Search papers, labs, and topics across Lattice.
100 papers published across 4 labs.
MLLMs exhibit significant weaknesses in following scientific instructions, especially in chemistry, where they fail to meet fine-grained constraints despite increased model size.
Uncovering a rich atlas of physical concepts in a neutrino model reveals that interpretability can dramatically enhance angular resolution in particle physics tasks.
A neural-operator surrogate enables rapid inversion of millions of airborne electromagnetic soundings, achieving near real-time results previously deemed infeasible.
Fine-tuning a single molecular foundation model can yield superior performance across multiple olfactory tasks, challenging the need for task-specific models.
VINCENT achieves a remarkable mean motif recall of 0.826, significantly outperforming existing models in identifying key molecular regions driving drug synergy.
MLLMs exhibit significant weaknesses in following scientific instructions, especially in chemistry, where they fail to meet fine-grained constraints despite increased model size.
Uncovering a rich atlas of physical concepts in a neutrino model reveals that interpretability can dramatically enhance angular resolution in particle physics tasks.
A neural-operator surrogate enables rapid inversion of millions of airborne electromagnetic soundings, achieving near real-time results previously deemed infeasible.
Fine-tuning a single molecular foundation model can yield superior performance across multiple olfactory tasks, challenging the need for task-specific models.
VINCENT achieves a remarkable mean motif recall of 0.826, significantly outperforming existing models in identifying key molecular regions driving drug synergy.
Predictive accuracy and uncertainty quantification improve significantly when using a row-wise PCA approach that respects physical constraints in multi-output Gaussian processes.
FlowMoDL achieves unprecedented accuracy in 4D flow MRI reconstruction, outperforming traditional and contemporary models even under extreme acceleration conditions.
A novel deep-learning framework uncovers 97 promising solid-state electrolytes, revealing that Li$^{+}$ jump-network connectivity is the key to high ionic conductivity.
Orthogonal projection reveals that protein language model embeddings capture critical biochemical features, with their removal leading to significant drops in predictive performance.
Achieving up to 49.4% faster computations for neutron diffusion problems without sacrificing accuracy could revolutionize reactor core analysis.
A generative model can slash the computational cost of simulating stochastic chemical reactions while preserving accuracy.
Low temperature continental climates deliver the highest discounted lifetime energy yield for photovoltaic modules, reshaping strategies for solar energy deployment.
Task-oriented sampling strategies can substantially elevate multimodal protein language model performance, revealing that default protocols are often inadequate.
LLMs misjudge research ideas as "medium novel" due to a systematic bias, but a new probing method boosts their accuracy by over 22%.
Evidence weighting in Homo-RAG boosts retrieval effectiveness, achieving near-perfect relevance scores while enhancing the quality of gene function predictions for non-model organisms.
Core-valence double ionization reveals that core holes dramatically alter molecular orbital configurations, challenging existing assumptions about electronic structure in SF6.
A new AI-integrated curriculum for mechanical engineering empowers students to tackle complex thermal engineering problems while enhancing their interdisciplinary skills.
Recovery rates of lithium and cobalt can be significantly impacted by the presence of H$_2$O$_2$ and the formation of passivating films during leaching processes.
Non-Dyson ADC methods may introduce significant errors in ionization potential calculations, challenging their reliability in correlated electron theories.
Achieving 97% accuracy in reconstructing complex cryo-EM mixtures could revolutionize how we analyze heterogeneous biological samples.
Enzymatic degradation of plastics may leave behind "fluffy" microparticle clusters that significantly impact recycling efficiency.
PAGS achieves sharper reconstructions in photoacoustic tomography by eliminating the need for calibrated speed-of-sound priors, even in heterogeneous media.
Delayed observations can dramatically enhance El Niño forecasts, but more complex models don't necessarily lead to better predictions.
oFM not only outperforms traditional prognostic models but also links treatment outcomes to biological mechanisms, revolutionizing patient care in oncology.
BioKERN achieves superior biological-neighborhood retrieval by leveraging an explicit biological kernel, outperforming existing methods without relying on larger models.
Quantum kernels can significantly boost fraud detection accuracy by aligning feature-map geometry with interaction-sensitive decision boundaries, outperforming traditional methods.
Calibration of uncertainty in structural monitoring can be dramatically improved, reducing coverage error by nearly 56% across diverse operational conditions.
Identifying distinct disease trajectories in T2DM reveals critical insights into patient progression and comorbidity risk that could transform clinical management.
Classical algorithms are fundamentally outmatched by quantum methods in sampling from Gibbs states, requiring exponentially more queries as dimensions increase.
