New framework maps the fractal geometry of random collisions
Preprint: A mathematical framework measures the dimensions and fine-scale mass of collision times and points for independent Markov processes.
Search
2,451 results
1861–1880
AI method tests a finer way to screen misinformation in answers
Preprint results show the strongest scores in controlled tests, while the method’s performance on natural misinformation and real deployments remains untested.
Recommendation model reports higher recall with less computing power
Preprint: The model reported higher offline recall and a 2.53% revenue lift in a one-month test, but key online experiment details were not reported.
AI Study Reports Fewer Reasoning Tokens With Math Accuracy Intact
An arXiv preprint reports a 16.9% average reduction across six matched settings, while its GPQA-Diamond accuracy result did not meet the same equivalence test.
AI system traces how charts were altered and why
Preprint: VizAnchor combines embedded metadata, spatial tamper clues and sequential visual-language-model analysis, while leaving important gaps.
Gold cluster study tracks a staged sequence of vibrations
Preprint: Ultrafast spectroscopy and simulations trace high- and low-frequency motions at a gold–ligand interface, while catalytic turnover was not measured.
Quantum-dot light varies across AlGaAs nanowire designs
Preprint: InGaAs emission shifted across growth designs; separate demonstrations showed two InGaAs dots, or a GaAs-InGaAs pair, in individual nanowires.
Simulation maps resonance frequency to radial position
The Preprint on arXiv reports a model in which localized resonance peaks move outward as frequency rises.
AI training method reports stronger factuality scores in tests
Preprint: FARCA links factual feedback to source tokens in tests of two 3B language models.
Study maps how people make sense of artificial intelligence
An arXiv preprint examines news, Verified-account tweets and practitioner interviews to describe AI frames and debates.
Power-outage models falter on unseen states and events
Preprint: Random train-test splits gave stronger scores, but tests on unseen states and events often did no better than a simple baseline.
Cosmology preprint reports possible dark-energy crossing
Across different data combinations, the strongest reported tensions were 3.55σ above and 3.22σ below w = −1.
Metalens scanner maps mechanical differences in cancer tissue samples
Preprint: A benchtop prototype found mechanical contrast in five mouse pancreatic specimens and freshly excised breast and liver tissue, but clinical accuracy remains untested.
Model suggests how beta-gallium oxide films relax without dislocations
Preprint simulations point to a one-plane gallium rearrangement at the interface between two gallium oxide forms on sapphire.
Evolutionary model links reliable expression to slower adaptation
Preprint: An abstract evolutionary model suggests that reliable phenotype expression can come with less access to alternative high-fitness regions.
AI system checks optimization reformulations with proof certificates
Preprint: FLARE matched a benchmark perfectly, but its guarantee depends on accurate formalization and does not automatically extend beyond the test set.
Theorems link function rings to properties of abstract spaces
A version 1 arXiv preprint reports theorem-level links between closure and density in continuous-function rings and properties such as compactness and discreteness.
Black-hole detector model finds thermal behaviour only in a narrow limit
This preprint finds that a falling detector’s outgoing response can become thermal, but the complete response still carries two-branch and scattering effects.
Small adapters show promise in machine-vision compression
This preprint reports favorable ImageNet and COCO2017 results from a frozen-codec approach with far fewer trainable parameters, but provides no uncertainty estimates.
GRIP model blend reports similar accuracy with fewer tokens
Preprint: GRIP matched SLERP’s average accuracy and used 14.5% fewer tokens in reported benchmark tests.