Approximating Analytically-Intractable Likelihood Densities with Deterministic Arithmetic for Optimal Particle Filtering

Orestis Kaparounakis, Yunqi Zhang, Phillip Stanley-Marbell · 2025

Read Approximating Analytically-Intractable Likelihood Densities with Deterministic Arithmetic for Optimal Particle Filtering by Orestis Kaparounakis, Yunqi Zhang, Phillip Stanley-Marbell (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-01023.txt.

Introduction

Orestis Kaparounakis, Yunqi Zhang, and Phillip Stanley-Marbell improve particle filtering in a 2025 paper. Particle filters track hidden states from noisy observations and power tools in self-driving cars, drones, warehouse robots, target tracking, and patient monitoring. They keep many candidate states, called particles, and weight them by how well each matches new data.

Their weakness is representing that match, the likelihood, which can cause particles to degenerate into a few survivors. The authors approximate hard likelihood densities with deterministic arithmetic for optimal filtering. Better likelihood handling matters most for fast, resource-limited systems where filters must run in real time.

Catalog record

CREATOROrestis Kaparounakis, Yunqi Zhang, Phillip Stanley-Marbell
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE28 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-01023

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Approximating Analytically-Intractable Likelihood Densities with Deterministic Arithmetic for Optimal Particle Filtering Orestis Kaparounakis, Yunqi Zhang, Phillip Stanley-Marbell Abstract Particle filtering algorithms have enabled practical solutions to problems in autonomous robotics (self-driving cars, UAVs, warehouse robots), target tracking, and econometrics, with further applications in speech processing and medicine (patient monitoring). Yet, their inherent weakness at representing the likelihood of the observation (which often leads to particle degeneracy) remains unaddressed for high-frequency and resource-constrained systems. Improvements such as the optimal proposal and auxiliary particle filter mitigate this issue under specific circumstances and with increased computational cost. This work presents a new particle filtering method and its implementation, which enables tunably-approximative representation of arbitrary likelihood densities as program transformations of parametric distributions. Our method leverages a recent computing platform that can perform deterministic computation on probability distribution representations (UxHw) without relying on stochastic methods. For non-Gaussian non-linear systems and with an optimal-auxiliary particle filter, we benchmark the likelihood evaluation error and speed for a total of 294 840 evaluation points. For such models, the results show that the UxHw method leads to as much as 37.7x speedup compared to the Monte Carlo alternative. For narrow uniform observation noise, the particle filter falsely assigns zero likelihood as much as 81.89% of the time whereas UxHw achieves 1.52% false-zero rate. The UxHw approach achieves filter RMSE improvement of as much as 18.9% (average 3.3%) over the Monte Carlo alternative. I Introduction Particle filters [ 1 ] have become a mainstay for real-time state estimation in embedded systems because they handle nonlinear dynamics and non-Gaussian noise better than classical Kalman filters and variants [ 2 , 3 ] . Yet, the accuracy and performance of particle filters hinge on an explicit likelihood model [ 4 ] , which in sensor-rich, data-driven environments is often unavailable or intractable to formulate [ 5 , 6 , 7 ] . Computing the likelihood for analytically-intractable models can be computationally expensive: each particle may require solving an optimization problem or simulating a costly transformation [ 8 , 9 ] . These computations are ill-suited for resource-constrained embedded platforms under tight power and latency budgets This letter introduces a new method for particle filter implementation that enables easy marginalization over noise distributions, facilitating the computation of the likelihood density. The approach capitalizes on recent advances in hardware architectures for performing arithmetic on digital representations of probability density functions ( pdf s) [ 10 , 11 ] . Particle filters compute the likelihood density at the measurement to adjust the particle weights and thus update the state estimate. When the likelihood is hard to compute, this step becomes a bottleneck for accurate and performant state estimation [ 12 ] . This article introduces a technique for approximating the measurement likelihood for analytically-intractable probability distributions. The technique leverages direct arithmetic on probability density functions and density evaluation via uncertainty-extended hardware (UxHw) [ 10 ] . Here, we apply