67639 research outputs found
Sort by
Towards an Extremely Robust Baby Robot With Rich Interaction Ability for Advanced Machine Learning Algorithms
No full textAdvanced machine learning algorithms require platforms that are extremely robust and equipped with rich sensory feedback to handle extensive trial-and-error learning without relying on overwhelming inductive biases. Traditional robotic designs, while well-suited for their specific use cases, are often fragile when used with these algorithms as they fail to address the intermediate sub-optimal posterior-based behavior these algorithms exhibit. To address this gap—and inspired by the vision of enabling curiosity-driven baby robots—we present a novel robotic limb designed from scratch. Our design features a semi-soft structure, a high degree of redundancy achieved through rich non-contact sensors (exclusively cameras), and strategically designed, easily replaceable failure points. Proof-of-concept experiments using two contemporary reinforcement learning algorithms on a physical prototype demonstrate that our design is able to succeed in a target-finding task even under simulated sensor failures, all with minimal human oversight during extended learning periods. Additional experiments on the robustness of the design show that it is able to withstand relatively large amounts of mechanical stress. We believe this design represents a concrete step toward more tailored robotic designs capable of supporting general-purpose, generally intelligent robots.The authors would like to thank Jan Przepiora for his assistance. This work was supported by the Center of Excellence for Generative AI at the King Abdullah University of Science and Technology (KAUST, Award Number 5940), the European Research Council (ERC, Advanced Grant Number 742870), and the Swiss National Science Foundation (SNF, Grant Number 200021 192356). Large language models were
used throughout the paper to improve readability
Nanofiller-confined spatial fluctuation in monomer diffusion synthesizing ultrafast reverse osmosis membranes driven by hydrogen-bonding networks.
No full textThin-film composite (TFC) reverse osmosis (RO) membranes encounter a significant trade-off between water permeability and selectivity. This study presents a mechanism to address this limitation by altering the structure of the polyamide (PA) layer. By incorporating layered double hydroxides (LDH) and sodium lignosulfonate (SL), we establish differential diffusion resistances during interfacial polymerization (IP). This approach facilitates the diffusion of m-phenylenediamine (MPD) across the interface while concurrently inhibiting it in the bulk phase, thereby inducing spatial fluctuations in monomer diffusion. The resulting heterogeneous polymerization dynamics yield a thin, highly wrinkled PA layer that promotes ultrafast water transport. Moreover, the hydrophilic sulfonic groups (-SO3-) present on the LDH nanosheets form a robust hydrogen-bonding network with water, further enhancing transport efficiency. The optimized membrane attains a water permeance of 4.00 LMH·bar-1 and a NaCl rejection rate of 99.4%, surpassing most current TFC/TFN RO membranes. This research offers insights into the control of the polymerization process, contributing to the design of next-generation RO membranes.This work is financially supported by the National Natural Science Foundation of China (No. 52270076, 92475205, 22178076), the National Key R&D Program (2023YFE0127000), and the Open Research Fund of Suzhou Laboratory (No. SZLAB-1308-2024-ZD007). Prof. Zhao would also like to thank the financial aid from the Fundamental Research Funds for the Central Universities (22120250217)
PhysGym: Benchmarking LLMs in Interactive Physics Discovery with Controlled Priors
No full textEvaluating the scientific discovery capabilities of large language model based agents, particularly how they cope with varying environmental complexity and utilize prior knowledge, requires specialized benchmarks currently lacking in the landscape. To address this gap, we introduce \textsc{PhysGym}, a novel benchmark suite and simulation platform for rigorously assessing LLM-based scientific reasoning in interactive physics environments. \textsc{PhysGym}'s primary contribution lies in its sophisticated control over the level of prior knowledge provided to the agent. This allows researchers to dissect agent performance along axes including the complexity of the problem and the prior knowledge levels. The benchmark comprises a suite of interactive simulations, where agents must actively probe environments, gather data sequentially under constraints and formulate hypotheses about underlying physical laws. \textsc{PhysGym} provides standardized evaluation protocols and metrics for assessing hypothesis accuracy and model fidelity. We demonstrate the benchmark's utility by presenting results from baseline LLMs, showcasing its ability to differentiate capabilities based on varying priors and task complexity.The research reported in this publication was supported by funding from King Abdullah University of Science and Technology (KAUST) - Center of Excellence for Generative AI, under award number 5940
