Study finds distinct immune-cell patterns in connective-tissue lung disease
In 217 CTD-ILD patients, lymphocyte counts were lower than in healthy controls, while SSc-ILD had mean T-cell and helper T-cell proportions of 76.2% and 49.39%.
Search
2,451 results
221–240
Abstract duality does not set the physical scale of momentum
An arXiv preprint argues that the de Broglie wavelength follows only after physical normalization fixes the momentum coordinate.
Sm-Doped BiFeO3 Powders Show Large, Frequency-Sensitive Response
Preprint: Pressed nanopowders showed broad dielectric peaks, while composition, temperature and frequency changed the response.
Models Find a Narrow Route for Venus to Lose a Moon
Preprint: Tidal calculations suggest a hypothetical Venus moon could be destroyed under a narrow set of starting spin and orbit conditions.
Preprint Models a Hot-to-Cold Bias in Molecular Motion
An idealized dilute-gas model argues that coarse-graining helps explain irreversibility, while leaving key assumptions untested.
GPU method nears best-observed scores in multi-objective MaxCut
Preprint reports rapid convergence on benchmark and synthetic graphs, but says the method may miss some best trade-offs.
AI method reports better balance when editing model facts
Preprint: MO-IKE reports stronger benchmark scores for changing target facts while retaining other knowledge in frozen language models.
Bayesian model shows less sensitivity to extreme data points
Preprint: The method kept quantile estimates steadier under severe contamination, while its uncertainty intervals still showed undercoverage.
Coding specifications often include implementation detail, study finds
This preprint reports that spec-driven development is used selectively, with participants describing clearer intent and more reliable agent work alongside extra planning and specification drift.
After AI-agent adoption, output rose as code quality diverged
Preprint: Repositories with more structured AI practices saw smaller increases in complexity and warnings, while commits rose across groups.
Twisted graphene shows signs of current-linked magnetic dynamics
This preprint reports transport signals in twisted graphene consistent with a proposed ferro-Josephson effect, while key evidence remains indirect.
Adaptive Robot Control Shows Higher Success in Transport Tests
Preprint: A robot that softened its response when sampled futures diverged completed more directional transport trials than two fixed-control comparisons.
Audio self-check spots one-word and two-word speech edits
Preprint findings show lower error rates than three passive detectors, but the proof-of-concept test covered only a narrow word-swapping attack.
AI Agents Struggle to Complete Full Biology Workflows
A preprint testing 13 frontier models found scores from 0.00 to 0.48, with large datasets and deep analysis chains linked to lower performance.
Delayed cache misses reshape the rules of paging
Preprint: Theoretical results preserve some online guarantees but expose weaknesses in familiar offline replacement rules.
Claim-Locked AI Reports Were More Stable Across Repeated Runs
Preprint: Reports built from fixed evidence-bound claims were more reproducible and less prone to directional errors, but the method did not validate the statistics behind them.
Preprint: Speech-act AI models lead in online derailment tests
A version 2 arXiv study reports the strongest benchmark scores for a model using speech-act information, especially with limited training data, while cross-dataset results varied.
Cross-market AI model ranks best in eight-market volatility forecasts
Preprint: PGA-Trans-HAR led average error rankings at five- and 22-day horizons and led on MAE at one day.
A Mandarin speech system shows gains on homophone wordplay
Preprint: An adapter lowered recognition errors in tests of ordinary and humorous homophones, but the humour set was small, synthetic and not evaluated by people.
Preprint proves blowup identities for topological-string counts
A mathematical proof covers local Hirzebruch geometries, while the local P2 result still rests on two unproved conjectures.