Search

Given the increasing use of synthetic data in language model (LM) post-training, an LM's ability to generate high-quality data has become nearly as crucial as its ability to solve problems directly. While prior works have focused on developing effective data generation methods, they lack systematic comparison of different LMs as data generators in a unified setting. To address this gap, we propose AgoraBench, a benchmark that provides standardized settings and metrics to evaluate LMs' data generation abilities. Through synthesizing 1.26 million training instances using 6 LMs and training 99 student models, we uncover key insights about LMs' data generation capabilities. First, we observe that LMs exhibit distinct strengths. For instance, GPT-4o excels at generating new problems, while Claude-3.5-Sonnet performs better at enhancing existing ones. Furthermore, our analysis reveals that an LM's data generation ability doesn't necessarily correlate with its problem-solving ability. Instead, multiple intrinsic features of data quality-including response quality, perplexity, and instruction difficulty-collectively serve as better indicators. Finally, we demonstrate that strategic choices in output format and cost-conscious model selection significantly impact data generation effectiveness., Comment: Work in Progress

For decades, corporations and governments have relied on scanned documents to record vast amounts of information. However, extracting this information is a slow and tedious process due to the sheer volume and complexity of these records. The rise of Vision Language Models (VLMs) presents a way to efficiently and accurately extract the information out of these documents. The current automated workflow often requires a two-step approach involving the extraction of information using optical character recognition software and subsequent usage of large language models for processing this information. Unfortunately, these methods encounter significant challenges when dealing with noisy scanned documents, often requiring computationally expensive language models to handle high information density effectively. In this study, we propose PatchFinder, an algorithm that builds upon VLMs to improve information extraction. First, we devise a confidence-based score, called Patch Confidence, based on the Maximum Softmax Probability of the VLMs' output to measure the model's confidence in its predictions. Using this metric, PatchFinder determines a suitable patch size, partitions the input document into overlapping patches, and generates confidence-based predictions for the target information. Our experimental results show that PatchFinder, leveraging Phi-3v, a 4.2-billion-parameter VLM, achieves an accuracy of 94% on our dataset of 190 noisy scanned documents, outperforming ChatGPT-4o by 18.5 percentage points., Comment: This paper has been accepted to IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2025

The properties of dilute neutron matter are mostly determined by the s-wave two-body (2N) interaction, while three-body (3N) interactions are suppressed by the Pauli principle. In a previous work, we showed that it can be advantageous to use the renormalization group based effective interactions with cutoffs scaled with the Fermi momentum, especially at low densities. In that case, induced 3N interactions may become important. In this work, we compute the 3N interaction induced by the similarity renormalization group flow of the s-wave 2N interaction. We work in the momentum-space hyperspherical partial wave basis and investigate its convergence properties. Then we study the effect of the induced 3N interaction on the equation of state of dilute neutron matter. We observe that the cutoff dependence of the equation of state is strongly reduced when the effect of induced 3N interaction is included., Comment: 11 pages, 10 figures

The ability of robotic grippers to not only grasp but also re-position and re-orient objects in-hand is crucial for achieving versatile, general-purpose manipulation. While recent advances in soft robotic grasping has greatly improved grasp quality and stability, their manipulation capabilities remain under-explored. This paper presents the DexGrip, a multi-modal soft robotic gripper for in-hand grasping, re-orientation and manipulation. DexGrip features a 3 Degrees of Freedom (DoFs) active suction palm and 3 active (rotating) grasping surfaces, enabling soft, stable, and dexterous grasping and manipulation without ever needing to re-grasp an object. Uniquely, these features enable complete 360 degree rotation in all three principal axes. We experimentally demonstrate these capabilities across a diverse set of objects and tasks. DexGrip successfully grasped, re-positioned, and re-oriented objects with widely varying stiffnesses, sizes, weights, and surface textures; and effectively manipulated objects that presented significant challenges for existing robotic grippers., Comment: 6 pages, 5 figures

A fundamental problem in decentralized networked systems is to coordinate actions of different agents so that they reach a state of agreement. In such applications, it is additionally desirable that the actions at various nodes may not be anticipated by malicious eavesdroppers. Motivated by this, we investigate the problem of secure multi-terminal strong coordination aided by a multiple-access wiretap channel. In this setup, independent and identically distributed copies of correlated sources are observed by two transmitters who encode the channel inputs to the MAC-WT. The legitimate receiver observing the channel output and side information correlated with the sources must produce approximately i.i.d. copies of an output variable jointly distributed with the sources. Furthermore, we demand that an external eavesdropper learns essentially nothin g about the sources and the simulated output sequence by observing its own MAC-WT output. This setting is aided by the presence of independent pairwise shared randomness between each encoder and the legitimate decoder, that is unavailable to the eavesdropper. We derive an achievable rate region based on a combination of coordination coding and wiretap coding, along with an outer bound. The inner bound is shown to be tight and a complete characterization is derived for the special case when the sources are conditionally independent given the decoder side information and the legitimate channel is composed of deterministic links. Further, we also analyze a more general scenario with possible encoder cooperation, where one of the encoders can non-causally crib from the other encoders input, for which an achievable rate region is proposed. We then explicitly compute the rate regions for an example both with and without cribbing between the encoders, and demonstrate that cribbing strictly improves upon the achievable rate region., Comment: Extended version of ISIT 2024 paper

