A Two-Part Error Bound for Agnostic Multiclass Transductive Learning
Preprint: A mathematical analysis links DS and Natarajan dimensions to separate terms in worst-population error bounds, which match only up to logarithmic factors.
Search
2,451 results
1121–1140
Optical probing recovers AI-chip data in a controlled test
Preprint: A Kintex-7 testbed produced complete register and BRAM readouts, while algebraic methods recovered missing bits in many tested cases.
Financial AI models rank differently across risk-review tasks
A preprint reports that a Chinese-language benchmark finds broad financial scores can miss how models handle specific risk and compliance work.
PINN tests point to different roles for precision and architecture
Preprint study: matched tests link alignment to some hard-convection recoveries, while FP64 mainly lowered error in selected runs.
Microwave Amplifier Delivers Gain in Fields Above 1 Tesla
Preprint: A two-device experiment found more than 10 dB of gain above 1 tesla in near-parallel alignment, but nearly all explored settings produced no net gain by 20 millitesla in the perpendicular setup.
Mathematicians Extend the Range of Complete Non-Quadratic Hypersurfaces
Preprint: The paper reports Euclidean-complete examples across a wider parameter interval and three types for specified two-dimensional Calabi affine surfaces.
Preprint Reports Exact Dimension Law for the Three-Dimensional Veronese Curve
The theorem compares generic and tangent-special approximants and places a change in the dominant term at λ = 2/3.
Two-dimensional Oldroyd-B model remains well posed with large initial stress
Preprint reports global existence and uniqueness in two dimensions when stress diffuses and the initial conformation is nonnegative.
Preprint reports lower AI help for relationship manipulation in tests
Under the primary GPT-4o-mini judging protocol, a two-stage gate recorded a 2.21% pooled harmful rate and a 93.04% pooled protective rate in synthetic conversations, with no real users or targets involved.
Models suggest twin signals from dark-matter halos around black holes
Preprint: A theoretical model projects linked gravitational-wave and radio signatures during a neutron-star and black-hole inspiral.
New theorem ties thin-set colorings to a formal bounding principle
Preprint: A formal proof over RCA_0 derives BΣ^0_{n+1} from RT^n_{<∞,ℓ} for every n and ℓ at least 1, while leaving stronger questions open.
New criteria identify when stationary Navier-Stokes solutions are zero
Preprint: Geometric criteria can imply that certain smooth, decaying stationary Navier-Stokes solutions in R3 are identically zero.
A galaxy can lose gas while its AGN keeps flickering
This preprint uses a single-galaxy simulation to suggest how rapid AGN flickering can coexist with slow gas depletion and star-formation quenching.
Video model maps moments and objects tied to driving-risk scores.
Preprint: CoRE turns changes in a coarse video-level prediction into direct estimates of the time regions and tracked entities that support it.
New sufficient test for generalized Alexander quandle admissibility
A preprint dated 26 Aug 2026 sets out the condition, checks one finite example and reports seven qualifying cases from a classification.
FPGA system reports lower job-runtime error, 17-fold speedup over AVX
Preprint: An arXiv study reports lower job-runtime error on three historical traces; a separate ALCF benchmark averaged a 17-fold speedup over AVX.
Bitcoin Signal Search Shows Positive Scores but Negative Net Sharpe
Preprint: A deterministic audit caught all 40 illegal templates and rejected none of 40 legal ones in a finite benchmark, while all 23 holdout runs had negative net Sharpe under primary costs.
J-lens Study Reports Higher AI Readout Scores With Energy Filtering
Preprint: In tests on Qwen3-8B, variants retaining the top 10% to 20% of positions by Jacobian energy had higher reported averages, while masking did not cleanly separate two readout modes.
Preprint finds exact dimension for a difficult number set
An arXiv preprint reports a Hausdorff dimension of 2/(τ+1) in the stated nonzero-shift setting, using a scale-sensitive game.
Robot-learning study finds listwise feedback can match pairwise methods
Preprint: In simulated robot tasks, listwise visual rankings were competitive overall, but results varied by task and random seed.