Proposes a non-intrusive correction method using VAE latent space adaptation to reduce discrepancies between CFD predictions and experiments for open tip clearance flow in compressor cascades, trained on 166 parametrically sampled CFD total pressure loss fields.
A region-held-out evaluation protocol for hydrogen embrittlement detection in 316L stainless steel, preventing data leakage from same specimen regions, comparing texture features vs deep learning for SEM micrograph classification.
This high-energy physics work develops a Parnassus conditional flow matching surrogate for the CLD detector, emulating full detector simulation and event reconstruction to address computationally expensive collider studies.
NewSee LICENSE fileHigh-energy physicsParticle detector simulationGenerative surrogate model
This paper investigates neural network solvers for infinity and p-Laplace problems in nonlinear analysis, applying PINNs and DeepONets to 2D and 3D domains with p values ranging from 2 to 1000.
A physics-informed deep learning framework performs single-snapshot 3D phase-only positioning in narrowband phase-coherent distributed MIMO networks under two-ray electromagnetic propagation.
This paper develops a surrogate-assisted stochastic multiobjective framework for photovoltaic hosting-capacity planning, jointly optimizing penetration and total electrical loss under reliability constraints.
This paper applies meta-adaptive filtering to active noise control when acoustic paths vary over time, constructing physics-informed optimizer features from a secondary-path estimate to support online filter updates.
This study evaluates five deep learning architectures for univariate significant wave height forecasting using buoy observations. Across a 47-buoy, 37-year evaluation corpus, the model families converge to nearly identical performance, with an between-family standard deviation of 0.0014 square meters.
This paper presents differentiable and hardware-accelerated implementations of the machine-learning surrogate models mlgw and mlgw-bns for gravitational-wave waveform modeling, enabling JIT compilation, vector parallelization, GPU acceleration, and automatic differentiation within JAX.
This paper studies gradient conflicts and multi-task optimization in Physics-Informed Neural Networks, with implications for more stable convergence in stiff and high-frequency partial differential equations.
Reveals compatibility conditions under which multi-symplectic Hamiltonian PDEs regularized by Brinkman-type penalization retain a multi-conformal symplectic structure, yielding exact local conservation laws for structure-preserving operator learning on complex geometries.
DNNsolver introduces a deep learning-based approach for modeling polarization-sensitive diffractive neural networks, addressing the limitation of thin element approximation in capturing intra-layer electromagnetic coupling for improved accuracy and efficiency in high-density optical computing.
A physics-informed GNN-based foundation model for AC-OPF, pretrained across 54 grid topologies (500-4,000 buses), achieving 2.45% zero-shot generation-cost error on 10,000-bus systems.
NewSee LICENSE fileAC-OPFGraph Neural NetworkPower System Optimization
This study develops a patient-specific image-based computational modeling framework to evaluate the hemodynamic impact of mitral and aortic valve repair in five LVAD-supported patients, addressing valve dysfunction threats to long-term LVAD therapy success.
LeafNet algorithm reformulates hyperbolic conservation laws as a coupled system with smooth solutions and approximates smooth functions via physics-informed neural networks, enabling accurate computation of multidimensional hyperbolic PDEs without diffusion.
PR-Smoother is an amortized smoother for prescribed-simulator data assimilation, supporting non-Gaussian posterior estimation over physical state variables in high-dimensional simulators with calibration compatibility.
RFlash uses differentiable acoustic simulation and radiance field decomposition to reduce bone-induced shadow artifacts in ultrasound imaging, improving symmetric assessment of fetal brain hemispheres.
Proposes SO(3) Equivariant Neural Kalman Networks (SENK) with response-state cascade to improve transferable vibrational spectral prediction across chemical spaces using equivariant transformer backbone.
SHRAV is an architecture-independent computational framework for physical modeling and inverse design, featuring a state-continuation core with declared reuse boundaries and explicit roles for learned evolution.
TM4FF introduces a physics-constrained operator learning framework combining Transformer attention with Mamba state-space models for accurate flow field prediction and denoising.
NewSee LICENSE fileFlow Field PredictionOperator LearningTransformer
Numerical investigation of how stationary distributions concentrate onto large scales in the fast-advection limit of stochastically forced 2D incompressible Navier-Stokes equations at N=128 resolution.
A reduced-order physical model describes re-ignition dynamics between consecutive discharges at the same position in dielectric barrier discharge arrangements, enabling time prediction without machine learning.
A topology-aware GCN model for fault classification in distribution networks with inverter-induced harmonic distortion using IEEE 34-bus system topology.
