High-dimensional particle filters get improved spectral correction method
Regularized free deconvolution for spectral repair of resampling distortion in high-dimensional particle filters
Machine Learning
Summary
Particle filters are tools to estimate changing systems but struggle when dealing with many variables at once, causing distortions in their calculations. The authors study how to recover and correct these distortions using a mathematical approach called regularized free deconvolution. They propose a new method that repairs the filter’s internal calculations without biasing the results, improving efficiency and accuracy. Their tests show the method greatly reduces errors and helps the filter work better in complex scenarios.
What this means in practice
- •For machine learning engineers: Improve state estimation in high-dimensional sensor fusion systems by reducing resampling distortions in particle filters.
- •For financial risk modelers: Increase accuracy and efficiency in particle filter-based models for complex market risk predictions involving many variables.
Authors
Deming Song, Chi-Kin Lam, Kefeng Li, Hui Yu, Xilin Li, Huagang Lin
Abstract
Resampling is the standard remedy for weight degeneracy in particle filters, but in high dimensions it distorts the spectrum of the ensemble covariance in a way that is computable from the importance weights. This paper studies the inverse problem: recovering the pre-resampling spectrum from the observed weights and the empirical spectrum alone, and using it without biasing the filter. We first show that the forward spectral law extends to random importance weights once one conditions on the weights together with the realized offspring counts, rather than on the weights alone. We then construct a regularized free-deconvolution estimator, prove its consistency with an explicit error budget at a fixed resolution scale, and certify model mismatch through a computable forward residual. The repaired spectrum enters the filter only as the preconditioner of a Metropolis-adjusted rejuvenation move, so that spectral error affects mixing efficiency but not the invariant target. For the full filtering cycle, we derive a geometrically summable error recursion under stable scalar dynamics and an explicit law for the cost of abstaining from repair. Pre-registered simulations show that the inversion reduces the spectral error by an order of magnitude, that the corrected preconditioner recovers about 90% of the oracle's sampling efficiency in an anisotropic regime, and that the cost of abstention depends on when abstentions occur rather than on how many. The repair recovers eigenvalue laws, not eigenvectors, and is certified only within an explicitly stated exogenous-weight regime.