Observational studies are identifying stars thought to be remnants from the earliest stages of the hierarchical mass assembly of the Milky Way, referred to as the proto-Galaxy. We use red giant stars with kinematics and [$\alpha$/M] and [M/H] estimates from Gaia DR3 data to investigate the relationship between azimuthal velocity and metallicity, aiming to understand the transition from a chaotic proto-Galaxy to a well-ordered, rotating (old) disc-like population. To analyse the structure of the data in [M/H]-v$_\phi$ space for both high- and low-$\alpha$ samples with carefully defined $\alpha$-separation, we develop a model with two Gaussian components in v$_\phi$: one representing a disc-like population and the other a halo-like population. This model is designed to capture the conditional distribution P(v$_\phi$ $\mid$ [M/H]) with a 2-component Gaussian Mixture Model with fixed azimuthal velocities means and standard deviations. To quantify the spin-up of the high-$\alpha$ disc population, we extend this two-component model by allowing the mean velocity and velocity dispersion to vary between the spline knots across the metallicity range used. We also compare our findings with existing literature using traditional Gaussian Mixture Modelling in bins of [M/H] and investigate using orbital circularity instead of azimuthal velocity. Our findings show that the metal-poor high-$\alpha$ disc gradually spins up across [M/H] $\sim$ -1.7 to -1.0, while the low-$\alpha$ sample exhibits a sharp transition at [M/H] $\sim$ -1.0. This latter result is due to the accreted debris dominating the metal-poor end, underscoring the critical role of [$\alpha$/M] selection in studying the (old) disc evolution of the Milky Way. These results indicate that the proto-Galaxy underwent a slow, monotonic spin-up phase rather than a rapid, dramatic spin-up at [M/H] $\sim$ -1.0, as previously inferred., Comment: Submitted to A&A. 19 pages + 11 figures, 6 pages + 5 figures in appendix. Comments are very welcome!

Understanding material failure is critical for designing stronger and lighter structures by identifying weaknesses that could be mitigated. Existing full-physics numerical simulation techniques involve trade-offs between speed, accuracy, and the ability to handle complex features like varying boundary conditions, grid types, resolution, and physical models. We present the first foundation model specifically designed for predicting material failure, leveraging large-scale datasets and a high parameter count (up to 3B) to significantly improve the accuracy of failure predictions. In addition, a large language model provides rich context embeddings, enabling our model to make predictions across a diverse range of conditions. Unlike traditional machine learning models, which are often tailored to specific systems or limited to narrow simulation conditions, our foundation model is designed to generalize across different materials and simulators. This flexibility enables the model to handle a range of material properties and conditions, providing accurate predictions without the need for retraining or adjustments for each specific case. Our model is capable of accommodating diverse input formats, such as images and varying simulation conditions, and producing a range of outputs, from simulation results to effective properties. It supports both Cartesian and unstructured grids, with design choices that allow for seamless updates and extensions as new data and requirements emerge. Our results show that increasing the scale of the model leads to significant performance gains (loss scales as $N^{-1.6}$, compared to language models which often scale as $N^{-0.5}$)., Comment: Accepted at NeurIPS 2024 "Foundation Models for Science: Progress, Opportunities, and Challenges" Workshop

Parenting brings emotional and physical challenges, from balancing work, childcare, and finances to coping with exhaustion and limited personal time. Yet, one in three parents never seek support. AI systems potentially offer stigma-free, accessible, and affordable solutions. Yet, user adoption often fails due to issues with explainability and reliability. To see if these issues could be solved using a co-design approach, we developed and tested NurtureBot, a wellbeing support assistant for new parents. 32 parents co-designed the system through Asynchronous Remote Communities method, identifying the key challenge as achieving a "successful chat". Aspart of co-design, parents role-played as NurturBot, rewriting its dialogues to improve user understanding, control, and outcomes. The refined prototype evaluated by 32 initial and 46 new parents, showed improved user experience and usability, with final CUQ score of 91.3/100, demonstrating successful interaction patterns. Our process revealed useful interaction design lessons for effective AI parenting support.

The classical `decision problem' asks whether a given formula of first-order logic is satisfiable. In this work we consider an extension of this problem to regular first-order theories, i.e. (infinite) regular sets of formulae. Building on the beautiful classification of syntactic classes as decidable or undecidable for the classical decision problem, we show that some classes (the EPR and Gurevich classes) which are decidable in the classical setting are undecidable for regular theories; on the other hand for each we show a subclass which remains decidable in our setting, leaving a complete classification as a challenge for future work. Finally, we observe that our problem generalises prior work on verification of uninterpreted programs, and give a semantic class of existential formulae for which the problem is decidable.

Understanding the decision-making processes of large language models (LLMs) is critical given their widespread applications. Towards this goal, describing the topological and geometrical properties of internal representations has recently provided valuable insights. For a more comprehensive characterization of these inherently complex spaces, we present a novel framework based on zigzag persistence, a method in topological data analysis (TDA) well-suited for describing data undergoing dynamic transformations across layers. Within this framework, we introduce persistence similarity, a new metric that quantifies the persistence and transformation of topological features such as $p$-cycles throughout the model layers. Unlike traditional similarity measures, our approach captures the entire evolutionary trajectory of these features, providing deeper insights into the internal workings of LLMs. As a practical application, we leverage persistence similarity to identify and prune redundant layers, demonstrating comparable performance to state-of-the-art methods across several benchmark datasets. Additionally, our analysis reveals consistent topological behaviors across various models and hyperparameter settings, suggesting a universal structure in LLM internal representations., Comment: 10+6 pages, 7 figures, 1 table. All comments welcome!