Proposes hybrid neural density estimation combining neural ratio estimation with flow-based models to address density normalization and efficient sampling in high-dimensional unbinned simulation-based inference.
NewSee LICENSE fileSimulation-Based InferenceNeural Density EstimationFlow Models
Proposes a physics-informed ML framework combined with multi-objective optimization for inverse design of Co-based superalloy compositions to enhance γ' solvus temperature while considering alloy density.
A periodic physics-informed neural network (P-PINN) framework is proposed for solving parametric periodic flows and analyzing finite-amplitude aeroelastic responses in a purely physics-driven manner.
Proposes Interface Autoencoder (IAE) for learning neural operators on geometry-dependent PDEs, representing interface-function states in a fixed domain via finite projection codes.
This paper formulates optimal biasing as an ergodic optimal control problem, recast as a linear eigenvalue problem for the infinitesimal generator of unbiased dynamics to learn bias potentials that accelerate transitions between metastable states in molecular dynamics.
Proposes active self-consistent field theory (ASCFT) for studying steady-state behavior of active Ornstein-Uhlenbeck polymers, extending classical polymer field theory to non-equilibrium systems.
NewSee LICENSE fileActive PolymersSelf-Consistent Field TheoryNon-Equilibrium Statistical Mechanics
This paper reveals the multi-conformal symplectic structure preserved by Hamiltonian PDEs regularized via Brinkman penalization, proving an exact local conservation law under a compatibility condition linking the symplectic matrix and penalization projection.
Investigates how load information quality affects power grid control using a DC power flow model and Grid2Op-inspired simulator, exploring data assimilation combined with reinforcement learning strategies.
Proposes a harmonic-regression combined ARMA model for fault detection in rotating machinery, addressing limitations of purely rational spectrum assumptions under varying operating conditions.
NewSee LICENSE filerotating machineryfault detectionARMA model
This paper develops an analytical and numerical framework for temporal scattering in electromagnetic metamaterials with shape-varying cylindrical inclusions, enabling dynamic control through abrupt geometric transformations of meta-atoms.
Proposes using ordinary multilayer perceptrons with finite-group averaging to achieve exact equivariant surrogate models for correlated electron systems, enabling efficient force and transition energy evaluations.
Uses patch-based autoencoder and neural ODE to build a distributed reduced-order model for studying relative periodic orbits in 2D Kolmogorov turbulence.
NewSee LICENSE fileKolmogorov turbulenceperiodic orbitreduced-order model
A self-driving laboratory strategy that simultaneously and dynamically assesses composition-dependent synthetic accessibility and target properties via aggregated cost functions to accelerate optimization of multicomponent NASICON fast ionic conductors while maintaining phase purity.
This survey covers brain-to-language decoding techniques, including neural signal acquisition, representation learning, and text generation, applicable to brain-computer interfaces and speech restoration.
NewSee LICENSE fileBrain-computer interfaceNeural decodingLanguage model
A physics-informed computational screening framework that decomposes effective second-harmonic response to optimize nonlinear optical properties of 2D semiconductor/organic molecule interfaces.
Investigates ferroelectric switching modeling by comparing four cost-effective DFT methods against high-accuracy ab initio approaches, aiming to establish reliable benchmarks for active learning workflows.
A benchmark dataset of complete numerical solutions for seven PDE families including Darcy, Poisson, and Helmholtz, generated by the benchmark's own solvers for scientific ML model evaluation.
See LICENSE filePDE benchmarknumerical solution datasetscientific ML
A context-conditioned molecule transfer model based on InternS1-Mini, designed for drug-induced liver injury (DILI) related molecular transformation tasks in chemical reaction prediction and drug discovery.
See LICENSE filemolecule transferdrug discoverychemical reaction prediction
A deep learning-based context-conditioned molecule transfer model designed for carcinogen-related molecular structure migration and generation in cheminformatics.
See LICENSE filemolecule transferchemistry discoverymolecular generation
HuggingFace blog tutorial on using NVIDIA Warp physics framework and MjWarp bridge for robotics simulation, covering rigid body dynamics and contact mechanics in learning workflows.
See LICENSE fileRobotics SimulationNVIDIA WarpMjWarp
FREESIA addresses multimodal posterior estimation under high-dimensional sparse observations by using covariance-aware posterior transport, improving inference accuracy for nonlinear or non-injective observation operators in dynamical systems.
See LICENSE filedata assimilationposterior transportcovariance-aware
Proposes an integrated CFD-ML framework for optimizing pod geometry in supersonic evacuated tube transport, leveraging Latin Hypercube Sampling and Bezier curve parameterization for aerodynamic design.
