MapReduce particle filtering with exact resampling and deterministic runtime

Abstract

Particle filtering is a numerical Bayesian technique that has great potential for solving sequential estimation problems involving non-linear and non-Gaussian models. Since the estimation accuracy achieved by particle filters improves as the number of particles increases, it is natural to consider as many particles as possible. MapReduce is a generic programming model that makes it possible to scale a wide variety of algorithms to Big data. However, despite the application of particle filters across many domains, little attention has been devoted to implementing particle filters using MapReduce. In this paper, we describe an implementation of a particle filter using MapReduce. We focus on a component that what would otherwise be a bottleneck to parallel execution, the resampling component. We devise a new implementation of this component, which requires no approximations, has O(N) spatial complexity and deterministic O (log N) 2 time complexity. Results demonstrate the utility of this new component and culminate in consideration of a particle filter with 2 24 particles being distributed across 512 processor cores.

View Priority Based Threads Allocation and Binary Search Oriented Reweight for GPU Accelerated Real-Time 3D Ball Tracking

“…, the computing complexity of the CDF building processing has always been O(N). Some work [17], [20], [21] proposed methods of sum reduction [26] in reweight and resampling, however, they only reduced time consumption in achieving P i=0 (W (i) k ). As Fig.…”

“…However, when assigning the multi-view algorithm, repeated work for each thread corresponding to each particle still exists. Lastly, some work [17], [20], [21] has been focusing on this problem in reweight task by applying sum reduction method [26]. Although they have reduced the time complexity by achieving partial parallelism summation for reweight, time complexity was still large.…”

Underwater Pure Orientation Target Tracking Based on Adaptive Box Length KLD PF Algorithm

“…Particle Filter (PF) is a filtering method based on Monte Carlo (MC) and recursive Bayesian algorithm [10], [11] which can more accurately represent the real-time state evolution of nonlinear stochastic processes [12]. Due to its characteristics of non-parametric and highly nonlinear conditions, PF greatly improves the accuracy, real-time and robustness of the algorithm, and is very suitable for dealing with pure azimuth tracking problems [13], [14].…”

Parallelised Bayesian optimisation for deep learning

“…The time complexity of the fully distributed cumulative sum is O(log N ). A deterministic redistribute method implies balanced computational load with deterministic runtime (e.g., [97] [49]), and a deterministic runtime is essential for real-time applications.…”

“…In Section 4.2, the parallel resampling algorithms is reviewed. Section 4.3 includes the proposed novel parallel resampling implemented in MapReduce, and Section 4.3 is based on [97]. The results indicate that Apache Spark provides significantly better runtime over the Apache Hadoop implementation as it enables the random access memory (RAM) to store the data.…”