Recent progress in underground hydrogen storage
No full textWith the global population anticipated to reach 9.9 billion by 2050 and rapid industrialization and economic growth, global energy demand is projected to increase by nearly 50%. Fossil fuels meet 80% of this demand, resulting in considerable greenhouse gas emissions and environmental challenges. Hydrogen (H2) offers a promising alternative due to its potential for clean combustion and integration into renewable energy systems. Underground H2 storage (UHS) enables long-term, large-scale storage to achieve equilibrium between seasonal supply and demand. This review synthesizes recent advancements in UHS, highlighting progress and persistent challenges. The review explores the complex mechanisms of H2 trapping and its implications for storage security and efficiency. The challenges these mechanisms present compared to other gases are discussed, emphasizing the unique properties of H2. The exploration covers interactions between H2 and geological formations, focusing on the wettability, interfacial tension, and sorption characteristics of rock–H2–brine systems. Advanced experimental methods are evaluated alongside the effects of critical parameters, including temperature, pressure, salinity, and organic contaminants. Findings from innovative imaging, core-flooding techniques, and computational methods (e.g., molecular dynamics simulations and machine learning) are incorporated. These approaches are vital for understanding H2 behavior in subsurface environments and developing robust, efficient storage solutions. This review offers a comprehensive update on recent progress, identifying and addressing the remaining gaps in UHS research. This work also highlights the significance of interdisciplinary research and technological innovation in overcoming these challenges. By providing insight into recent theoretical research, practical applications, and technological development, the findings support the successful incorporation of H2 into the global energy infrastructure, contributing to implementing a sustainable H2 economy successfully and fostering energy security and environmental protection for future generations
Enhanced Prediction of S-wave Velocity and Geomechanical Properties using Depth-Ordered Recurrent Neural Networks - A Case Study in West Texas
No full textSince pore fluids have less influence on shear waves, S-wave velocity ( VS) directly measures a rock frame's stiffness. However, direct VS measurements are scarce in vintage wells of the Delaware Basin. While empirical and traditional machine learning methods have been used to predict VS from conventional well logs, their estimates often depend on specific geological formations or boundary conditions. In this study, we develop a self-attention Bidirectional Long Short-Term Memory (BiLSTM) model with depth-ordered sequences and automated hyperparameter optimization to overcome these limitations, requiring no prior geological information. Trained on P-wave velocity, density, total porosity, and gamma-ray logs from 123 wells, the model captures nonlinear relationships in the data, achieving an R-squared of 0.85 and surpassing existing empirical and regression-based approaches. Further validation in blind tests with eight wells in different counties demonstrates accurate VS predictions throughout the basin where direct VS logs are missing. To interpret the model's "black box," we compute Shapley values to quantify each input's contribution to VS predictions. In addition, the BiLSTM's superior performance extends to geomechanical properties, including bulk modulus ( K), shear modulus ( G), Young's modulus ( E), and Poisson's ratio ( v), highlighting its practical utility in seismic applications such as energy exploration and development, geothermal energy, carbon and hydrogen storage, and induced seismicity monitoring
CCDC 2418684: Experimental Crystal Structure Determination :
No full textAn entry from the Cambridge Structural Database, the world’s repository for small molecule crystal structures. The entry contains experimental data from a crystal diffraction study. The deposited dataset for this entry is freely available from the CCDC and typically includes 3D coordinates, cell parameters, space group, experimental conditions and quality measures
Semantic Meta-Split Learning: A TinyML Scheme for Few-Shot Wireless Image Classification