In this work, we present an autonomous inspection framework for remote sensing tasks in active open-pit mines. Specifically, the contributions are focused towards developing a methodology where an initial approximate operator-defined inspection plan is exploited by an online view-planner to predict an inspection path that can adapt to changes in the current mine-face morphology caused by route mining activities. The proposed inspection framework leverages instantaneous 3D LiDAR and localization measurements coupled with modelled sensor footprint for view-planning satisfying desired viewing and photogrammetric conditions. The efficacy of the proposed framework has been demonstrated through simulation in Feiring-Bruk open-pit mine environment and hardware-based outdoor experimental trials. The video showcasing the performance of the proposed work can be found here: https://youtu.be/uWWbDfoBvFc, Comment: Accepted for publication in IEEE ROBIO 2024

The rapid evolution of Large Language Models (LLMs) has transformed them from basic conversational tools into sophisticated entities capable of complex reasoning and decision-making. These advancements have led to the development of specialized LLM-based agents designed for diverse tasks such as coding and web browsing. As these agents become more capable, the need for a robust framework that facilitates global communication and collaboration among them towards advanced objectives has become increasingly critical. Distributed Agents in a Worldwide Network (DAWN) addresses this need by offering a versatile framework that integrates LLM-based agents with traditional software systems, enabling the creation of agentic applications suited for a wide range of use cases. DAWN enables distributed agents worldwide to register and be easily discovered through Gateway Agents. Collaborations among these agents are coordinated by a Principal Agent equipped with reasoning strategies. DAWN offers three operational modes: No-LLM Mode for deterministic tasks, Copilot for augmented decision-making, and LLM Agent for autonomous operations. Additionally, DAWN ensures the safety and security of agent collaborations globally through a dedicated safety, security, and compliance layer, protecting the network against attackers and adhering to stringent security and compliance standards. These features make DAWN a robust network for deploying agent-based applications across various industries.

When selecting applicants for scholarships, universities, or jobs, practitioners often aim for a diverse cohort of qualified recipients. However, differing articulations, constructs, and notions of diversity prevents decision-makers from operationalising and progressing towards the diversity they all agree is needed. To understand this challenge of translation from values, to requirements, to decision support tools (DSTs), we conducted participatory design studies exploring professionals' varied perceptions of diversity and how to build for them. Our results suggest three definitions of diversity: bringing together different perspectives; ensuring representativeness of a base population; and contextualising applications, which we use to create the Diversity Triangle. We experience-prototyped DSTs reflecting each angle of the Diversity Triangle to enhance decision-making around diversity. We find that notions of diversity are highly diverse; efforts to design DSTs for diversity should start by working with organisations to distil 'diversity' into definitions and design requirements., Comment: 32 pages, 11 figures

The context window of large language models (LLMs) has been extended significantly in recent years. However, while the context length that the LLM can process has grown, the capability of the model to accurately reason over that context degrades noticeably. This occurs because modern LLMs often become overwhelmed by the vast amount of information in the context; when answering questions, the model must identify and reason over relevant evidence sparsely distributed throughout the text. To alleviate the challenge of long-context reasoning, we develop a retrieve-then-reason framework, enabling LLMs to reason over relevant evidence collected during an intermediate retrieval step. We find that modern LLMs struggle to accurately retrieve relevant facts and instead, often hallucinate "retrieved facts", resulting in flawed reasoning and the production of incorrect answers. To address these issues, we introduce ALR$^2$, a method that augments the long-context reasoning capability of LLMs via an explicit two-stage procedure, i.e., aligning LLMs with the objectives of both retrieval and reasoning. We demonstrate the efficacy of ALR$^2$ for mitigating performance degradation in long-context reasoning tasks. Through extensive experiments on long-context QA benchmarks, we find our method to outperform competitive baselines by large margins, achieving at least 8.4 and 7.9 EM gains on the long-context versions of HotpotQA and SQuAD datasets, respectively.

Electrolytes mediate interactions between the cathode and anode and determine performance characteristics of batteries. Mixtures of multiple solvents are often used in electrolytes to achieve desired properties, such as viscosity, dielectric constant, boiling point, and melting point. Conventionally, multi-component electrolyte properties are approximated with linear mixing, but in practice, significant deviations are observed. Excess quantities can provide insights into the molecular behavior of the mixture and could form the basis for designing high-performance electrolytes. Here we investigate the excess density of commonly used Li-ion battery solvents such as cyclic carbonates, linear carbonates, ethers, and nitriles with molecular dynamics simulations. We additionally investigate electrolytes consisting of these solvents and a salt. The results smoothly vary with mole percent and are fit to permutation-invariant Redlich-Kister polynomials. Mixtures of similar solvents, such as cyclic-cyclic carbonate mixtures, tend to have excess properties that are lower in magnitude compared to mixtures of dissimilar substances, such as carbonate-nitrile mixtures. We perform experimental testing using our robotic test stand, Clio, to provide validation to the observed simulation trends. We quantify the structure similarity using SOAP fingerprints to create a descriptor for excess density, enabling the design of electrolyte properties. To a first approximation, this will allow us to estimate the deviation of a mixture from ideal behavior based solely upon the structural dissimilarity of the components.

The superposition of complex optical fields in three-dimension (3D) is the basis of several non-trivial wave phenomena. Significant among them are the non-uniform polarization distribution and their topological character, leading to the emergence of transverse spin, transverse momentum, and their dynamics. These aspects are experimentally measured in the nonparaxial focal region of a circularly-polarized Gaussian input beam. A dielectric mirror, kept in the focal region, is axially scanned to obtain the phase and polarization variations in the retro-reflected output beam using an interferometer and spatially-resolved Stokes parameter measurements. The identification of phase and polarization singularities in the beam cross-section and their behaviour as a function of the mirror position enabled us to map and study the phase-polarization variations in the nonparaxial focal region. The lemon-monstar type polarization patterns surrounding the C-point singularity in the output beam are identified and tracked to study the transverse spin dynamics for right- and left- circular polarization of the input beam. Direct measurement of the input beam polarization helicity-independent and helicity-dependent aspects of the transverse and longitudinal spin dynamics in the nonparaxial focal region are the significant findings reported here. The proposed and demonstrated measurement method allows us to investigate the nonparaxial focal region in more detail and has the potential to unravel other intricate optical field effects., Comment: 22 pages including Supporting Information

