Fleming-Viot Selection of the Yaglom Limit for Age-Structured Bellman-Harris Processes, with Application to Livestock Epidemic Surveillance

What the math gives to ML

The paper develops a Fleming–Viot particle system that approximates the Yaglom or quasi-stationary distribution of an absorbing, age-structured branching process. Its transferable mechanism is to maintain an ensemble of stochastic trajectories, kill members that enter an absorbing failure state, and replace each killed member by a copy of a surviving member. A neural implementation can use this mechanism to maintain robust optimizer or inference trajectories conditioned on avoiding divergence, NaNs, excessive loss, or constraint violations. The quantitative signatures are exponential convergence of the conditional survivor law and polynomial reduction of finite-ensemble error with particle count.