Random-feature rule reaches oracle rate up to logarithmic factors under conditions
A preprint reports NESR-KRR-RF, which compares neighboring fits with linear comparison complexity, alongside synthetic and benchmark tests.
Search
2,111 results
841–860
Odor playback ratings varied with scent and capture material
Preprint: In a 12-person evaluation, ScentEcho replays received different similarity ratings across materials, while perceived similarity and intensity were strongly associated.
Prompt wording linked to wide gaps in multimodal AI jailbreak tests
An arXiv preprint testing 16 multimodal models found the widest harmful-response gap in prompt framing, while task-relevant images were associated with higher attack success rates.
Preprint model tops traffic forecasts even with 80% data missing
An arXiv preprint reports lower forecast errors and shorter runtimes on two Caltrans traffic datasets, but the tests used randomly simulated gaps.
Pose-linked motion cues show higher action accuracy on benchmarks
Preprint: A joint-anchored optical-flow feature was tested against skeleton-only models across three action-recognition datasets, with the paper focused on earlier prediction from partial video.
Crab Nebula model predicts a much wider PeV gamma-ray glow
Preprint: A spatial model predicts a sharp size difference that could help distinguish hadronic from leptonic gamma-ray emission.
AI agents score higher in tests with graph-based skill retrieval
Preprint: A graph-based system posted the highest benchmark scores for six language-model backbones in simulated household and science tasks.
New model tackles heat-flow gaps in layered metamaterials
Preprint: Simulations suggest a generalized EMA can track heat flow in high-contrast layered systems embedded in a surrounding medium.
Model predicts a switch in particle production in proton-oxygen collisions
A preprint model comparison with eight centrality classes and preliminary ALICE data suggests the peak could move from the oxygen-going side to the proton-going side.
Preprint develops a new way to handle quarks and gluons near x = 1
A theoretical QCD study factorizes and resums next-to-leading-power terms in deep-inelastic scattering, while leaving non-perturbative inputs and higher-logarithmic work open.
Free spin waves match 2D Heisenberg magnet's leading thermodynamics
Preprint: A fixed-spin theorem for a square-lattice Heisenberg ferromagnet finds the free-boson coefficient for its leading low-temperature free energy.
Preprint Finds Higher Workplace Well-Being Scores Associated With App Use
A three-month workplace field study reported a modest rise in self-reported PERMA scores, but its nonrandomized design leaves the result as an exploratory association.
Preprint reports high-fidelity gates and five-qubit entanglement
The reported processor reached 99.957(5)% median single-qubit fidelity and 90.0(2)% five-qubit GHZ fidelity, while readout ranged from a 94% median to 85.5% at its lowest.
Degree rules split the difficulty of finding spanning trees
Preprint: A mathematical analysis reports tractable and W[1]-hard cases across several graph-structure measures.
Higgs-Pair Rates Could Exceed the Standard Model in New Scan
Preprint: A U(1)_X model predicts 60 to 100 fb at the 14 TeV LHC, above the paper’s 35 to 40 fb Standard Model benchmark.
Models estimate hundreds of supermassive black-hole pair candidates
Preprint: A population model estimates hundreds of candidate systems, while joint spectral and periodic tests select far fewer.
Study predicts tiny rates for rare top-quark meson decays
Preprint calculates tiny Standard-Model rates and an almost absent right-handed vector signal.
Welding cable model changes how robot joint loads are calculated
Preprint: A planar simulation found that a modeled welding umbilical can noticeably alter a robot’s dynamic response despite its much lower mass.
ClueWeaver Leads Local Methods on Four Evaluated Narrative Benchmarks
A preprint reports that ClueWeaver recorded 59.0% overall accuracy across four long-context narrative benchmarks, with paragraph-level evidence traces.
Near-Field Beamforming Nearly Matches Full SVD at Lower Cost
This arXiv preprint reports near-full-SVD rates and much lower CPU time for a geometry-based dual-UPA design.