The spin-orbit interaction in a focused-reflected beam of light results in spatially non-uniform polarization in the beam cross-section due to the superposition of orthogonal field components and polarization-dependent interface reflection coefficients. Polarization filtering the output beam leads to an interchangeable transformation of - or + 2 charge vortex into two - or + unit charge vortices, for + or - circular polarization of the input Gaussian beam. This transformation follows a trajectory, named optical vortex trajectory, that depend on the input beam spin and hence the vortex charge and reflecting surface characteristics. The OVT is used here to quantify both the sign and the magnitude of the chiral parameter of a quartz crystal. The Jones matrix-based simulation anticipates the chirality-dependent OVT that matches with experimental measurements., Comment: 5 pages, 4 figures

Molecular Foundation Models are emerging as powerful tools for accelerating molecular design, material science, and cheminformatics, leveraging transformer architectures to speed up the discovery of new materials and drugs while reducing the computational cost of traditional ab initio methods. However, current models are constrained by closed-vocabulary tokenizers that fail to capture the full diversity of molecular structures. In this work, we systematically evaluate thirteen chemistry-specific tokenizers for their coverage of the SMILES language, uncovering substantial gaps. Using N-gram language models, we accessed the impact of tokenizer choice on model performance and quantified the information loss of unknown tokens. We introduce two new tokenizers, smirk and smirk-gpe, which can represent the entirety of the OpenSMILES specification while avoiding the pitfalls of existing tokenizers. Our work highlights the importance of open-vocabulary modeling for molecular foundation models and the need for chemically diverse benchmarks for cheminformatics., Comment: 26 pages, 6 figures

Context: In the context of Galactic archaeology, the outer halo remains relatively unexplored with respect to its metallicity distribution, merger debris, and the abundance of known very/extremely metal-poor ([Fe/H]<-2.5) stars. Aim: We utilize the Pristine survey's publicly available, Pristine data release 1 (PDR1) and Pristine-Gaia synthetic (PGS) catalogues of photometric metallicities, to select Red Giant Branch (RGB) stars in the outer halo. Methods: The RGB selection pipeline selects giants based on the absence of a well-measured parallax in the brightness range where dwarfs have reasonable parallax estimate from Gaia DR3 data. The photometric distances are calculated using a BaSTI-isochrone fitting code and the photometric metallicities. Results: Photometric distances derived from PDR1- and PGS-giants show typical uncertainties of 12% and a scatter of up to 20% and 40% respectively, when validated against inverted-parallax and Starhorse-code distances. The PDR1-giants catalogue provides a low-to-no bias view of the metallicity structure versus distance compared to the PGS-giants catalogue (with a distance-metallicity selection bias), while the PGS-giants catalogue provides an all-sky view of the outer halo, especially in the VMP end. The PDR1-giants catalogue is used to study the metallicity distribution function of the halo out to ~100 kpc, using Gaussian Mixture Model decomposition. Additionally, we use the PDR1-giants with radial velocities from spectroscopic surveys to map the metallicity view of the integrals-of-motion space where accreted dwarf galaxy debris conserve their orbital parameters for a long time. Using the PGS-giants catalogue, we associate 41 stars tentatively to the stellar counterpart of the Magellanic stream in the VMP end, out to 70 kpc. Conclusions: We publish two RGB catalogues between -4.0<[Fe/H]<+0.1 with reliable photometric distances and metallicities., Comment: Submitted to A&A, 25 pages (19 Figures) + 4 pages (6 Figures) in Appendix. Comments are very welcome, enjoy reading! The catalogues will be published upon acceptance

With recent rapid growth in online shopping, AI-powered Engagement Surfaces (ES) have become ubiquitous across retail services. These engagement surfaces perform an increasing range of functions, including recommending new products for purchase, reminding customers of their orders and providing delivery notifications. Understanding the causal effect of engagement surfaces on value driven for customers and businesses remains an open scientific question. In this paper, we develop a dynamic causal model at scale to disentangle value attributable to an ES, and to assess its effectiveness. We demonstrate the application of this model to inform business decision-making by understanding returns on investment in the ES, and identifying product lines and features where the ES adds the most value., Comment: 10 pages, 5 figures. Accepted at Applied Data Science track of KDD 2024, Barcelona, Spain

Embedding Based Retrieval (EBR) is a crucial component of the retrieval stage in (Ads) Recommendation System that utilizes Two Tower or Siamese Networks to learn embeddings for both users and items (ads). It then employs an Approximate Nearest Neighbor Search (ANN) to efficiently retrieve the most relevant ads for a specific user. Despite the recent rise to popularity in the industry, they have a couple of limitations. Firstly, Two Tower model architecture uses a single dot product interaction which despite their efficiency fail to capture the data distribution in practice. Secondly, the centroid representation and cluster assignment, which are components of ANN, occur after the training process has been completed. As a result, they do not take into account the optimization criteria used for retrieval model. In this paper, we present Hierarchical Structured Neural Network (HSNN), a deployed jointly optimized hierarchical clustering and neural network model that can take advantage of sophisticated interactions and model architectures that are more common in the ranking stages while maintaining a sub-linear inference cost. We achieve 6.5% improvement in offline evaluation and also demonstrate 1.22% online gains through A/B experiments. HSNN has been successfully deployed into the Ads Recommendation system and is currently handling major portion of the traffic. The paper shares our experience in developing this system, dealing with challenges like freshness, volatility, cold start recommendations, cluster collapse and lessons deploying the model in a large scale retrieval production system., Comment: Major rewrite of paper in progress