Neural emulators can outperform traditional PDE solvers by leveraging insights from both paradigms, revealing a surprising synergy that enhances simulation efficiency.
Deep learning can sift through imaging noise to reveal thousands of potential dual active galactic nuclei previously dismissed as false positives.
Integrating physics directly into neural operators can cut prediction errors by up to 2.2 times, transforming how we model complex dynamical systems.
Conditional memory can either enhance or hinder scientific reasoning, and knowing when to activate it is crucial for optimal performance.
A mesh-free deep energy method achieves near-perfect accuracy in brittle fracture modeling, outperforming traditional approaches on complex crack patterns.
EMFE achieves 94.6% accuracy in malaria cell classification while remaining computationally efficient and interpretable, challenging the dominance of deep learning in this space.
Transforming a conical intersection into an avoided crossing reveals that topology can drastically alter molecular dynamics, leading to distinct hopping rates even on identical energy surfaces.
Soot concentration in equi-diffusive flames increases with pressure, but soot production rates are exclusively pressure-dependent, revealing a nuanced relationship that challenges existing combustion models.
Achieving DFT-level precision in NMR shielding calculations with a scalable, efficient framework could revolutionize how we interpret NMR spectra across diverse chemical systems.
Photoexcitation of gold nanoclusters triggers a rapid and complex reorganization of their ligand interfaces, revealing new pathways for energy transfer and catalysis.
A novel minimal chemical oscillator reveals a kink-like discontinuity in thermodynamic quantities at the Hopf bifurcation, challenging previous numerical-only insights.
High partial scores in scientific workflows can be misleading, as models often claim completion without delivering results, revealing a critical flaw in current evaluation methods.
AI usage in mathematical research surged from 1.39% to 14.09% of submissions in just five months, with Combinatorics leading the charge.
A novel self-consistent field framework reveals that finite quantum-chemical clusters can effectively model polarization and charge transfer under constant potential boundary conditions.
NEGF-MCPDFT reveals that carbon wire junctions exhibit transport properties that closely match experimental observations, challenging existing models of electron correlation.
Best-effort computing can achieve a 92% scaling efficiency in evolutionary models while gracefully handling hardware failures—transforming our approach to high-performance digital evolution.
The ability to design RNA sequences with short helices opens new avenues for RNA engineering, challenging previous limitations in structural RNA design.
Weak Raman bands can arise from strong atomic contributions canceling each other out, challenging traditional interpretations of Raman intensity.
Altering the reordering rate can change nucleation rates by several orders of magnitude, revealing a surprising turnover effect in multi-step nucleation processes.
Collective intermolecular Coulombic decay can occur in gases, enabling multiple donors to ionize an acceptor even with insufficient individual energy.
3D conformational fluctuations can significantly enhance polymer chain rupture rates, but bending stiffness may reverse this effect, leading to unexpected dependencies on bond position.
A new framework categorizes molecular LLM agents into four levels of autonomy, revealing critical gaps and risks in their deployment for scientific discovery.
Local Neural Operators can reconstruct complex solar wind states with unprecedented accuracy, transforming how we model heliospheric dynamics.
Text supervision can significantly enhance the organization of complex Cell Painting data, but matching genetic perturbations to compounds remains a challenging frontier.
Influence functions can now pinpoint the most impactful training samples without requiring labels, revolutionizing data attribution in scientific missions.
Four new long-period exoplanet candidates reveal the hidden dynamics of planetary systems previously thought to be incomplete.
A structured framework reveals five critical constants that could dictate the architecture of successful quantitative trading systems.
Clustering count data just got a major upgrade with 3CPO, which not only delivers reliable clusters but also highlights the most relevant features for enhanced interpretability.
Maximizing the distance between end members allows for unprecedented clarity in unmixing highly mixed grain-size distribution data, outperforming traditional methods.
Unraveling the complexities of single-cell responses, DeMixPert achieves unprecedented accuracy in predicting gene expression changes from unseen genetic perturbations.
Achieving accurate parameter recovery in quantum systems with a single moment-level PINN could revolutionize our approach to modeling quantum Brownian motion.
By integrating machine learning with traditional simulation methods, this framework boosts resonant antenna designs from 40% to 52%, showcasing a leap in design efficiency.
Neural Boltzmann Equations enable unprecedented precision in modeling early universe particle dynamics, overcoming traditional computational bottlenecks.
Machine learning models reveal that satellite data can unlock critical insights into phytoplankton community structure beyond traditional chlorophyll-a metrics.
ADDA revolutionizes data assimilation by seamlessly integrating simulation and observation with automatic differentiation, making it easier to tackle complex dynamical systems.
QRC combined with physics-informed corrections can dramatically enhance forecasting accuracy for chaotic systems, outperforming traditional methods in key metrics.