the technique for computing the optimal likelihood for the auxiliary particle filter [ 13 ] . This letter makes the following contributions to the state of the art: ➀ A tunably-approximative approach to computing likelihoods from arbitrary densities in particle filters. This method uses Turing-complete transformations on random variables in conventional programming language syntax. Building on functionality enabled by UxHw, it keeps likelihood models simple, digestible, and with clear semantic mapping to the sensor physics. The approach offers deterministic time guarantees, paving the way for use in real-time systems, without having to rely on complex Monte Carlo simulation schemes. (Section II ) ➁ Experimental validation of the accuracy, bare-metal speed, robustness, and convergence, of the proposed method for the optimal likelihood on the Gordon–Salmond–Smith system [ 1 ] , across 540 system configurations including Gaussian and non-Gaussian noise, for a total of 294\,840\text{\,}\mathrm{e} valuation points. (Section III ) The results show that the UxHw method enables Bayesian filtering for non-Gaussian non-linear analytically-intractable models on resource-constrained systems, such as drones and autonomous robots, as much as 37.7x faster than the status quo. This paves the way for real-time use of a new generation of powerful likelihood models. Listing 1: Baseline pointwise approach. ⬇ 1 double lyk_proxy_pointwise ( double z , double x , double t ) { 2 double expected_x = transition_model_ideal ( x , t ); 3 double expected_z = observation_model_ideal ( expected_x ); 4 return evaluate_density_analytical ( z , expected_z ); 5 } Listing 2: Monte Carlo simulation. ⬇ 1 double lyk_proxy_mc ( double z , double x , double t ) { 2 double accum = 0.0; 3 double expected_x = transition_model_ideal ( x , t ); 4 for ( size_t i = 0; i < M ; i ++) { 5 double sim_noise = sample ( transition_noise_model ()); 6 double sim_x = expected_x + sim_noise ; 7 double sim_expected_z = observation_model_ideal ( sim_x ); 8 double lk = evaluate_density_analytical ( z , sim_expected_z ); 9 accum += lk ; 10 } 11 return accum / ( double ) M ; // Return average Listing 3: UxHw approach. ⬇ 1 double lyk_proxy_uxhw ( double z , double x , double t ) { 2 double x_density = transition_model_ideal ( x , t ) + ➊ 3 transition_noise_model (); 4 double z_density = observation_model_ideal ( x_density ) + ➊ 5 observation_noise_model (); 6 return UxHwDoubleEvaluatePDF ➋ ( z , z_density ); 7 } Fig. 1 : C code for three approaches of the proxy likelihood evaluation: pointwise (Listing 1 ), Monte Carlo simulation (Listing 2 ) , and UxHw-based computation (Listing 3 ). ➊ On an uncertainty-tracking processor [ 10 ] such as the one in Figure 4 , the + operator adds the conventional values but also performs addition between the associated distributions for these values. ➋ UxHwDoubleEvaluatePDF is not a C function but rather a function-wrapped UxHw-microarchitecture instruction for evaluating the density. II Predictive-Lookahead Auxiliary Particle Filter Let x_{k} be the state of an one-dimensional Markov system at time k and let z_{k} denote the observation. Let \upsilon_{k}\sim\mu_{\upsilon} and \nu_{k}\sim\mu_{\nu} denote transition and observation noise respectively, with arbitrary distributions. Equations 1 and 2 are the state-space transition and observation models for the system: \displaystyle x_{k} \displaystyle=f(x_{k-1},\upsilon_{k}), (1) \displaystyle z_{k} \displaystyle=h(x_{k},\nu_{k}). (2) Let superscript (i) denote the particle index. The auxiliary particle filter computes, in a pointwise manner, the likelihood of the expected observation conditioned on the expected transition ( proxy likelihood , m_{k}^{(i)} ) to find the current particles that are probably useful for the next iteration (Listing 1 ). To quantify this it assigns respective relative weights ( lookahead weights , w_{k}^{(i)} ) (Equation 3 ) [ 13 , 14 ] . w_{k}^{(i)}\propto w_{k-1}^{(i)}\;p({z}_{k}\mid\mathbb{E}_{\mu_{\upsilon}}[X_{k}^{(i)}]). (3) In this letter, we examine an enhancement for the auxiliary particle filter which replaces this pointwise likelihood with a predictive likelihood marginalized on the transition noise distribution rather than conditioned on the transition noise expected value (Equation 4 ). This leads to proxy likelihood weights which are fully-informed on the transition model uncertainty and yields estimate posteriors that are more faithful to the system model under non-Gaussian non-linear system dynamics: \displaystyle w_{k}^{(i)} \displaystyle\propto w_{k-1}^{(i)}\;p({z}_{k}\mid{x}_{k-1}^{(i)}),\text{\hskip 10.00002ptwhere} (4) \displaystyle p({z}_{k}\mid{x}_{k-1}^{(i)}) \displaystyle=\int p({z}_{k}\mid{x}_{k})p({x}_{k}\mid{x}_{k-1}^{(i)})dx_{k}. (5) In general, Equation 5 does not have an analytic solution, so applications default to Equation

Read the complete text