Safe, all-solid-state lithium metal batteries enable high energy density applications, but suffer from instabilities during operation that lead to rough interfaces between the metal and electrolyte and subsequently cause void formation and dendrite growth that degrades performance and safety. Inspired by the morphogenetic control of thin lamina such as tree leaves that robustly grow into flat shapes -- we propose a range of approaches to control lithium metal stripping and plating. To guide discovery of materials that will implement these feedback mechanisms, we develop a reduced order model that captures couplings between mechanics, interface growth, temperature, and electrochemical variables. We find that long-range feedback is required to achieve true interface stability, while approaches based on local feedback always eventually grow into rough interfaces. All together, our study provides the beginning of a practical framework for analyzing and designing stable electrochemical interfaces in terms of the mechanical properties and the physical chemistry that underlie their dynamics.

The SSPACE Astrobiology Payload (SAP) series, starting with the SAP-1 project is designed to conduct in-situ microbiology experiments in low earth orbit. This payload series aims to understand the behaviour of microbial organisms in space, particularly those critical for human health, and the corresponding effects due to microgravity and solar/galactic radiation. SAP-1 focuses on studying Bacillus clausii and Bacillus coagulans, bacteria beneficial to humans. It aims to provide a space laboratory for astrobiology experiments under microgravity conditions. The hardware developed for these experiments is indigenous and tailored to meet the unique requirements of autonomous microbiology experiments by controlling pressure, temperature, and nutrition flow to bacteria. A rotating platform, which forms the core design, is innovatively utilised to regulate the flow and mixing of nutrients with dormant bacteria. The technology demonstration models developed at SSPACE have yielded promising results, with ongoing efforts to refine, adapt for space conditions, and prepare for integration with nanosatellites or space modules. The anticipated payload will be compact, approximately 1U in size (10cm x 10cm x 10cm), consume less than 5W power, and offer flexibility for various microbiological studies., Comment: 20 Pages, Submitted to Advances in Space Research

This work establishes links between the Ising model and elliptic curves via Mahler measures. First, we reformulate the partition function of the Ising model on the square, triangular and honeycomb lattices in terms of the Mahler measure of a Laurent polynomial whose variety's projective closure defines an elliptic curve. Next, we obtain hypergeometric formulas for the partition functions on the triangular and honeycomb lattices and review the known series for the square lattice. Finally, at specific temperatures we express the free energy in terms of a Hasse-Weil $L$-function of an elliptic curve. At the critical point of the phase transition on all three lattices, we obtain the free energy more simply in terms of a Dirichlet $L$-function. These findings suggest that the connection between statistical mechanics and analytic number theory may run deeper than previously believed., Comment: Preprint submitted to Physical Review E

This study aims to assess the potential of the upcoming PLATO mission to investigate exoplanet populations around stars in diverse Galactic environments, specifically focusing on the Milky Way thin disk, thick disk, and stellar halo. We aim to quantify PLATOs ability to detect planets in each environment and determine how these observations could constrain planet formation models. Beginning from the all-sky PLATO Input Catalog, we kinematically classify the 2.4 million FGK stars into their respective Galactic components. For the sub-sample of stars in the long-observation LOPS2 and LOPN1 PLATO fields, we estimate planet occurrence rates using the New Generation Planet Population Synthesis (NGPPS) dataset. Combining these estimates with a PLATO detection efficiency model, we predicted the expected planet yields for each Galactic environment during a nominal 2+2 year mission. Based on our analysis, PLATO is likely to detect at least 400 exoplanets around the alpha-enriched thick disk stars. The majority of those planets are expected to be Super-Earths and Sub-Neptunes with radii between 2 and 10 Earth radii and orbital periods between 2 and 50 days, ideal for studying the link between the radius valley and stellar chemistry. For the metal-poor halo, PLATO is likely to detect between 1 and 80 planets with periods between 10 and 50 days, depending on the potential existence of a metallicity threshold for planet formation. The PLATO fields contain more than 3,400 potential target stars with [Fe/H] < -0.6, which will help to improve our understanding of planets around metal-poor stars. We identify a specific target list of 47 (kinematically classified) halo stars in the high-priority, high-SNR PLATO P1 sample, offering prime opportunities in the search for planets in metal-poor environments., Comment: 20 pages, 15 figures, accepted in A&A, code can be found at https://github.com/ChrisBoettner/plato, catalogue at https://doi.org/10.5281/zenodo.11428968

Large language models (LLMs) hold the promise of solving diverse tasks when provided with appropriate natural language prompts. However, prompting often leads models to make predictions with lower accuracy compared to finetuning a model with ample training data. On the other hand, while finetuning LLMs on task-specific data generally improves their performance, abundant annotated datasets are not available for all tasks. Previous work has explored generating task-specific data from state-of-the-art LLMs and using this data to finetune smaller models, but this approach requires access to a language model other than the one being trained, which introduces cost, scalability challenges, and legal hurdles associated with continuously relying on more powerful LLMs. In response to these, we propose SELF-GUIDE, a multi-stage mechanism in which we synthesize task-specific input-output pairs from the student LLM, then use these input-output pairs to finetune the student LLM itself. In our empirical evaluation of the Natural Instructions V2 benchmark, we find that SELF-GUIDE improves the performance of LLM by a substantial margin. Specifically, we report an absolute improvement of approximately 15% for classification tasks and 18% for generation tasks in the benchmark's metrics. This sheds light on the promise of self-synthesized data guiding LLMs towards becoming task-specific experts without any external learning signals., Comment: Accepted by COLM 2024