Surrogates that are cheaper per call can still fail to accelerate simulations, revealing fundamental limits in their deployment for complex multiphysics problems.
Hierarchical supervision can boost uncertainty estimation in LLMs, enhancing decision-making in biodiversity monitoring by improving prediction accuracy significantly.
Current scientific agents struggle to maintain a coherent narrative across evidence and calculations, with only 34.81% achieving strict accuracy in complex tasks.
Qwen3-32B may be the most reliable at verdicts, but GPT-5 outshines it in mimicking human reasoning paths during scientific fact-checking.
Language models can now be rigorously evaluated on their ability to generate falsifiable research ideas, not just stylistic fluency.
Treating a frozen large language model as the decision-maker in molecular design dramatically enhances candidate selection and optimization efficiency.
Transforming legacy scientific applications into cloud-native workflows is now feasible without losing original functionality, thanks to an innovative AI-assisted framework.
Automating the entire medical imaging model-development pipeline can yield competitive baselines with minimal engineering effort, challenging traditional iterative approaches.
AnaDiffusion allows for precise and controllable editing of 3D brain MRIs, achieving the lowest FID scores while preserving anatomical integrity.
The joint density of parked cars can be expressed as a rational function, revealing intricate spatial laws that govern parking dynamics.
Local residue couplings can accurately capture complex protein conformational dynamics, revealing hidden correlations and enabling efficient sampling along activation landscapes.
Binding affinity between Pdots and ovalbumin protein is dramatically enhanced at pH = 1, revealing a stark contrast in electrostatic interactions that could inform nanotechnology applications.
A comprehensive sustainability strategy in a major scientific computing department reveals actionable insights for aligning digital infrastructure with climate goals.
RSA cryptography faces a 50% vulnerability threshold by 2032, urging immediate action for PQC migration.
Visualizing high-probability differential clusters in lightweight ciphers reveals critical structural insights, achieving zero false positives in the process.
LLMCrater transforms the way we generate FAIR metadata by continuously enriching it throughout the research lifecycle, not just at publication time.
Accurate computation in stochastic networks is fundamentally limited by thermodynamic principles, requiring either high energy dissipation or long-lasting memory.
Importance-reweighted FS-VMC achieves accurate variational energies across diverse electronic-correlation regimes, redefining efficiency in quantum Monte Carlo methods.
Discovering significant correlations in dwell-times of TRPM8's kinetic states could redefine our understanding of cold sensation mechanisms.
LUCAID outperforms experienced pathologists by achieving 93% concordance in lung cancer diagnostics, setting a new standard for AI in precision oncology.
Gen2Physics doubles the accuracy of material segmentation in generated 3D meshes, enabling them to be immediately usable in interactive simulations and gaming.
LLM-SPICEMixer boosts circuit design efficiency, achieving up to 93.3% accuracy by combining LLM insights with traditional genetic algorithms.
Learning relative thermal stability differences can achieve over 85% accuracy, outperforming conventional methods in a field plagued by experimental variability.
Achieving 242.97 GFLOPS on a RISC-V accelerator not only outpaces traditional CPUs but also does so with a fraction of the power consumption.
cGraphGANFed generates high-quality molecules that not only meet user-defined metrics but also show improved resilience against common pitfalls in generative modeling.
Achieving a 94% correlation between predicted and actual RNA yields, this framework revolutionizes how we prioritize mRNA sequence designs for therapeutic applications.
Achieving Kohn-Sham DFT accuracy without the computational burden of orbital diagonalization opens the door to simulating massive electronic systems efficiently.
Label-free optical mapping reveals that electrically isolated regions can suppress interfacial charging signals, challenging assumptions about connectivity in electrochemical systems.
Strain engineering can significantly tune the electronic band gap of lithium chalcohalide antiperovskites, enhancing their potential as solid electrolytes.
Stable trapping of non-laser-coolable ions for over six hours reveals a threshold behavior that could transform studies in ultracold ion-neutral interactions.
Transport efficiency in photosynthetic proteins is governed not just by environmental fluctuations but by the intricate distribution of spectral weight across frequency space.
The formation of molecular hydrogen from water occurs through three distinct pathways, with the reaction's timing hinging on an unexpected symmetry-breaking motion.
Ion transport can be finely tuned in two dimensions, enabling hybrid devices that outperform traditional single-dimensional control methods.
A causal mapping framework reveals how to reconstruct sources in real-time QEDFT, transforming our approach to open quantum systems.
Extrapolation of interatomic potentials can be dramatically improved by integrating high-order many-body correlations, outperforming existing models in extreme conditions.
