA, for language adaptation. Alternatively, we propose the usage of vocabulary adaptation to enhance efficiency. Lastly, we evaluate the safety of Cendol and showcase that safety in pre-training in one language such as English is transferable to low-resource languages, such as Indonesian, even without RLHF and safety fine-tuning., Comment: Cendol models are released under Apache 2.0 license and will be made publicly available soon

Mechanical ventilation is a critical life-support intervention that uses a machine to deliver controlled air and oxygen to a patient's lungs, assisting or replacing spontaneous breathing. While several data-driven approaches have been proposed to optimize ventilator control strategies, they often lack interpretability and agreement with general domain knowledge. This paper proposes a methodology for interpretable reinforcement learning (RL) using decision trees for mechanical ventilation control. Using a causal, nonparametric model-based off-policy evaluation, we evaluate the policies in their ability to gain increases in SpO2 while avoiding aggressive ventilator settings which are known to cause ventilator induced lung injuries and other complications. Numerical experiments using MIMIC-III data on the stays of real patients' intensive care unit stays demonstrate that the decision tree policy outperforms the behavior cloning policy and is comparable to state-of-the-art RL policy. Future work concerns better aligning the cost function with medical objectives to generate deeper clinical insights.

Although commonsense reasoning is greatly shaped by cultural and geographical factors, previous studies have predominantly centered on cultures grounded in the English language, potentially resulting in an Anglocentric bias. In this paper, we introduce IndoCulture, aimed at understanding the influence of geographical factors on language model reasoning ability, with a specific emphasis on the diverse cultures found within eleven Indonesian provinces. In contrast to prior work that has relied on templates (Yin et al., 2022) and online scrapping (Fung et al., 2024), we create IndoCulture by asking local people to manually develop a cultural context and plausible options, across a set of predefined topics. Evaluation of 27 language models reveals several insights: (1) the open-weight Llama-3 is competitive with GPT-4, while other open-weight models struggle, with accuracies below 50%; (2) there is a general pattern of models generally performing better for some provinces, such as Bali and West Java, and less well for others; and (3) the inclusion of location context enhances performance, especially for larger models like GPT-4, emphasizing the significance of geographical context in commonsense reasoning., Comment: Accepted at TACL