Graph-Based Optimiser Reports Zero Final Error on Selected Problems
Preprint: Mycelial Search matched other optimisers on selected CEC 2022 functions, but its results varied with landscape structure and information flow.
241–260
Preprint: Mycelial Search matched other optimisers on selected CEC 2022 functions, but its results varied with landscape structure and information flow.
Preprint: The system reported lower perplexity and higher benchmark accuracy than several baselines near three-bit compression, but AWQ was faster on the tested GPU.
Preprint: A conditional model ties tail risk, search costs and position sizing to its assumptions, but its one-index test found no robust position.
Preprint calculations suggest that coherent J/ψ production could make shell filling visible in diffractive patterns, while ϕ production is more sensitive to saturation effects.
Preprint analysis shows how reported energy, carbon emissions and costs change when a climate-model workload is counted in different ways.
Preprint: A proof gives an if-and-only-if result for one Toeplitz operator, while a constructed product breaks the same test.
Preprint: LUCAID combined nine analysis tools and matched an expert-panel reference in 93% of prospective clinical decisions, but routine clinical benefit remains unestablished.
Preprint compares fixed-capacity stationary, mobile and hybrid AED deployments using emergency incidents, AED inventories and vehicle trips in New York City and Toronto.
Preprint: Tests of a learned overlay reported lower attack success on frozen Qwen and Llama models while returned passages remained highly similar.
A preprint uses recursive constructions to show a depth advantage for every adjacent fixed-depth pair, while leaving the relevant regularity regime unresolved.
Preprint: A mathematical construction links position and momentum through Fourier duality and uses the canonical commutation relation to set the physical scale.
Preprint: The system reported higher task success than listed baselines in simulation and physical manipulation, alongside stronger future-video prediction scores.
ExMesh++ builds a mesh first, then estimates surface materials and lighting on fixed UV maps for later editing and relighting.
A formal study recovers Palatini dynamics when a scale condition is met, but leaves the wider non-integrable sector without a clear physical meaning.
Preprint: A mathematical model finds that a hinged swimmer can show directed motion when its angle and stiffness are cycled together.
Preprint: A lookahead approach cut approximation error in tests and restored the exact arm ranking at depth two, but the evidence is computational and limited.
A preprint finds frequent but shallow nonconvexity in two-dimensional tests and develops certified radial bounds under stated assumptions.
Preprint: The method outperformed RQ-KMeans in an industrial search test and across four public recommendation benchmarks.
An arXiv note reports a conditional theorem: strong local stability follows after an Artin–Schreier base-change when every Frobenius iterate meets the stated condition.
Preprint: In one robot-based test, a model-based system produced semantically enriched measurements, with most generation within 500 microseconds.