Search papers, labs, and topics across Lattice.
100 papers published across 8 labs.
Task-driven Bayesian experimental design just got a major upgrade, enabling efficient learning of design policies without the complexity of posterior estimation.
Patient-aware contrastive learning can achieve an AUROC of 0.989 while preserving individual patient structures that traditional methods overlook.
Achieving a 4.3% prediction error, SuperCond-GNN transforms how we simulate and analyze high-temperature superconducting circuits, paving the way for real-time monitoring and design optimization.
HSPINN achieves exact boundary enforcement and faster convergence, revolutionizing how we solve PDEs with neural networks.
Non-canonical SMILES disrupt molecular representations more than invalid ones, revealing deeper insights into how cLMs understand molecular grammar and semantics.
Task-driven Bayesian experimental design just got a major upgrade, enabling efficient learning of design policies without the complexity of posterior estimation.
Patient-aware contrastive learning can achieve an AUROC of 0.989 while preserving individual patient structures that traditional methods overlook.
Achieving a 4.3% prediction error, SuperCond-GNN transforms how we simulate and analyze high-temperature superconducting circuits, paving the way for real-time monitoring and design optimization.
HSPINN achieves exact boundary enforcement and faster convergence, revolutionizing how we solve PDEs with neural networks.
Non-canonical SMILES disrupt molecular representations more than invalid ones, revealing deeper insights into how cLMs understand molecular grammar and semantics.
Federated learning in health research just got easier with FLKit, a structured onboarding toolkit that demystifies the process for diverse teams.
ChemBack reveals that even with rigorous admission protocols, chemically valid backdoors can still pose significant risks to molecular graph neural networks.
Ultra-peripheral collisions can be transformed into a powerful tool for nuclear imaging, revealing critical insights into nuclear structure through deep learning.
Embedding physical laws into neural networks not only enhances predictive accuracy but also unlocks interpretable insights from complex wood thermal behaviors.
Generative models for crystal design are largely recycling known structures, with 81-92% of outputs being duplicates or substitution variants.
AI agents can yield correct results while relying on fundamentally flawed reasoning, raising critical questions about their reliability as scientific co-authors.
Achieving accurate cosmological inference with just 60 high-fidelity simulations could revolutionize the cost-effectiveness of weak lensing studies.
SPADE reduces correct-prior regret from 10.3% to just 2.6% while achieving 100% accuracy in structure selection, revolutionizing how we incorporate physical priors in machine learning.
LISDD reveals that localized model errors can be pinpointed and corrected with unprecedented accuracy, transforming how we diagnose discrepancies in hybrid models.
A unified modeling framework for TENGs reveals the intricate interplay of charge states, transforming how we simulate and design energy-harvesting devices.
Achieving optimal quantum key distribution without a shared reference frame could revolutionize secure satellite communications.
A penalty-free approach to LWE cryptanalysis enables efficient quantum-classical hybrid solutions that could redefine post-quantum cryptography.
PHAST-Net achieves unprecedented accuracy in time-frequency analysis by integrating attention mechanisms with physics-informed learning, redefining how we approach signal representation.
Thermomechanical dynamics can now be seamlessly integrated into 3D scene rendering, enabling realistic simulations of melting and solidification processes.
Ocean4D achieves unprecedented stability and consistency in underwater 4D reconstruction, overcoming the limitations of traditional methods that fail to account for medium effects.
A novel reduced-order model reveals that woven structures can exhibit programmable mechanical responses, challenging existing modeling paradigms.
GPU offloading can dramatically enhance throughput and energy efficiency for scientific applications, but the benefits hinge on problem size and granularity.
Increasing NaCl concentration can paradoxically reduce the availability of dissociated chloride ions in supercritical fluids, challenging conventional assumptions about electrolyte behavior.
Charge transfer energy estimates can vary by nearly 1 eV depending on the computational method used, with standard approaches like B3LYP falling short.
A single-layer Vision Transformer can effectively reduce terabyte-scale data from X-ray detectors in real-time, proving that less can be more in high-speed scientific environments.
A novel electrochemical sensor can accurately distinguish between DNA intercalators and minor groove binders, streamlining drug candidate screening.
Achieving near-AO-DMET accuracy while retaining computational efficiency, this new LO-based DMET method transforms how we approach excited states in strongly correlated systems.
Universal MLIPs can be transformed into efficient configuration-space generators, enabling the creation of accurate material-specific models with minimal DFT calculations.
Cancer survivors show significantly impaired autonomic regulation during physical activity, with HR and HRV metrics revealing stark contrasts to healthy controls.
Identifying and resolving parameter degeneracies can cut simulation costs by up to 10x while enhancing our understanding of complex models.
