pith. sign in

arxiv: 2209.12245 · v3 · submitted 2022-09-25 · 📡 eess.SP

Decentralised possibilistic inference with applications to target tracking

Jeremie Houssineau , Han Cai , Murat Uney , Emmanuel Delande This is my paper

Pith reviewed 2026-05-24 11:35 UTC · model grok-4.3

classification 📡 eess.SP
keywords decentralised inferencepossibility theorytarget trackingBernoulli filtersensor fusiondata associationGaussian mixture
0
0 comments X

The pith

Possibility theory supplies a fusion rule for sensor networks that recovers the exact centralised posterior as the number of sources grows.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to replace ad-hoc averaging of probability distributions with a fusion operation drawn from possibility theory. This operation is constructed so that the combined result approaches the posterior that would have been obtained by a single central processor given all the raw data. The rule is then inserted into a hierarchical Bernoulli filter that performs joint data association and state estimation for multiple targets. The resulting decentralised tracker keeps each local posterior independent of the others during fusion and still records lower cardinality and localisation errors than either geometric or arithmetic averaging of Gaussian-mixture approximations.

Core claim

The authors derive a fusion rule inside possibility theory that is asymptotically exact: when every local filter processes the same sequence of measurements, the fused possibility distribution converges to the posterior of the optimal centralised possibilistic filter while the local posteriors remain independent throughout the process.

What carries the argument

the possibilistic fusion rule that combines local possibility distributions without destroying source independence

If this is right

  • Local posteriors stay independent after fusion, removing the need for covariance-consistency corrections.
  • Joint data association and state estimation occur inside the same hierarchical structure at every node.
  • The method remains applicable once each local filter is replaced by a Gaussian-mixture representation.
  • Cardinality and position errors fall below those of both geometric-mean and arithmetic-mean probabilistic fusion under identical mixture approximations.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same fusion rule could be inserted into other recursive filters that admit a possibility-theoretic formulation.
  • Communication load could be further reduced by transmitting only the parameters of the fused possibility distribution rather than raw measurements.
  • The asymptotic exactness property supplies a concrete benchmark against which any future decentralised rule can be tested by scaling the sensor count.

Load-bearing premise

The Gaussian-mixture approximations needed to run the hierarchical Bernoulli filter do not erase the claimed performance advantage over probabilistic fusion methods.

What would settle it

A Monte-Carlo experiment in which the cardinality and localisation error of the decentralised possibilistic tracker fails to approach the error of the corresponding centralised possibilistic tracker as the number of sensors is increased while all sensors observe the same target trajectory.

Figures

Figures reproduced from arXiv: 2209.12245 by Emmanuel Delande, Han Cai, Jeremie Houssineau, Murat Uney.

Figure 1
Figure 1. Figure 1: 3-stage process to split and merge the information in a [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Performance averaged over 1000 Monte Carlo runs. problems, significant gains were observed for decentralised networks. Future work includes improving the robustness to potential calibration errors of some of the sensors, hence leveraging the capabilities of possibility theory in terms of robust inference [34], [35]. REFERENCES [1] M. Cetin, L. Chen, J. W. Fisher, A. T. Ihler, R. L. Moses, M. J. Wainwright,… view at source ↗
read the original abstract

Fusing and sharing information from multiple sensors over a network is a challenging task, partly due to the absence of a foundational rule for fusing probability distributions that preserves the independence of sources. To address this, we propose a decentralised inference framework based on possibility theory. Unlike probabilistic approaches that rely on ad-hoc averaging, we derive a principled fusion rule that is proven to be asymptotically exact, meaning it recovers the posterior of the optimal centralised possibilistic approach. We apply this rule to the possibilistic Bernoulli filter, leveraging its hierarchical nature to jointly infer data association and state estimation, distinct from standard decentralised Kalman filtering. We demonstrate that the proposed approach maintains the independence of local posteriors during fusion and, even under necessary approximations to handle Gaussian mixtures, significantly outperforms probabilistic geometric and arithmetic average fusion baselines in terms of cardinality and localisation error.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes a decentralised possibilistic inference framework for multi-sensor fusion. It derives a principled fusion rule claimed to be asymptotically exact (recovering the centralised possibilistic posterior while preserving local posterior independence) and applies it to a possibilistic Bernoulli filter for target tracking, enabling joint data association and state estimation. The approach is reported to outperform probabilistic geometric and arithmetic average fusion baselines even after necessary Gaussian mixture approximations.

