2026-10-11
Monte-Carlo-VarianceDoes variance reduction change Monte Carlo's 1/√N error rate, or only its constant?
Complete
Maskana Labs is a small AI, machine learning and mathematics house. We formalise a problem, try ideas against it, and keep what survives contact with the evidence.
Models, training dynamics and evaluation. We care about why something works as much as whether it does.
Optimisation, probability and structure. Clear statements and proofs where we can get them, careful numerics where we can't.
Systems that reason, plan and act. We build small, inspectable versions first and scale only what earns it.
Write the question down in a sentence anyone on the team can challenge.
Turn it into something precise: an objective, a metric, a conjecture, a baseline to beat.
Run the smallest test that could prove us wrong. Cheap, fast, and many of them.
Prove it, ablate it, or reproduce it from scratch. Results that only work once don't count.
Record what we tried, what failed and what held, so the next experiment starts further along.
Tell us what you are stuck on, or what you wish someone would test. We read everything.
hello@maskanalabs.com