A new constrained variable-projection approach achieves superior efficiency in structured data-science models, outperforming conventional optimization techniques.
Federated learning can outperform local training in survival analysis, with Random Survival Forest emerging as the top model for heterogeneous healthcare data.
SurfBind's innovative surface-centric approach outperforms traditional methods, revealing the critical role of molecular surface interactions in epitope prediction.
GRACE reduces false positives by 99% while recovering nearly all true causal edges in complex time series data, revolutionizing causal discovery efficiency.
Reinforcement learning can boost signal efficiency in high-throughput scientific facilities, achieving a 56% improvement in real-time event filtering at the Large Hadron Collider.
MAE-3D not only surpasses 2D methods in single-cell tasks but also sets new benchmarks for protein localization and interaction, showcasing the power of 3D modeling in microscopy.
A hybrid AI model outperforms traditional deep learning approaches in real-time melt pool monitoring, achieving high accuracy with minimal inference latency.
Bypassing cubic scaling walls, this method achieves over 14,100 times speedup in non-smooth NML estimation, revolutionizing large-scale statistical inference.
Achieving a basin-mean correlation of 0.94 with half to a tenth of the predictors used by traditional models reveals the power of graph neural networks in reconstructing historical terrestrial water storage data.
Fragment-conditioned lead optimization is revolutionized by Sesame, which enables targeted molecular growth from partial structures while maintaining high generation quality.
Achieving optimal small-set expanders could revolutionize key exchange protocols in a post-quantum world by maximizing neighbor connections in bipartite graphs.
The rise of militarized language in scientific abstracts could be eroding the credibility and funding prospects of research, especially in the Global South.
Cavity QED can shift molecular dissociation energies by several inverse centimeters, revealing critical nuclear dynamics often overlooked in quantum systems.
Non-Markovian dynamics can arise from Markovian processes due to noise, complicating entropy production estimates in single-molecule tracking.
Data isolation in decentralized operations can optimize chemical plant scheduling while keeping operational secrets safe, revealing surprising emissions dynamics.
A novel QUBO/Ising approach enables efficient design of cyclic peptides by reducing the complexity of residue representation without sacrificing interaction fidelity.
The charge-transfer reaction between ammonia and propylene oxide cations proceeds at a surprisingly slow rate, challenging existing theoretical predictions and reshaping our understanding of chiral molecule formation in space.
Ten-million-scale exploration reveals 279 promising ammonia synthesis catalysts, including new families previously overlooked in conventional searches.
Disciplinary siloing in research is starkly revealed through a novel citation graph that links claims to their sources, reshaping our understanding of knowledge evolution in AI fields.
Funding for IoMT research not only boosts scholarly output but also correlates with better health outcomes, highlighting the critical role of financial support in advancing healthcare technology.
Tracking marine animals with just one hydrophone per AUV achieves unprecedented accuracy, reducing data loss and operational complexity in aquatic research.
Electronic noise spectroscopy reveals nanoscale molecular parameters with unprecedented resolution, opening new avenues for chemical sensing in quantum devices.
Finer grain sizes may mitigate pitting but paradoxically accelerate defect propagation under mechanical stress, complicating corrosion management strategies.
Surprisingly, lower bulk thermal diffusivities in liquids can lead to faster interfacial thermal spreading, challenging conventional understanding of heat transport.
Achieving 0.8% error in glottal volume flow, this PINO-based method revolutionizes speech production analysis by eliminating the need for pre-computed training data.
Contextual tunneling in LLMs can be overcome, leading to more reliable and physically grounded materials discovery through the innovative ARIA framework.
Conventional computing systems are ill-equipped to handle the explosive growth of biological data, necessitating a paradigm shift in computer architecture for healthcare applications.
A single multimodal model outperforms specialized approaches across 80 biological tasks, redefining the landscape of biological AI applications.
Agentic models may resolve citations, but they still mislink to the wrong papers 15.9% of the time, exposing a critical flaw in current AI evaluation benchmarks.
Static reports are out; BioInsight's interactive system empowers researchers to dynamically explore and refine biomedical evidence like never before.
Transforming dense scientific papers into reader-friendly formats could revolutionize how researchers engage with complex literature.
Achieving 98.6% tracking efficiency with a 0.8% fake rate, HEPTv2 revolutionizes particle tracking by eliminating the need for graph construction and auxiliary processing.
Uncertainty-quantified neural surrogates can drastically enhance Bayesian inference efficiency for complex inverse problems, outperforming traditional methods in high-dimensional settings.
Evolutionary algorithms can drastically improve the accuracy of Physics-Informed Neural Networks by efficiently navigating the complex hyperparameter landscape.