See LICENSE fileEvacuated Tube TransportPod OptimizationBezier Curve
This paper presents a privacy-preserving hybrid quantum-classical computing framework that integrates parameterized quantum circuits with classical components, targeting quantum computing proxy applications such as chemistry simulation.
See LICENSE fileQuantum Machine LearningHybrid Quantum-Classical ModelsPrivacy-Preserving
This paper presents an unconditionally stable second-order numerical method for fluid-poroelastic structure interaction, where free fluid governed by incompressible Stokes equations couples with poroelastic medium via the Biot system, reformulated using Robin-type interface conditions.
See LICENSE filefluid-structure interactionporoelasticityRobin boundary conditions
A deep-learning surrogate model based on residual CNN predicts Earth-induced transition probabilities for solar neutrinos across energy and zenith angle parameter space.
See LICENSE fileneutrino oscillationEarth matter effectdeep learning surrogate
Proposes finite-state automata encoding recursive lattice geometry for exact MPO construction of tight-binding Hamiltonians across fractals, quasicrystals, and hyperbolic lattices.
See LICENSE filetight-binding modelmatrix product operatorfinite-state automaton
A probabilistic physics-informed framework using WaveNet and ParamNet to solve the inverse problem of recovering Woods-Saxon nuclear potential parameters from sparse spectral data.
See LICENSE fileNuclear PhysicsPhysics-Informed Neural NetworksParameter Identification
A modal integrator framework (MIISO) is proposed to stabilize non-physical outlier modes at the top of the discrete spectrum in high-order isogeometric discretizations, improving numerical solutions in wave propagation and structural dynamics.
See LICENSE fileisogeometric analysisstructural dynamicswave propagation
DNNsolver proposes a deep learning approach for accurate and efficient modeling of polarization-sensitive diffractive neural networks, addressing limitations of thin element approximation in capturing intra-layer electromagnetic coupling.
See LICENSE fileDiffractive Neural NetworksDeep Learning ModelingPolarization-Sensitive Optics
Investigates the relationship between residual scale and learnability in Δ-learning, demonstrating through molecular graph neural network evaluation on total energy targets that smaller residual targets do not necessarily yield better learning outcomes.
See LICENSE fileDelta learningresidual learningmolecular graph neural network
DissipNet enforces dissipativity constraints to model dissipative dynamical systems, balancing prediction accuracy with physical property preservation for control theory and complex dynamics analysis.
See LICENSE filedissipative systemsrecurrent neural networkdynamical systems
JAMB proposes jointly denoising bimanual actions and motion states, explicitly modeling geometric interactions between arms in shared 3D scenes to enable coordinated manipulation.
See LICENSE filebimanual manipulationdiffusion policyaction generation
KATOsuper presents an objective-agnostic framework that couples neural-reparameterized topology optimization with sensitivity-consistent Fourier neural operators, addressing gradient inconsistency between predicted objectives and sensitivities for stable structural optimization.
See LICENSE fileTopology OptimizationFourier Neural OperatorSensitivity Analysis
A high-performance PDE neural solver based on Manifold Function Encoder, featuring resolution-independent memory footprint and arbitrary PDE input support for large-scale physical simulations.
See LICENSE fileNeural OperatorPDE SolverMemory Efficiency
Proposes a geometric constellation shaping (GCS) framework to enhance sensing privacy in OFDM-ISAC systems, addressing the privacy risk where passive eavesdroppers exploit ISAC data signals for unauthorized target sensing, revealing receiver-specific ranging performance dependent on constellation design.
PISCES is a convolutional autoencoder trained without catalog labels on OMNI solar wind measurements under physics constraints, designed for detecting solar wind transients and providing space weather early warning at the Sun-Earth L1 point.
See LICENSE fileSpace weatherSolar windAnomaly detection
Proposes a Runge-Kutta aligned oscillation-eliminating DG framework that fixes the disruption of error structures caused by nonlinear stabilization, enabling superconvergence analysis for fully discrete shock-capturing DG schemes.
See LICENSE fileDiscontinuous GalerkinSuperconvergenceShock-capturing
Presents AKAPINN, a mesh-free KAN-PINN method for nonlinear strain-limiting PDEs with exact hard boundary ansatz and pseudo-Huber loss, evaluated across three material-parameter pairs.
See LICENSE filePINNKolmogorov-Arnold NetworkQuasilinear PDE
Proposes a coupled Maxwell-cable framework integrating FDTD solutions with Hodgkin-Huxley and Fitzhugh-Nagumo dynamics to simulate action potential propagation in branched neuronal geometries.
