Frequency-aware audio model leads on five of seven music reconstruction metrics
Preprint: The stereo autoencoder also posted higher point estimates on all 12 reported automatic generation metrics.
Search
2,111 results
1601–1620
Preprint translates a geodesic identity into CFT equations
An arXiv version 1 preprint derives formal scale equations for a scalar two-point function and finite entanglement between disjoint spherical regions.
Preprint Finds Randomized Caps Improve Delegation's Worst-Case Guarantee
A theoretical analysis reports that random upper limits can outperform fixed caps when a decision-maker knows only the mean of an agent's private information.
Preprint reports faster benchmark runs for fixed-penalty optimizer
The method keeps the classical nonlinear-residual update while offering conditional complexity guarantees for smooth nonconvex equality-constrained problems.
Study tests camera-only drone turn guidance from 360-degree video
Preprint: A CNN trained on views extracted from 360-degree footage predicted continuous yaw corrections, and an offboard test traversed an unseen indoor laboratory.
VDM proof tools report lower unchecked share in corpus test
Preprint: An expanded VDM method adds support for loop, operation-call and recursive-measure cases, while several difficult scenarios remain unresolved.
One-at-a-time data may help predict complex interactions
A preprint reports that DExtrI led on several synthetic and benchmark tests, while its mathematical guarantees depend on stated model assumptions.
Mathematicians show needles can be optimal for some convex curves
An arXiv preprint proves sharp inequalities for selected planar curves but leaves open a higher-dimensional conjecture about flat, disk-like shapes.
New graph-colouring proof establishes two cases of a signed bound
Preprint: A theorem covers treewidth 3, and treewidth 4 when maximum degree is at least 10.
NAPE audio model ties AudioSet-2M and tops listed IEMOCAP baseline
Preprint: NAPE matched SSLAM on AudioSet-2M, came close on AudioSet-20K and ESC-50, and scored higher than the strongest listed baseline on IEMOCAP.
New statistical method tackles treatment choices with hidden confounding
A preprint reports better performance under simulated confounding, but results weakened when instruments were weak or treatment choices exceeded instrument levels.
Model Finds Different Instability Patterns in W and Z Fields
Preprint: A mathematical scan finds broader longitudinal instability regions and extra low-momentum structures for W modes, but does not calculate particle production.
CeNiAsO shows evidence of tilted magnetism and spin splitting
Preprint: Nuclear-resonance measurements and calculations associate a small canted moment with unusual spin splitting and a phase-specific Hall response, while field-driven reorientation remains inferred.
Some AI models can leak secrets they refuse to reveal, study finds
Preprint tests found that adaptive analysis could reconstruct short numeric secrets and infer memory attributes from routine outputs, with results varying by model.
Graph anomaly detector adapts to changing data in benchmark test
An arXiv preprint reports stronger scores than listed baselines, but the evidence comes from computational tests on selected graph streams.
Planned repairs could lower costs in a modeled service queue
Preprint: A mathematical model finds lower long-run costs when repair costs stay below the value of one temporary control period, but simulations remain uncertain.
Current reveals mixed metal-insulator state in bulk organic crystal
Nuclear magnetic resonance measurements support metallic and insulating regions coexisting while the crystal’s average temperature stays near its sharp transition under current.
Model finds momentum transfer matters more than skewness in pions and kaons
Preprint: A covariant NJL calculation finds stronger momentum-transfer dependence than skewness dependence in modeled pion and kaon valence distributions.
Quantum Method Cuts Qubit Use in Two Molecular-Docking Tests
Preprint: A full-basis encoding used fewer qubits and recovered optimal solutions more often than a ZX-basis comparison in two graph-based docking tests.
Preprint reports Marton’s one-letter bound can fall short
A mathematical construction shows that Marton’s one-letter inner bound can miss a higher achievable sum-rate for at least one finite two-receiver channel.