CNNs can achieve high entropy estimation accuracy in multi-qutrit systems with only a fraction of the measurements needed for traditional methods, marking a significant leap in scalability.
ASYS reveals a novel way to automate the discovery of analytical forms for PDEs, producing interpretable solutions where none existed before.
A trajectory-based event detection method using topological data analysis outperforms traditional techniques in monitoring high-dimensional dynamic processes.
Integrating genomic data with ecological constraints allows for unprecedented accuracy in predicting microbial dynamics and carbon cycling in soil systems.
Multimodal ensemble approaches for cfDNA analysis could revolutionize early cancer detection, but standardization is crucial for progress.
TESSERA embeddings outperform traditional methods in fine-scale urban climate mapping, achieving remarkable accuracy and scalability.
SO-RaNN achieves exact mass matching and positivity in PNP systems while ensuring incompressibility in velocity fields, setting a new standard for accuracy in neural network-based simulations.
Achieving FEM-level accuracy with a 60x reduction in evaluation time, this framework revolutionizes localized response prediction for long-span bridges.
A neural network that accelerates PDE solutions by learning HODLR matrix inverses outperforms traditional solvers and existing neural operators in speed and accuracy.
Context-aware features can slash IVF prediction errors by over 60%, revealing critical insights hidden in environmental data.
Unsupervised algorithms can slash semantic segmentation labeling time from 170 hours to just 37 hours, revolutionizing data annotation in materials science.
Evaluation design choices can drastically alter the perceived risks of AI systems in biological research, revealing a hidden complexity that demands attention.
A structured coarse space can dramatically improve the efficiency of solving complex fluid dynamics equations on exascale systems, outperforming traditional AMG methods.
An ideal model of amorphous silicon reveals a pristine electronic structure that aligns perfectly with experimental observations, challenging traditional approximations.
Achieving Hartree-Fock precision in solid-state energy calculations could redefine benchmarks for computational materials science.
A new thermal-based definition of slip velocity clarifies the complexities of fluid transport at the nanoscale, offering a more robust framework for understanding solid-fluid interactions.
Ground states in ultrastrong light-matter interactions can be accurately modeled with a new nonperturbative variational approach that outperforms traditional methods.
The authors reveal that the Hamiltonian emerges naturally from a minimal set of axioms, reshaping our understanding of quantum dynamics.
Contour-constrained deformable registration can cut target registration errors in head and neck surgeries by nearly half, revolutionizing intraoperative guidance.
QCPIKAN accelerates PDE solutions with exponential convergence rates while embedding physical laws directly into the learning process.
TerraMARS transforms unstructured Mars research into actionable insights, paving the way for advanced terraforming strategies.
Encrypted genomic queries can now be conducted without revealing sensitive data, significantly enhancing privacy in genomic research.
Achieving state-of-the-art LVFP classification while enhancing interpretability, HypOProto transforms how we leverage echocardiography in heart failure diagnostics.
Achieving Hamiltonian cycle decompositions with minimal switches in non-coprime Eisenstein--Jacobi networks could redefine how we approach cycle-splicing problems in advanced interconnection networks.
The stationary spin entanglement entropy serves as a crucial observable that reveals unexpected relationships between entanglement structures and dynamical phases in the sub-Ohmic spin-boson model.
DFT-trained neural network potentials can accurately model hydration and exchange kinetics of magnesium ions, but struggle with solvation free energy predictions.
The nonlinear index of entangled polymers reveals a surprising geometric limit, challenging traditional views on shear deformation behavior.
Claim drift in automated research can lead to significant discrepancies, but Xcientist ensures that every generated mechanism remains accountable and traceable back to its evidential roots.
Machine learning resolves 20,000 ambiguous X-ray source matches, revealing the limitations of traditional spatial cross-matching methods.
Achieving high-resolution cardiac simulations with a method that guarantees reduced super-resolution error through physics-based constraints and Koopman regularization.
AI agents struggle to make reliable preclinical pharmacology decisions, with the best models achieving less than 60% accuracy in real-world evaluations.
Achieving up to 37% faster convergence on complex flow problems by harnessing graph neural networks to optimize algebraic multigrid solvers.
Interactive candidate selection in bioprocess optimization reveals trade-offs between performance, uncertainty, and robustness, empowering experts to refine criteria as models evolve.
Tailoring follow-up intervals for Type 2 Diabetes patients can cut costs by over 34% while enhancing care based on individual risk profiles.
DIPHINE reveals the intricate information dynamics of complex systems, outperforming traditional methods and enabling insights from real-world data without distributional constraints.
Achieving a 93% recovery rate for Earth-size transits, TransitNet outperforms traditional methods while maintaining a compact model size and high inference speed.