See LICENSE fileNeuroelectromagneticsAction PotentialFDTD
This paper rigorously establishes inf-sup stability of the Scott-Vogelius finite element on three-dimensional Freudenthal meshes for polynomial degree k ≥ 4, completing the theoretical framework for this classical divergence-free mixed element used in Stokes flow computations.
See LICENSE filefinite element methodsStokes probleminf-sup stability
Investigates how uncertainty quantification approaches perform when adapted to geometry-conditioned neural surrogates for CAE, comparing closed-form and sampling-based methods across varying geometries and engineering quantities of interest.
See LICENSE fileneural surrogatesuncertainty quantificationCAE simulation
MARC-TS introduces a two-stage framework that learns endpoint-conditioned continuous reaction paths, queries at arbitrary resolution, and refines transition-state candidates via local path context, with T1x-IRC-8K dataset of 8,209 reactions and 1.09M path-resolved geometries.
See LICENSE filereaction pathtransition statechemical reaction
Proposes a tensor product multilevel method (TPML) combining Smolyak's sparse grids with kernel-based interpolation, coupling kernel scale to fill distance for uniform condition number bounds in high-dimensional function approximation.
See LICENSE filesparse gridskernel interpolationhigh-dimensional functions
This paper introduces ICG-MTO, a framework that leverages numerical foundational models to improve inter-task coupling estimation in few-shot multi-task optimization scenarios, addressing negative transfer issues.
See LICENSE filemulti-task optimizationnumerical foundational modelsfew-shot learning
Molecular dynamics simulations investigate how local chemical clustering affects structural and functional properties of Ba(Zr_xTi_{1-x})O_3 in ferroelectric and relaxor regimes, revealing mechanisms of chemical inhomogeneity on material performance.
See LICENSE fileMolecular DynamicsBZT MaterialChemical Clustering
Proposes adaptive interaction strategies for companion robots using deep reinforcement learning, overcoming traditional fixed-position tracking limitations to enable flexible robot accompaniment in dynamic workspaces.
See LICENSE fileDeep Reinforcement LearningHuman-Robot InteractionCompanion Robot
Presents a physics-informed ML framework for inverse design of Co-based superalloys, using atomic size mismatch and other descriptors with multi-objective optimization to enhance gamma-prime solvus temperature while constraining density.
See LICENSE fileCo-based superalloygamma-prime solvus temperatureML inverse design
This study evaluates physics-informed neural networks for atmospheric potential-temperature forecasting, constrained by pressure-coordinate thermodynamic advection-source equations to address short-horizon predictions under sparse thermal observation conditions.
See LICENSE filePhysics-Informed Neural NetworksAtmospheric Temperature ForecastingDigital Twins
This paper proposes a separated-section constitutive law for Cosserat rod modeling of trimmed helicoid soft robot arms, evaluating each helix domain's mechanical response in its local frame and mapping it back to the global coordinate system for improved modeling accuracy.
See LICENSE filesoft robotCosserat rodconstitutive law
This paper studies the convergence properties of the cyclic steepest descent (CSD) method for minimizing strictly convex quadratic functions in arbitrary dimensions, proving doubly exponential convergence rates and extending prior theoretical results limited to two-dimensional cases.
See LICENSE filenumerical optimizationconvex optimizationcyclic steepest descent
A physics-informed self-supervised learning framework that jointly performs wire calibration and interaction position reconstruction in multi-wire parallel plate avalanche counters, enhancing adaptability to changing operating conditions and detector aging.
See LICENSE fileparticle detectorwire calibrationposition reconstruction
Presents a fully implicit PIC method extended to electromagnetic cases with exact conservation of global energy and local charge, enabling plasma equilibrium simulations at high (up to extreme) β in open magnetic traps.
See LICENSE fileplasma simulationPIC methodelectromagnetic simulation
A technical Q&A dataset covering fundamental and advanced topics in materials science, including material properties, synthesis methods, characterization techniques, and engineering applications, designed for training technical response models.
See LICENSE filematerials sciencetechnical Q&Asynthesis methods
A benchmark dataset for PDE surrogate tasks within the ai4sci-tasks evaluation framework, containing training trajectories and evaluation splits for assessing neural operator and surrogate modeling performance.
See LICENSE filePDE SurrogateBenchmark DatasetScientific ML
Proposes specification-conditioned human demonstration collection for embodied AI in construction, bridging the gap between design model configurations and physical environments.
