One model could identify tracks and versions in benchmark tests
An arXiv preprint reports unified operation in a tested 10-second track-identification setting; scores varied between clean and altered audio and between segment durations.
Search
2,111 results
1661–1680
Study finds gains on some AI tasks when training matches sparse inference
Preprint: A 10-task test found strong gains on some benchmarks, mixed results on four core tasks and about 30% higher training cost.
Preprint sets conditions for derived kernels to have ordinary ring models
A theoretical study shows how higher-term tilting constructions can be represented by exact categories and, under specific conditions, by ordinary rings.
New York audit flags gaps in predictive lead-line records
An arXiv preprint says New York City's model-cleared records showed no lead or Unknown entries, but the estimate is not a direct accuracy test.
Filesystem design catches every tested storage attack, but performance varies
Preprint: ShieldZFS detected all tested storage attacks, while performance losses varied by workload.
Study links singular q-Dirac spectra to scattering theory
A preprint develops a theorem-based framework for dissipative q-Dirac operators, with results on dilation, scattering, functional models and eigenvector completeness.
GPS receiver study links longer integration to lower frequency fluctuation
Preprint: Simulations and a one-satellite GPS test linked longer integration and reduced spacing with lower reported fluctuations, while filter behavior varied under dynamic conditions.
Model links warm ions to changes in the Moon’s surface sheath
A preprint model reports its strongest changes at higher lunar latitudes, with possible implications for dust charging and motion above the surface.
Radio double J0011+3443 may be a short-lived cosmic relic
The preprint favors a compact symmetric object, but does not rule out gravitational lensing or a dual AGN.
2D anisotropic fluid model obeys energy balance for weak solutions
Preprint: A mathematical proof also finds weak–strong uniqueness when one solution has extra vertical regularity and the initial data match.
Australian Soil-Water Model Performs Better in Western Regions
A preprint reports that SWEB matched observed soil-moisture patterns across most regions, but its proposed uses for yield estimation and fertiliser recommendations have not been tested.
Gaussian Model Outperforms Graph Model in Heart CT Shape Test
This preprint reports better shape completion than a graph-based comparator on aligned CT meshes, but expert-surface checks covered only part of the heart and did not establish clinical use.
Parallel web library shows big speedups but not everywhere
An arXiv preprint reports speedups of up to 414 times in tests, while some variants ran slower than the sequential baseline.
ShadowPath phone tests show a wide gap in credential proof times
Preprint: ShadowPath moves credential-specific status lookup to the holder; Groth16 proving took 2.93 to 3.11 seconds on two tested phones, while PLONK took 39.12 to 55.29 seconds.
Open-source speech models may favor benchmark text over audio
Preprint: Tests of 11 open-source ASR models found benchmark-linked output could persist when speech was masked, altered or replaced.
Two language models outperform Wazuh and OpenSearch in a lab test
A preprint study found Meta Llama and Qwen detected more Linux authentication anomalies, while GPT-OSS often failed to return valid structured output.
Numerical method tracks topology changes in magnetic bands
Preprint: A finite-basis calculation follows Chern-number changes across weak, intermediate and strong periodic-potential regimes.
Neutron-star universal relations can shift near sharp phase transitions
Preprint: A mathematical model links the relations' stability to fading interior information and finds larger departures around sharp density jumps.
Quantum method reconstructs imaginary-time results from real-time data
A preprint reports excellent agreement for diffusion and good agreement at short scattering times, but two-qubit tests failed after a few Trotter steps.
A High-Energy Gravity Series Diverges, but Its Integral Stays Finite
Preprint: A high-energy gravity model yields a momentum-kick series with zero convergence radius; its Borel sum is identified with a finite integral when alpha1 is below two.