Six pieces of research: particle physics, post-quantum cryptography, machine-learning fairness, and three studies that sit outside those fields.
Searches for the dark photon across the 100–150 MeV mass window, through the exotic decay π⁰ → γA′.
A tungsten target, a dipole magnetic spectrometer, Delay Wire Chambers for tracking and a 4×4 electromagnetic calorimeter, simulated end to end in Geant4. The mass range is the one high-energy colliders cover least well.
Maps 104,640 BoardGameGeek titles across six historical periods since 1930.
Publication growth, the distribution of mechanics and themes by decade, and diversity measured with Shannon's index. Kruskal–Wallis tests with post hoc comparisons separate the periods by rating, complexity, playing time, participation and player count.
Trains the contrast between a high-entropy dreamer agent and a web-grounded evaluator into hypotheses that survive physical and institutional constraints.
A Rust multi-agent orchestration turns speculative model outputs into testable scientific hypotheses through an Epistemological Friction loop. An exploratory ablation across six conditions finds direct prompting among the weakest, but no general superiority over simple self-reflection — each architecture shifts the balance between originality, feasibility, diversity and grounding.
Cuts counterfactual gender bias from 21.18% to 7.66%.
A single-neuron auditor exposes the spurious features a network would otherwise lean on — Capital-Gain in the Adult Census set. Pruning them forces the model into a higher-capacity regime where it decides on merit. Beats L1 regularization at a fraction of the cost of Just Train Twice.
Across N = 243 European secondary students, self-rated digital skill outruns measured ability.
Students report high self-efficacy in passive consumption and score markedly lower on creation and algorithmic logic — a Dunning–Kruger gap. The technological gender gap reaches significance only inside Technology-oriented classrooms (p = 0.046), and 76.5% asked for more hands-on instruction.
Kyber beats RSA and ECC at key generation and shared-secret derivation, on Intel and on ARM alike.
A speed and storage comparison of the post-quantum algorithm against the two schemes it is meant to replace, run on commodity CPUs. Kyber is not a future contingency — on this evidence it is deployable now.