AI model posts faster, higher scores on 3D MRI questions
Preprint reports benchmark results for SeVeR, a system that selects visual information from multi-sequence breast MRI.
Search
2,451 results
1361–1380
Geometry-based model improves hard-sphere predictions in confinement
Preprint: A parameter-free mapping was tested against simulations of unmixing, cavity pressure and nanoscale force profiles.
Factory tests find two distinct signal patterns at 318 GHz
Preprint: Measurements at 11 line-of-sight receiver locations found that strong-reflection and random clusters followed different delay, power and angular patterns.
E-commerce search model reports modest GMV change in test
Preprint: In a 41-day A/B comparison, DCEO showed a 0.36% relative GMV change; the evidence comes from one search system.
Particle-model scan finds strong transitions in LISA simulations
Preprint modeling finds simulated three-year LISA signal-to-noise ratios up to 88, while benchmark thermal choices sharply change the forecast.
Model maps how chemical fuel reshapes a Brownian heat engine
Preprint: An exact mathematical model links chemical fuel, heat and mechanical load, but its predictions have not been tested in a physical motor.
AutoVerifier tops four benchmarks for answer verification
Preprint: The system reached 93.05 macro accuracy; its method turns recurring errors into replay-tested code modules.
Euclidean-Positive Kernels Can Fail on Even-Dimensional Spheres
Preprint authors build a smooth counterexample at every allowed radius and report that direct transfer holds exactly in odd dimensions.
Robot navigation system reports higher success in indoor trials
Preprint: EgoNav records higher success and path-efficiency scores than a matched baseline in simulation and physical office tests.
Computer model finds three-atom copper clusters bind gases on hBN/SiC
This preprint reports calculations in which three-atom copper clusters had the strongest adsorption energies among Cu1 to Cu5, with O2 bond changes consistent with molecular activation.
LDAC-Net scores higher in drifting gas-sensor tests
Preprint: LDAC-Net was tested on raw sensor windows across SmellNet and eNose-Drift benchmarks; uncontrolled deployment was not evaluated.
Preprint: Symbolic robot planner leads VLMs in kitchen simulations
A simulated benchmark reported higher task success and much shorter planning times for a planner that encoded object-to-object constraints, but no physical robot was tested.
AI Distillation Study Reports Uneven Gains From Reweighting Surprising Tokens
This preprint reports that low-probability tokens carried much of the update budget, while its benchmark gains varied by model size and task.
Quantum amplifier reports 0.06 photons of added noise
Preprint: A cryogenic SQUID-based amplifier reached 94 ± 12 millikelvin of system noise in phase-sensitive mode, below the reported 288-millikelvin standard quantum limit.
Preprint Reports Eight Modular Cases From Platonic Symmetry
The version-1 arXiv paper follows the tetrahedron, octahedron and icosahedron into elliptic surfaces, period sequences and arithmetic tests, with important exceptions.
Quantum Model Finds Clusters Can Retain Coherence
Preprint: A one-dimensional simulation finds clustered states with quantum coherence, while evidence for phase separation remains limited by the extended model’s size.
Preprint links Yang-Mills hierarchy to continuous and lattice systems
A version 1 arXiv preprint dated 26 Aug 2026 presents a Sato-theoretic construction for studying the Ward conjecture through continuous and discrete reductions.
AI model synthesizes contrast-enhanced breast MRI from pre-contrast scans
Preprint: The authors selected a parameter-efficient setup after comparing tumor-focused and global image metrics in 10 axial development instances.
A math framework links symmetric functions to three matrix models
Preprint: A proposed algebra connects stable series, shifted Jack functions and Gaussian checks, but its higher-degree formulas remain conjectural.
AI roof model beats one rival, but does not yet make 3D roofs
Preprint: RoofDiT uses a two-stage diffusion-transformer pipeline to generate planar roof graphs, with results varying by its input.