Preprint finds rising-systole paths for selected geometric surfaces
Explicit deformations connect square-tiled, hexagon and octagon constructions to maximal surfaces, but the result does not cover every translation surface.
Search
2,451 results
2421–2440
Preprint finds early warning signals for Solana memecoin rug pulls
Models using five minutes of trading data performed better within the same platform than when moved between Raydium and PumpFun, but the authors say the results are not ready for real-world deployment.
Preprint: DeepFOCUS takes two-photon images 1.2 millimeters into the mouse brain
The demonstrations reached hippocampal neurons and blood vessels, but the report describes one mouse for each.
Preprint maps energy shifts in a curved-space quantum oscillator
An algebraic calculation reports exact spectra and wave functions for two potential models, but uses no experimental or observational data.
Preprint extracts proton and neutron amplitudes from DVCS data
Two modeling strategies broadly agree, while large uncertainty remains around the nucleon’s real response.
Preprint reports benchmark lead in reconstructing a human in 4D from one video
The 4DAnyone system generated 16 target-view videos from a single front-view source and beat listed baselines on two test sets, while still struggling with loose clothing and inaccurate pose estimates.
Preprint reports robot opening and traversing doors from one phone video
The system averaged 96.57% success on five real doors and 80.95% on three structurally similar unseen doors, without additional training or trajectory generation.
Preprint: AI model often beats benchmarks in simulated epidemic forecasts
GENIE more often produced the best hospitalisation and death forecasts in held-out simulations, but it has not been tested against real-world surveillance data.
Preprint Proves Tighter Upper Bounds for Binary-Code Rates
A two-row honeycomb construction improves on a whole-cube benchmark and extends to finite certificates and larger matrix levels, where reported numerical gains remain uncertified.
Preprint reports better identity preservation in AI-generated group images
WithEveryone scored higher than GPT-Image 2 on automated identity measures in a 210-example test of groups with five to ten reference identities.
How lipid data are standardized can change which differences researchers see
A methods tutorial using neonatal piglet liver found that four processing approaches produced very different differential-lipid lists and sample patterns.
A reverse-date screening method could lighten the workload of evidence reviews
The approach uses quotas to move from database records to full-text analysis sooner, but its authors warn that useful studies may be missed.
Software offers a way to compare diagnostic tests when evidence is incomplete
NMADTA uses Bayesian models for up to five tests; an illustrative fit put ultrasonography’s sensitivity above D-dimer’s but its specificity below it.
Onion-peel zinc oxide particles linked to lower viability in lung-cancer cells
In laboratory tests, O3 was associated with apoptosis-related changes, higher reactive oxygen species and stronger nuclear p53 signals in A549 cells; the study did not test people or animals.
AI Struggles With Statistical Data in Meta-Analysis Benchmark
Across 58 manuscripts from randomized controlled trials, tailored prompts recorded the largest reported recall change, but statistical results remained the hardest information to extract.
Exact methods more reliable for sparse diagnostic-test meta-analyses, study finds
The clearest differences appeared in small or sparse studies, while methods were broadly similar in large, non-sparse datasets.
Preprint reports thick origami structures that fold and carry static loads
A modular hinged-panel design retained one coordinated folding motion in prototypes, including demonstrations supporting about 11 kg and 96 kg.
Preprint: Reasoning model matches engineered controller in simulated plant tests
A bounded system stayed within hard constraints across its primary benchmark runs, but the study does not establish performance on a physical plant.
Preprint: Brain Researcher evaluation finds higher AI tool-selection scores, but evidence gaps remain
Compared with the no-platform condition, the researcher-governed neuroimaging harness scored higher on route selection, capability coverage and handoffs, while verified citation grounding stayed low.
Preprint maps workflow gaps in China’s mental-health check-in tools
A preliminary Huixin EMAI deployment recorded responses to 76.6% of scheduled prompts; participants rated it favorably, while researchers reported persistent operational frictions.