Preprint sees starspot shifts in extreme contact binaries
A study of 11 eclipsing systems reports light-curve reversals, mixed orbital-period changes and models that miss the shortest-period targets.
Search
2,073 results
21–40
New bounds map a threshold in composite hypothesis testing
Preprint: A mathematical framework gives finite-sample bounds and identifies when Type II error falls toward zero or converges to one as the Type I decay rate lies below or above a threshold.
AI model scores higher at finding camouflaged copper-tube defects
Preprint: CF-YOLO scored higher than YOLOv11n on three CTDD measures, while results were mixed on the external NEU-DET benchmark.
Trace BSA linked to disappearing satellite drops in liquid threads
Preprint: A laboratory study reports a centered neck and no satellite drops at 0.1 g/L BSA, with surface measurements and modeling pointing to a distinct late-stage thinning regime.
RISC-V FFT Tests Show juFFTe Gains, While Zen 5 Leads
Preprint: Tests found juFFTe ahead on several RISC-V configurations, but AMD Zen 5 led across the full transform-size range.
Particle-learning method reports better mixing in model tests
A preprint says its partially collapsed sampler matched a higher-cost full marginal method in simulations and mixed better than JAGS in one Norway Alpha-outbreak analysis.
Wireless AI System Cuts Image Data by Generating Missing Tokens
Preprint: The adaptive method reports strong semantic image scores at very low bitrates, but its receiver decodes far more slowly than SwinJSCC.
Graph structures reveal exact values for the diagonal F-threshold
Preprint: The study gives exact formulas for cycles, bipartite graphs and other families, while bounding the value for arbitrary graphs.
Non-Centred Volatility Models Score Better on Forecast Distributions
Preprint finds non-centred volatility scores 6% to 15% higher on density forecasts, while point-forecast differences stay below 1%.
A new sketching bound narrows a gap in tensor algorithms
Preprint: A mathematical analysis reports near-linear sketch dimensions for fixed-order Khatri-Rao embeddings under specific distribution assumptions.
Atmospheric-Neutrino Simulation Predicts Thousands of Single-Signal Events
Preprint: A Monte Carlo study estimates how atmospheric neutrinos could produce prompt-only events in a large liquid-scintillator detector, and where model uncertainty is greatest.
Preprint reports H-band interferometer signal at 188 pW per nanometre
A laboratory proof of concept reached a signal-to-noise ratio above 2 in 0.4 milliseconds, but it tested only one H-band spectral channel.
Released models improve gas adsorption predictions in MOFs
Preprint: A broad dataset and two machine-learning models match selected tests while improving predictions of carbon-dioxide adsorption.
Preprint reports faster simulations of graphene heat transport
A synthetic iterative scheme converged faster than conventional source iteration in deterministic tests, while matching the same discrete results.
Simple color forecast beat trained models on unseen copper
Preprint: A two-recording test found that a parameter-free color extrapolator beat trained forecasters on an unseen copper specimen.
Review Links Lead Perovskite Transport to a Soft, Shifting Lattice
Preprint review says one soft-polar framework may connect unusual transport, screening and lattice motion, but defects and polar order remain unresolved.
Robot Study Finds Gesture Commands Hold Up Better in Unseen Tests
Preprint: One pretrained robot model had higher success with language-plus-gesture and gesture-only commands than with language alone in several unseen tests.
Blue Field Ultra-Diffuse Galaxies Resemble Dwarfs in Star Formation
Preprint: Spectroscopy confirmed nine of 17 blue field candidates as ultra-diffuse galaxies, while dust and aperture corrections affected classification and estimated star formation.
Preprint: Photon emission varies more by exit charge than target material
A laboratory coincidence experiment tracked photons from helium ions transmitted through carbon, silicon and silicon carbide.
Preprint reports higher LLM scores with NVFP4 scale refinement
Tests across text-only Qwen3 and LLaMA configurations reported higher suite averages while inference cost stayed unchanged.