Pediatric central nervous system tumors are the leading cause of cancer-related deaths in children. The five-year survival rate for high-grade glioma in children is less than 20%. The development of new treatments is dependent upon multi-institutional collaborative clinical trials requiring reproducible and accurate centralized response assessment. We present the results of the BraTS-PEDs 2023 challenge, the first Brain Tumor Segmentation (BraTS) challenge focused on pediatric brain tumors. This challenge utilized data acquired from multiple international consortia dedicated to pediatric neuro-oncology and clinical trials. BraTS-PEDs 2023 aimed to evaluate volumetric segmentation algorithms for pediatric brain gliomas from magnetic resonance imaging using standardized quantitative performance evaluation metrics employed across the BraTS 2023 challenges. The top-performing AI approaches for pediatric tumor analysis included ensembles of nnU-Net and Swin UNETR, Auto3DSeg, or nnU-Net with a self-supervised framework. The BraTSPEDs 2023 challenge fostered collaboration between clinicians (neuro-oncologists, neuroradiologists) and AI/imaging scientists, promoting faster data sharing and the development of automated volumetric analysis techniques. These advancements could significantly benefit clinical trials and improve the care of children with brain tumors.

Modelling complex deformation for soft robotics provides a guideline to understand their behaviour, leading to safe interaction with the environment. However, building a surrogate model with high accuracy and fast inference speed can be challenging for soft robotics due to the nonlinearity from complex geometry, large deformation, material nonlinearity etc. The reality gap from surrogate models also prevents their further deployment in the soft robotics domain. In this study, we proposed a physics-informed Neural Networks (PINNs) named PINN-Ray to model complex deformation for a Fin Ray soft robotic gripper, which embeds the minimum potential energy principle from elastic mechanics and additional high-fidelity experimental data into the loss function of neural network for training. This method is significant in terms of its generalisation to complex geometry and robust to data scarcity as compared to other data-driven neural networks. Furthermore, it has been extensively evaluated to model the deformation of the Fin Ray finger under external actuation. PINN-Ray demonstrates improved accuracy as compared with Finite element modelling (FEM) after applying the data assimilation scheme to treat the sim-to-real gap. Additionally, we introduced our automated framework to design, fabricate soft robotic fingers, and characterise their deformation by visual tracking, which provides a guideline for the fast prototype of soft robotics.

Semantic communication, when examined through the lens of joint source-channel coding (JSCC), maps source messages directly into channel input symbols, where the measure of success is defined by end-to-end distortion rather than traditional metrics such as block error rate. Previous studies have shown significant improvements achieved through deep learning (DL)-driven JSCC compared to traditional separate source and channel coding. However, JSCC is impractical in existing communication networks, where application and network providers are typically different entities connected over general-purpose TCP/IP links. In this paper, we propose designing the source and channel mappings separately and sequentially via a novel multi-level reliability interface. This conceptual interface enables semi-JSCC at both the learned source and channel mappers and achieves many of the gains observed in existing DL-based JSCC work (which would require a fully joint design between the application and the network), such as lower end-to-end distortion and graceful degradation of distortion with channel quality. We believe this work represents an important step towards realizing semantic communications in wireless networks.

One of the most reliable ways to create deployable models for specialized tasks is to obtain an adequate amount of high-quality task-specific data. However, for specialized tasks, often such datasets do not exist. Existing methods address this by creating such data from large language models (LLMs) and then distilling such knowledge into smaller models. However, these methods are limited by the quality of the LLMs output, and tend to generate repetitive or incorrect data. In this work, we present Retrieval Based Distillation (ReBase), a method that first retrieves data from rich online sources and then transforms them into domain-specific data. This method greatly enhances data diversity. Moreover, ReBase generates Chain-of-Thought reasoning and distills the reasoning capacity of LLMs. We test our method on 4 benchmarks and results show that our method significantly improves performance by up to 7.8% on SQuAD, 1.37% on MNLI, and 1.94% on BigBench-Hard.

In this work, we describe a novel approach to building a neural PDE solver leveraging recent advances in transformer based neural network architectures. Our model can provide solutions for different values of PDE parameters without any need for retraining the network. The training is carried out in a self-supervised manner, similar to pretraining approaches applied in language and vision tasks. We hypothesize that the model is in effect learning a family of operators (for multiple parameters) mapping the initial condition to the solution of the PDE at any future time step t. We compare this approach with the Fourier Neural Operator (FNO), and demonstrate that it can generalize over the space of PDE parameters, despite having a higher prediction error for individual parameter values compared to the FNO. We show that performance on a specific parameter can be improved by finetuning the model with very small amounts of data. We also demonstrate that the model scales with data as well as model size.

Multimodal Retrieval Augmented Generation (MMRAG) is a powerful approach to question-answering over multimodal documents. A key challenge with evaluating MMRAG is the paucity of high-quality datasets matching the question styles and modalities of interest. In light of this, we propose SMMQG, a synthetic data generation framework. SMMQG leverages interplay between a retriever, large language model (LLM) and large multimodal model (LMM) to generate question and answer pairs directly from multimodal documents, with the questions conforming to specified styles and modalities. We use SMMQG to generate an MMRAG dataset of 1024 questions over Wikipedia documents and evaluate state-of-the-art models using it, revealing insights into model performance that are attainable only through style- and modality-specific evaluation data. Next, we measure the quality of data produced by SMMQG via a human study. We find that the quality of SMMQG-generated synthetic data is on par with the quality of the crowdsourced benchmark MMQA and that downstream evaluation results using both datasets strongly concur., Comment: Accepted to EMNLP 2024 Findings; Camera Ready

