Technical Benchmark: BIDSim vs. LIIS
Year: 2026
Type: article
Abstract: This benchmark evaluates BIDSim (Chained Virtual Urns Simulator) against Laplace-Informed Importance Sampling (LIIS) across high-dimensional categorical probability spaces, simplex boundary constraints, Small Data robustness, computational efficiency, and client-side edge deployment suitability. While LIIS provides a local continuous Gaussian proposal around the maximum a posteriori (MAP) mode, it suffers from severe importance weight collapse, Gaussian tail leakage across simplex boundaries, and optimizer instability on Small Data. In contrast, BIDSim achieves zero autocorrelation (lag-1 = 0.000), 100% Effective Sample Size (ESS) efficiency, exact physical probability conservation, and microsecond execution speeds in a sub-154 KB WebAssembly binary.
Author: Pargamin, Bernard
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225752768
FWCI:
Cites:
Cited by:
Related to: