221–240
In a one-seed comparison, the 160-million-parameter model reached 1.76 times the throughput of a matched all-attention twin at 2,048 tokens, while downstream scores were essentially tied.
A theorem gives uniform bounds for hyperplane families and extends to certain bounded-degree divisors, but only under strict growth and geometric conditions.
A mathematical construction shows that the best dimension-dependent factor approaches √e as the number of dimensions grows.
In a controlled simulation, a partial-order method matched all 1,000 evaluation traces at nine executions; sequential discovery did so only around 24.
The best model scored 99.64% balanced accuracy against 17 known generators, while its built-in feature rankings broadly tracked SHAP; unseen sources were not tested.
In one illustrative MIMO model, optimized waveforms matched a target pattern more closely than the starting set, but the work has not been validated with hardware or field measurements.
Tests on two operational datasets found that spatial context performed best in one, while a combined encoding led in the other.
In a single-field scalar Gauss–Bonnet inflation model, the lattice prediction rose above the perturbative result when primordial curvature fluctuations reached about 10^-2.
The theoretical paper matches twisted gamma-factor formulas on the automorphic and Galois sides, with its stated result restricted to essentially tame middle supercuspidal representations.
A benchmark of 10 language models found no system that reliably identified gaps in incomplete legal questions while answering complete ones directly.
An exact Brownian-walker analysis predicts nonlinear splitting probabilities and diffusion-dependent long-time tails; a numerical fractional-Brownian extension changes the exponents.
A two-stage model matched its target waveform components closely, but recovery tests exposed systematic shifts when signals came from a different waveform family.
The formal framework represents belief, disbelief, conflict, reliability and uncertainty, while connecting several probability descriptions and update rules.
A mathematical analysis says nested covariance formulas behave differently from slope-type denominators when a denominator becomes weak under Gaussian sampling.
A theoretical analysis finds gains across several continual-release tasks, but relative error does not ease the hardest adaptive MinSelect case.
The framework combines qubits with U(1) rotors and applies its constructions to gauge-covariant hopping, direct-angle phase estimation and rotor Fourier circuits.
PROMISE-CNN scored above U-Net and PROMISE-Txformer generally above UNETR; cardiac agreement was better in one analysis, but clinical performance was not tested prospectively.
The SPT-3G+ receiver targets a six-year survey with deeper maps, faster mapping, stronger lensing reconstruction and more cluster detections.
A formal model links human welfare to ownership of machine production and treats its 30% to 100% growth range as an illustration, not a forecast.
A constrained four-body model produces persistent oscillations from one starting pattern, while the proposed atomic implementation remains untested.