Preprint reports sublinear regret with feasible queries only
The theoretical method uses one feasible query each round, but has not been tested on data or real-world problems.
Search
2,111 results
1441–1460
Quantum kernel tops synthetic tests, but trails on one fraud benchmark
Preprint reports leading scores on interaction-defined synthetic tasks and one fraud dataset, but a classical kernel led on another.
Atom-thin lens shows 30% modulation in focused light
Preprint: A monolayer WSe2 zone-plate lens showed about 30% nonlinear focal modulation and a picosecond-scale response in laboratory tests.
AI mapping system leads urban segmentation benchmark
Preprint: EVEREST posts the strongest reported overall mask scores, but the benchmark comparison does not establish a causal advantage.
TEM Models Diverge in Heavy Crystals, but Aperture Helps
Preprint simulations found larger gaps between diffraction models in strontium titanate than in diamond, while a 110 mrad detector aperture improved agreement.
Graphene Flat Bands Shift Across Moiré Sites Beyond Roughly 10 Nanometres
Preprint: STM/STS measurements found period-dependent energy shifts and gap patterns, while modeling points to atomic corrugation as a possible link to topological order.
AI-Assisted Preprint Claims Proof of Finite-Group Conjecture
The arXiv manuscript presents a formal proof claim about non-vanishing elements in finite solvable groups.
SkillForge reports higher benchmark scores for language-model agents
A preprint compares SkillForge on ALFWorld, WebShop and AppWorld, but reports no task counts, independent-run counts or uncertainty intervals.
Simulation Associates Local Sensing With Three Particle Patterns
Preprint: A two-dimensional model linked sensing range and activity to holes, stripes and labyrinths; higher density accompanied stronger local order and smoother interfaces.
Uneven electric fields linked to motion in some liquid-crystal droplets
Preprint: In laboratory tests, RM734 droplets moved in approximately 30% of attempts under fresh PTFE conditions, while the remaining attempts showed deformation or instability.
Preprint Reports More Balanced Synthetic Faces With One-Step Guidance
A lightweight boundary learned late in denoising was applied once at the start, with reported gains in balance and mostly steady quality and speed.
AI Text Detectors Need Different Signals for Different Writing
Preprint analysis finds spectral signals gain strength in long, continuous AI text, while confidence measures lead on short or edited passages.
Combined strategy setup led the main AI benchmark comparison
An arXiv preprint reports that CoT+Ours performed best in the main comparison, while strategy-only solving and supervised fine-tuning produced mixed results.
Retinal vessel rings match an AI model, but lag it elsewhere
Preprint: A ring-based classifier matched RETFound on HRF but scored lower on FIVES and SUSTech-SYSU.
New test targets false alarms in multi-arm clinical trials
Preprint proposes adjusted statistics for trials that balance covariates during randomization but may omit them from the final analysis.
Preprint reports 54% smaller optimization gaps for SHSP
The structure-aware approach reported lower gaps across several solver comparisons, but the tests used selected benchmarks and a time-limited reference solution.
WAVE reports lower depth errors at 32× in benchmark tests
An arXiv preprint reports lower RMSE for WAVE than SPFNet at 32×, alongside tests across datasets, model components and shifted RGB guides.
AI-guided equation search reports strong benchmark results
Preprint: A large language model guides an equation search through semantic clues and structural features, with reported scores on physics, scientific-domain and real-world benchmarks.
Agent-runtime study finds faster runs but exposes safety limits
An arXiv preprint reports faster matched tool runs, but its tests stop short of showing broad safety, rollback or model competence.
AI Study Finds Some Vision-Model Heads Matter More Than Others
A preprint reports a 13.7-point shallow-layer POPE F1 gap when selected visual heads are compared with all attention heads.