No full textSemantic and goal-oriented (SGO) communication is an emerging technology that only transmits significant information for a given task. Semantic communication encounters many challenges, such as computational complexity at end users, availability of data, and privacy-preserving. This work presents a TinyML-based semantic communication framework for few-shot wireless image classification that integrates split-learning and meta-learning. We exploit split-learning to limit the computations performed by the end-users while ensuring privacy-preserving. In addition, meta-learning overcomes data availability concerns and speeds up training by utilizing similarly trained tasks. The proposed algorithm is tested using a data set of images of hand-written letters. In addition, we present an uncertainty analysis of the predictions using conformal prediction (CP) techniques. Simulation results show that the proposed Semantic-MSL outperforms conventional schemes by achieving a 20% gain in classification accuracy using fewer data points yet less training energy consumption
VLMs Play StarCraft II: A Benchmark and Multimodal Decision Method
No full textWe introduce VLM-Attention, a multimodal StarCraft II environment that aligns artificial agent perception with the human gameplay experience. Traditional frameworks such as SMAC rely on abstract state representations that diverge significantly from human perception, limiting the ecological validity of agent behavior. Our environment addresses this limitation by incorporating RGB visual inputs and natural language observations that more closely simulate human cognitive processes during gameplay. The VLM-Attention framework consists of three integrated components: (1) a vision-language model enhanced with specialized self-attention mechanisms for strategic unit targeting and battlefield assessment, (2) a retrieval-augmented generation system that leverages domain-specific StarCraft II knowledge to inform tactical decisions, and (3) a dynamic role-based task distribution system that enables coordinated multi-agent behavior. Our experimental evaluation across 21 custom scenarios demonstrates that VLM-based agents powered by foundation models (specifically Qwen-VL and GPT-4o) can execute complex tactical maneuvers without explicit training, achieving comparable performance to traditional MARL methods that require substantial training iterations. This work establishes a foundation for developing human-aligned StarCraft II agents and advances the broader research agenda of multimodal game AI. Our implementation is available at https://github.com/camel-ai/VLM-Play-StarCraft2
Machine Learning Computer Vision Point of Care Decision Support of Echocardiographic Identification of Hypertrophic Cardiomyopathy
No full textBackground: Hypertrophic cardiomyopathy (HCM) remains underdiagnosed, and artificial intelligence tools for echocardiographic recognition have been hampered by lack of insight into drivers of model predictions and ease of implementation. Objectives: The purpose of this study was to train and validate a machine learning model with visualization of model prediction, optimized for implementation, to identify HCM from echocardiography. Methods: 1,601 HCM cases from 2000 to 2021 were matched on age, sex, year of echo, and ejection fraction to 7,103 controls from Duke Medical Center. Multivendor echocardiograms were used to train and validate a convolutional neural network (CNN) identifying HCM. Saliency maps were produced for insight into model predictions. CNN performance was evaluated by receiving operating characteristic and precision recall and additionally investigated among a subset of 232 patients with cardiac magnetic resonance imaging-based HCM morphologic grading. Results: Among the 1,601 HCM cases, the median age was 61, with 46.9% male. The median ejection fraction was 55% with 12.4% having an ejection fraction 0.95 for all morphologies with precision between 0.16 and 0.72. Saliency maps demonstrated maximum intensity within ventricular myocardium. Conclusions: A machine learning CNN with model prediction visualization optimized for clinical implementation identifies all HCM morphologic subtypes. Further work is necessary to validate model performance in external data and real-world use.This project was funded by a grant from Cytokinetics. PROMISE was funded by grants R01HL098237, R01HL098236, R01HL98305, and R01HL098235 from the National Heart, Lung, and Blood Institute. Dr Vemulapalli has received grants/contracts from American College of Cardiology, Society of Thoracic Surgeons, National Institutes of Health (R01, UG3/UH3), Food and Drug Administration, Cytokinetics, and Abbott Vascular; as consultant/honoraria/advisory board for AstraZeneca, Boehringer Ingelheim, Medtronic, Edwards, Cytokinetics, HeartFlow, Veralox Therapeutics, American College of Physicians, Icon, and Total CME. Dr Zhang has received grants/contracts from National Institutes of Health R01HL169347. Dr Shah has received grants/contracts from nference. Dr Wang has received research grants/contracts from Cytokinetics, Bristol Myers Squibb; is a consultant/advisor for Bristol Myers Squibb, and BioMarin; has received speaker fees from Bristol Myers Squibb; is on the steering committee for Cytokinetics and Bristol Myers Squibb. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose