Orestis Kaparounakis submitted a paper presenting a Bayesian filtering technique using processor-native uncertainty tracking for inference in dynamic systems. The approach achieves "deterministic approximate filtering" with up to 805x average speedup over direct Monte Carlo methods at matched result quality. Benchmarks across three nonlinear state-space systems show Pareto-dominant accuracy-latency trade-offs for posterior inference while remaining competitive in RMSE with baseline particle filters.
No score is assigned. Sources and their independence are shown in the citation chain below.