This study introduces a hybrid fluid simulation approach that integrates generative diffusion models with physics-based simulations, aiming at reducing the computational costs of flow simulations while still honoring all the physical properties of interest. These simulations enhance our understanding of applications such as assessing hydrogen and CO$_2$ storage efficiency in underground reservoirs. Nevertheless, they are computationally expensive and the presence of nonunique solutions can require multiple simulations within a single geometry. To overcome the computational cost hurdle, we propose a hybrid method that couples generative diffusion models and physics-based modeling. We introduce a system to condition the diffusion model with a geometry of interest, allowing to produce variable fluid saturations in the same geometry. While training the model, we simultaneously generate initial conditions and perform physics-based simulations using these conditions. This integrated approach enables us to receive real-time feedback on a single compute node equipped with both CPUs and GPUs. By efficiently managing these processes within one compute node, we can continuously evaluate performance and stop training when the desired criteria are met. To test our model, we generate realizations in a real Berea sandstone fracture which shows that our technique is up to 4.4 times faster than commonly used flow simulation initializations.

Long-text generation is seemingly ubiquitous in real-world applications of large language models such as generating an email or writing a review. Despite the fundamental importance and prevalence of long-text generation in many practical applications, existing work on personalized generation has focused on the generation of very short text. To overcome these limitations, we study the problem of personalized long-text generation, that is, generating long-text that is personalized for a specific user while being practically useful for the vast majority of real-world applications that naturally require the generation of longer text. In this work, we demonstrate the importance of user-specific personalization for long-text generation tasks and develop the Long-text Language Model Personalization (LongLaMP) Benchmark. LongLaMP provides a comprehensive and diverse evaluation framework for personalized long-text generation. Extensive experiments on LongLaMP for zero-shot and fine-tuned language tasks demonstrate the effectiveness of the proposed benchmark and its utility for developing and evaluating techniques for personalized long-text generation across a wide variety of long-text generation tasks. The results highlight the importance of personalization across a wide variety of long-text generation tasks. Finally, we release the benchmark for others to use for this important problem.

Vibrational properties of solids are key to determining stability, response and functionality. However, they are challenging to computationally predict at Ab-Initio accuracy, even for elemental systems. Ab-Initio methods for modeling atomic interactions are limited in the system sizes and simulation times that can be achieved. Due to these limitations, Machine Learning Interatomic Potentials (MLIPs) are gaining popularity and success as a faster, more scalable approach for modeling atomic interactions, potentially at Ab-Initio accuracy. Even with faster potentials, methodologies for predicting entropy, free energy and vibrational properties vary in accuracy, cost and difficulty to implement. Using the Covariance of Atomic Displacements (CAD) to predict entropy, free energy and finite-temperature phonon dispersions is a promising approach but thorough benchmarking has been hampered by the cost of Ab-Initio methods for sampling. In this work, we use a MLIP and the CAD to characterize the convergence of the predicted properties and determine optimal sampling strategies. We focus on solid lithium at zero pressure, showing that the MLIP-CAD approach reproduces experimental entropy, phonon dispersions and the martensitic transition while also comparing to more established methods.

Through the chemodynamical characterisation of metal-poor stars, one can efficiently probe the early history of the Milky Way. We aim at decontaminating a sample of $\sim$ 3M giant stars with Gaia DR3 XP-based \textit{Pristine-Gaia} metallicities, to investigate a subset of very metal-poor stars ([Fe/H] < -1.7) with disc-like orbits. We construct a statistically robust sample of $\sim$ 36 000 very metal-poor giants, using APOGEE and LAMOST to estimate and remove contamination from high $V_\phi$ stars. We investigate the spatial and kinematic properties of the decontaminated sample, using $V_\phi$ and the action space, both powerful to disentangle stellar populations. As in previous works, we find a pronounced asymmetry in $L_z$ and $V_\phi$ in favour of prograde stars. This excess is mostly made of prograde-planar stars (10% of the very metal-poor population), and contains stars with $V_\phi$ > 180 km s$^{-1}$ and $Z_{\text{max}}$ < 1.5 kpc, down to [Fe/H] = -2.9 at a 2$\sigma$ confidence level. While the overall orbital distributions of our sample match that of a halo, the highly prograde and planar subset (2% of the very metal-poor population) also bears characteristics classically associated with a thick disc, i.e., a spatial distribution compatible with a short-scaled thick disc, and a thick disc-like $Z_{\text{max}}$ - $R_{\text{max}}$ distribution. Additionally, assuming a stationary or prograde halo with $\overline{V_\phi}$ $\sim$ 30-40 km.s$^{-1}$ is not sufficient to suppress the kinematic signature of the highly prograde and planar subset. These results rule out any link with a thin disc, and instead, support a contribution from a \textit{metal-weak thick disc}., Comment: Accepted for publication in A&A on June 24, 2024. The abstract has been modified to comply with arXiv's abstract requirements (word limit). The paper numbering has been modified from XXV to XXIV to remain consistent with the recent submissions of the Pristine collaboration

All optical systems, which involve the collimation of a reflected, transmitted or scattered wave subsequent to tight focusing, are subject to two kinds of deviations. One is the wavefront curvature due to inaccurate focal placement of the interface or scatterer particle under consideration, and the other is the diffraction caused by the finite lens aperture. In the present paper we explore these phenomena in detail by considering a rigorous simulated model and an appropriate experimental setup. We hence demonstrate the complicated intensity profiles and optical singularity characteristics of the observed far field. Then we describe ways to minimize these deviations in a general experiment. But more importantly, our analysis proves that these deviations by themselves are significant optical phenomena of fundamental interest. The observed complex field profiles have similarities to standard diffraction-limited tight focal fields, though our field detection is different from the standard schemes. This indicates the relevance of these complex fields to a larger class of systems involving wavefront curvature and aperture diffraction. The detailed analysis and results of the present paper already serve as core explorations of these optical phenomena; and we also suggest future research directions where these system aspects can be purposefully created and explored further., Comment: 17 pages, 17 figures

