arXiv Nightly

Built tonight

Projects

• sw-cdf: sort-free Sliced-Wasserstein micro-library + live demo — a tiny self-contained JS/typed-array library implementing the paper's CDF-based Sliced-Wasserstein estimator, replacing the usual "project, sort, compare quantiles" sequential bottleneck with evaluating and integrating empirical CDFs on a fixed grid — an embarrassingly-parallel operation the paper built for federated learning (aggregable across machines without sharing raw samples). Pitch ships with a demo page: two draggable 2D point clouds, a rotating random-projection line, and a real-time SW-distance readout computed two ways side-by-side — classic sort-based vs. the CDF-grid method — with a Web Worker fan-out toggle showing the CDF method scale across cores while sort-based doesn't (arXiv:2606.30310).

Startups

• HalfFull/HalfEmpty Bankroll Optimizer for Sports Bettors — the paper generalizes the Half-Full/Half-Empty behavioral model to positive/negative lotteries, formulates portfolio selection as a nonconvex problem provably equivalent to a tractable Mixed-Integer Linear Program, and beats Prospect Theory, pure risk-minimization, and equal-weighting on real-data risk/return. Pitch: a bankroll-allocation tool for a slate of simultaneous bets/props — paste in implied prob + offered odds + correlation notes, drag the two HF/HE loss-aversion sliders, and watch optimal stake allocation shift against Kelly and equal-weight baselines in real time, solved client-side via a small MILP/LP relaxation — a personalization layer on top of standard Kelly staking (arXiv:2312.10749).

YouTube

• The Evolutionary Traffic Jam: How Simple Genes Build Impossible Bottlenecks — an ~8 minute explainer built around one animated visual: a 2D fitness landscape generated from the paper's global-epistasis model (fitness = nonlinear function of a simple additive trait) that starts smooth, then as you dial up nonlinearity two high-fitness basins pull apart and a narrow "bottleneck" corridor of viable mutational paths appears between them — the paper's headline counterintuitive result that rugged, hard-to-cross fitness landscapes can emerge WITHOUT any gene-gene interaction at all, just from a nonlinear fitness function on additive traits (arXiv:2505.14166).

Runner-up demos

• The Insurability Cliff: Price Your AI Agent — five sliders (autonomy, operational authority, permission exposure, governance maturity, dependency concentration) feed a real implementation of the paper's risk-state-to- premium mapping, with a live gauge and a "cliff" chart: crank permission exposure or autonomy past the governance-maturity line and the panel visibly flips into an "infeasible / uninsurable" red zone, reproducing the paper's proven monotone feasibility collapse (arXiv:2607.13230).
• The Sweet Spot: Optimal Complexity in Smart Materials — draggable log-scale points explore the paper's I1/I2/I3 complexity framework for responsive/smart materials, reproducing its reported order-of-magnitude bands (10¹⁸–10²⁰, 10¹⁰–10¹⁶, 10⁹–10¹⁰, 10⁵–10⁸, ER lipid ~10¹¹) exactly, while candidly labeling the illustrative curve where the paper gives no closed-form I3(N) (arXiv:2605.19795).