Exact Norm Maximization and ICNN Lipschitz Problems Stay Hard
Preprint: Exact Lp norm maximization and a two-layer ICNN Lipschitz problem are W[1]-hard when dimension is the parameter.
Search
2,451 results
1921–1940
Economic model rankings differ under a data-compression test
A version-1 arXiv preprint dated 25 August 2026 compares Minimum Description Length with AIC, BIC and cross-validation in simulations and three economic reanalyses.
Smart-Home Simulator Passes Did Not Match Every Physical Check
Preprint: An audit found that simulated success varied with the property and timing boundary checked in physical deployment.
Forecast-explanation method scores well, but results vary by dataset
An arXiv preprint reports strong flow-only faithfulness scores across four benchmarks, while the full pipeline loses ground on three.
Capra recovers major IBEX sky patterns in a new preprint
An arXiv preprint tests the HEALPix-native pipeline on IBEX-Hi data, reporting broad feature recovery, close numerical preservation and a 12.23 speedup with 16 kernels.
ROMNet tops neural baselines in wave-speed reconstruction tests
Preprint: In held-out wave-speed reconstruction tests, ROMNet beat two learning-based comparators, but direct ROM and an oracle had lower errors.
Mathematicians classify when finite semigroups have complete mappings
Preprint: The study reduces the question to principal factors and gives exact tests across several finite algebraic systems.
PBMC tests report higher ARI and NMI in weak-transport runs
An arXiv Preprint reports higher ARI and NMI than an IGW comparison in a one-donor PBMC benchmark, while macro-F1 was not uniformly higher.
Carbon-wire simulations show model choices change electron transport
An arXiv preprint reports that CASSCF orbital optimization gave the strongest qualitative transport match in modeled carbon-wire junctions.
New sensor-calibration method handles tilted ground better
A 2026 arXiv preprint reports stronger repeatability in several datasets, though results varied with terrain and initialization.
Study reports conditional entropy stability in selected moving-wall tests
This preprint reports entropy-conserving or entropy-stable behavior in mathematical analyses and selected computations under stated moving-wall assumptions.
AI method assigns new knowledge-graph entities without retraining
Preprint: The study reports that SNAP-KG assigns arriving entities to clusters from raw features, narrowing candidate searches but exposing a trade-off for link prediction.
Camera-image planner shows lower error and latency than BEV in scooter test
Preprint: An offline campus-video evaluation found a direct image-space planner with lower path-centering error and far lower CPU latency than the best BEV method, but it did not test closed-loop scooter use.
Preprint finds conditional barrier to quantum security proofs
An arXiv preprint says restricted reductions cannot establish security for QMA witness encryption or classical-output null-qIO if a stated gap problem exists.
Approximate Bayesian updates can preserve fast prediction rates
Preprint: A theoretical study says approximate updates can retain fast regret when they track the exact posterior’s contraction.
Fairness-Aware CLIP Tuning Reports Better Subgroup Scores
Preprint: A label-free test-time method reports better subgroup metrics in benchmarks but incurs extra processing time and remains untested in real-world deployment.
Tests find Exradin A30 response varies in ultra-high-dose-rate electron beams
Preprint: Laboratory measurements found dose-per-pulse and grounding differences; the authors conclude the chamber is suitable for UHDR reference dosimetry when appropriate CCE corrections are applied.
Preprint reports stronger results from a multimodal recommender
The authors report top results across three offline Amazon review datasets, but the study does not test real-world user outcomes.
Preprint Reports MPNO Staying Bounded in Long Physics Rollouts
MPNO completed a single shock-tube test without divergence, but its formal spectral bound applies only to the model's linear propagator.
Pruned network nears optimal regression rates in theory
Preprint: The analysis reports a lower-dimensional rate, but tests the method only on synthetic data.