Dimerized quantum magnets are exotic crystalline materials where Bose-Einstein condensation of magnetic excitations can happen. However, known dimerized quantum magnets are limited to only a few oxides and halides. Here, we unveil 9 dimerized quantum magnets and 11 conventional antiferromagnets in ternary metal borides MTB$_4$ (M = Sc, Y, La, Ce, Lu, Mg, Ca, Al; T = V, Cr, Mn, Fe, Co, Ni). In this type of structure, 3d transition-metal atoms T are arranged in dimers. Quantum magnetism in these compounds is dominated by strong antiferromagnetic interactions between Cr (both Cr and Mn for M = Mg and Ca) atoms within the structural dimers, with much weaker interactions between the dimers. These systems are proposed to be close to a quantum critical point between a disordered singlet spin-dimer phase, with a spin gap, and the ordered conventional N\'eel antiferromagnetic phase. This new family of dimerized quantum magnets greatly enriches the materials inventory that allows investigations of the spin-gap phase. All the quantum-, conventionally-, and non-magnetic systems identified, together with experimental synthesis methods of a phase suitable for characterization, provide a platform with abundant possibilities to tune the magnetic exchange coupling by doping and study this unconventional type of quantum phase transition. This work opens up new avenues for studying the quantum magnetism of spin dimers in borides and establishes a theoretical workflow for future searches for dimerized quantum magnets in other families or types of materials.

Communication technologies play a crucial role in battlefields. They are an inalienable part of any tactical response, whether at the battlefront or inland. Such scenarios require that the communication technologies be versatile, scalable, cost-effective, and stealthy. While multiple studies and past products have tried to address these requirements, none of them have been able to solve all the four challenges simultaneously. Hence, in this paper, we propose a tactical solution that is based on the versatile, scalable, and cost effective 5G NR system. Our focus is on the initial-access phase which is subject to a high probability of detection by an eavesdropper. To address this issue, we propose a novel approach that involves some modifications to the initial access procedure that lowers the probability of detection while not affecting standards compliance and not requiring any modifications to the user equipment chipset implementation. Further, we demonstrate that with a simple downlink power control algorithm, we reduce the probability of detection at an eavesdropper. The result is a 5G NR based initial-access method that improves stealthiness when compared with a vanilla 5G NR implementation., Comment: Submitted to IEEE MILCOM 2024

The Pristine-\textit{Gaia} synthetic catalogue provides reliable photometric metallicities for $\sim$30 million FGK stars using the Pristine survey model and Gaia XP spectra. We perform the first low-to-medium-resolution spectroscopic follow-up of bright (G<15) and distant (up to 35 kpc) very and extremely metal-poor (V/EMP, [Fe/H]<-2.5) red giant branch stars from this. We use Isaac Newton Telescope/Intermediate Dispersion Spectrograph (INT/IDS) observations centred around the calcium triplet region ideal for V/EMP stars. We find that 76\% of our stars indeed have [Fe/H]<-2.5 with these inferred spectroscopic metallicities and only 3\% are outliers with [Fe/H] > -2.0. We report a success rate of 77\% and 38\% in finding stars with [Fe/H]<-2.5 and -3.0 respectively. This will allow for 10,000-20,000 homogeneously analysed EMP stars using the WEAVE survey follow-up of Pristine EMP candidates. We associate 20\%, 46\%, and 34\% of the stars to be confined to the disc plane, or to have inner and outer halo orbits, respectively. We also associate these V/EMP stars to known accretion events such as Gaia-Enceladus-Sausage (GES), LMS-1/Wukong, Thamnos, Helmi streams, Sagittarius, Sequoia, etc. For the stars that orbit close to the disc plane, we find that the prograde region with low vertical action is overdense with a significance of 4$\sigma$ as compared to its retrograde counterpart. We also find three new (brightest) members of the most metal-poor stellar stream, C-19, one of which is 50$^\circ$ away from the main body of the stream. Our measured mean metallicity, velocity dispersion, and stream width are consistent with the literature, but our results favour a higher distance ($\sim$21.5 kpc) for the stream. We publish a catalogue (and 1D spectra) of 215 V/EMP stars from this spectroscopic follow-up and showcase the power of chemokinematic analysis of V/EMP end., Comment: Submitted to A&A. 17 pages (9 figures) + 3 pages (3 figures) in Appendix. Comments are very welcome! The catalogue and 1D spectra will be made available public after acceptance and before upon reasonable request to the first author

5G NR is touted to be an attractive candidate for tactical networks owing to its versatility, scalability, and low cost. However, tactical networks need to be stealthy, where an adversary is not able to detect or intercept the tactical communication. In this paper, we investigate the stealthiness of 5G NR by looking at the probability with which an adversary that monitors the downlink synchronization signals can detect the presence of the network. We simulate a single-cell single-eavesdropper scenario and evaluate the probability with which the eavesdropper can detect the synchronization signal block when using either a correlator or an energy detector. We show that this probability is close to $ 100\% $ suggesting that 5G out-of-the-box is not suitable for a tactical network. We then propose utilizing the uplink channel-state-information to beamform the downlink synchronization-signals towards the tactical user-equipment (UE) to lower the eavesdropper detection probability while not compromising the performance of the legitimate tactical UE., Comment: Submitted to IEEE MILCOM 2024