See LICENSE fileEmbodied AIConstructionHuman Demonstration
Presents a general theory for rolling arbitrary 2D crystals into nanotubes, implemented as open-source NTBuilder with desktop and web interfaces, including a catalog of 20 million nanotube structures.
See LICENSE filenanotube2D crystalmaterials discovery
A focused survey of PINN improvements via transfer learning and meta-learning, addressing PINN's notorious convergence difficulty and the need to re-optimize on every PDE parameter change - the engineering reference for productionizing PINNs.
arXiv preprint (copyright held by authors)PINNTransfer LearningMeta Learning
A survey on embedding physics priors in robot learning, covering how physical constraints like contact mechanics, rigid-flexible coupling, and multibody dynamics improve sample efficiency and reliability of learned policies.
See LICENSE filerobot learningphysics priorsmultibody dynamics
NVIDIA tutorial demonstrating GPU acceleration techniques for ROS 2 computational graphs using Isaac ROS and AI Agent integration to speed up robotics workloads.
See LICENSE fileROS 2NVIDIA Isaac ROSRobotics Acceleration
A survey of data-driven and physics-informed neural operators, organizing the architecture landscape (DeepONet / FNO / GNO / PINO / Transformer-based / diffusion) along three axes: scalability, generalizability, and data sensitivity.
arXiv preprint (copyright held by authors)Neural OperatorScientific MLDeepONet
A practical focused review of neural operators covering DeepONet, FNO and PCANet with hands-on training strategies; benchmarked on Poisson and linear elasticity, and demonstrated as surrogates for Bayesian inversion - an engineering-ready primer.
arXiv preprint (copyright held by authors)Neural OperatorDeepONetFNO
A systematic survey of physics-knowledge + ML methods for PDE prediction, splitting the field into architectural-level (loss, structured prediction, augmentation) and data-level (multi-task, meta, contextual) approaches, with industrial perspective and open-source ecosystem review.
arXiv preprint (copyright held by authors)Physics-informed MLPDEMulti-task Learning
ICML 2026 paper proposing APT, a geometry-/mesh-/physics-agnostic neural operator that fuses graph-based local feature extraction with global attention for long-range physical effects; achieves 4× saturation error reduction on 2D faulted CO2 storage vs prior SOTA, 19,873× speedup over GEOS and 6,800× over ECLIPSE.
arXiv preprint (copyright held by authors)Neural OperatorSubsurface EnergyCO2 Storage
ICLR 2026 paper proposing ATOM, a quasi-equivariant pretrained transformer neural operator that requires no explicit molecular graph and uses temporal attention for parallel decoding of multiple future states; curates TG80 (80 compounds, 2.5M fs trajectories) and demonstrates zero-shot generalization on MD17/RMD17/MD22.
arXiv preprint (copyright held by authors)Neural OperatorMolecular DynamicsTransformer
Reframes Transolver's Physics-Attention as a special case of linear attention and proposes LinearNO (Linear Attention Neural Operator); achieves SOTA on six PDE benchmarks with 40.0% fewer parameters and 36.2% less compute, plus superior performance on AirfRANS and Shape-Net Car industrial datasets.
arXiv preprint (copyright held by authors)Neural OperatorTransformerLinear Attention
Presents FRPN, a Full-Resolution Polymer Network that retains both atom-level details and monomer-level connectivity for multi-scale polymer property prediction.
See LICENSE filepolymermulti-scale modelingmaterials discovery
Investigating nonequilibrium phase-ordering kinetics in binary mixtures of uniaxial ellipsoidal particles using large-scale molecular dynamics simulations with anisotropic Gay-Berne potential, revealing two-stage ordering processes across different time scales after temperature quench.
See LICENSE filemolecular dynamicsphase-ordering kineticsanisotropic potential
LPINNs propose multiplying the first hidden activation of a dense PINN by input-dependent localization functions, endowing first-layer units with receptive fields to improve optimization on long domains and for high-order operators without domain partitioning.
See LICENSE filePhysics-Informed Neural NetworksPINN EnhancementNeural Network Localization
This paper proposes a data-driven policy construction method for dynamical systems using non-expert MPC demonstrations, introducing a memory-based nonparametric policy with Lipschitz-regularized upper envelope of the finite-horizon surrogate value function to approximate infinite-horizon optimal control.
See LICENSE fileModel Predictive ControlData-Driven ControlReinforcement Learning
This paper investigates using artificial neural networks as surrogate models for black-box optimization, enabling efficient optimization when analytical function descriptions are unavailable by obtaining objective values through numerical simulation.
See LICENSE fileBlack-Box OptimizationSurrogate ModelNeural Network