Significance. If the asymptotic exactness is established and shown to be robust to the required approximations, the work supplies a non-ad-hoc alternative to averaging-based fusion rules in decentralised settings. The use of the hierarchical Bernoulli filter structure for joint inference is a concrete strength, and the emphasis on preserving source independence addresses a known limitation of probabilistic methods.

major comments (2)
  1. [Abstract] Abstract: the central claim that the derived fusion rule is 'proven to be asymptotically exact' and recovers the centralised posterior is load-bearing for the contribution. The same paragraph states that the rule is demonstrated 'even under necessary approximations to handle Gaussian mixtures,' yet provides no indication that the proof was re-derived or shown invariant under truncation and merging steps. If the proof relies on closure properties that the GM representation violates, the recovery guarantee does not hold as the number of sensors grows.
  2. [Application to possibilistic Bernoulli filter] Application section (Bernoulli filter implementation): the hierarchical structure is used for joint data association and state estimation, but the manuscript supplies no error analysis, invariance argument, or bound quantifying how the GM approximations affect the asymptotic property or the reported performance gains over the probabilistic baselines. This omission directly undermines the claim that outperformance is maintained under the approximations.
minor comments (2)
  1. [Fusion rule derivation] Clarify the precise definition of the possibilistic fusion operator and its relation to the centralised posterior in the derivation; notation for the possibility measure should be consistent across sections.
  2. [Numerical results] Experimental results: report the number of Monte Carlo trials, exact parameter settings for the GM merging thresholds, and the specific cardinality and localisation error metrics used for the baselines.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight important points regarding the scope of our theoretical claims under practical approximations. We address each major comment below and will revise the manuscript to improve clarity.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim that the derived fusion rule is 'proven to be asymptotically exact' and recovers the centralised posterior is load-bearing for the contribution. The same paragraph states that the rule is demonstrated 'even under necessary approximations to handle Gaussian mixtures,' yet provides no indication that the proof was re-derived or shown invariant under truncation and merging steps. If the proof relies on closure properties that the GM representation violates, the recovery guarantee does not hold as the number of sensors grows.

    Authors: We agree that the asymptotic exactness proof applies to the exact possibilistic fusion rule and is established prior to introducing any approximations. The Gaussian mixture approximations are a standard computational necessity in the Bernoulli filter to handle the mixture representations arising from the update step; they are applied after the fusion rule derivation. The proof does not depend on GM closure properties, as it operates at the level of general possibility distributions. To address the concern, we will revise the abstract to explicitly distinguish the exact theoretical guarantee from the empirical results obtained under approximation, and add a clarifying sentence in the application section. revision: yes

  2. Referee: [Application to possibilistic Bernoulli filter] Application section (Bernoulli filter implementation): the hierarchical structure is used for joint data association and state estimation, but the manuscript supplies no error analysis, invariance argument, or bound quantifying how the GM approximations affect the asymptotic property or the reported performance gains over the probabilistic baselines. This omission directly undermines the claim that outperformance is maintained under the approximations.

    Authors: The referee is correct that the manuscript provides no formal error analysis, invariance argument, or quantitative bound on the effect of the GM truncation and merging steps on the asymptotic exactness or on the observed performance advantage. We will add a dedicated paragraph in the application section acknowledging this limitation, noting that the approximations follow standard GM filter practices and that the reported gains are empirical. Deriving a rigorous bound lies outside the scope of the present work. revision: partial

Circularity Check

0 steps flagged

Derivation of fusion rule presented as independent from possibility theory principles

full rationale

The paper states that a principled fusion rule is derived and proven asymptotically exact, recovering the centralised possibilistic posterior while preserving source independence. No quoted equations or steps in the abstract reduce this result to a fitted parameter, self-defined quantity, or load-bearing self-citation chain. The approximations for Gaussian mixtures are flagged as implementation necessities but are not shown to be inputs that force the exactness claim by construction. The central claim therefore remains self-contained against external benchmarks in possibility theory.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Possibility theory is treated as established background; the new contribution is the fusion rule itself. No free parameters, invented entities, or additional axioms are identifiable from the abstract alone.

axioms (1)
  • domain assumption Possibility theory supplies a fusion operation that preserves source independence where probability does not.
    Invoked to motivate and derive the decentralised rule.

pith-pipeline@v0.9.0 · 5676 in / 1218 out tokens · 48831 ms · 2026-05-24T11:35:09.701945+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    Distributed fusion in sensor networks,

    M. Cetin, L. Chen, J. W. Fisher, A. T. Ihler, R. L. Moses, M. J. Wainwright, and A. S. Willsky, “Distributed fusion in sensor networks,” IEEE Signal Processing Magazine , vol. 23, no. 4, pp. 42–55, 2006

    work page 2006

  2. [2]

    Topology for distributed inference on graphs,

    S. Kar, S. Aldosari, and J. M. Moura, “Topology for distributed inference on graphs,” IEEE Transactions on Signal Processing , vol. 56, no. 6, pp. 2609–2613, 2008

    work page 2008

  3. [3]

    Monte Carlo optimization of decentralized estimation networks over directed acyclic graphs under communication constraints,

    M. Uney and M. Cetin, “Monte Carlo optimization of decentralized estimation networks over directed acyclic graphs under communication constraints,” IEEE Trans. on Signal Proc. , vol. 59, no. 11, pp. 5558– 5576, 2011

    work page 2011

  4. [4]

    Consensus + innovations distributed inference over networks: cooperation and sensing in networked systems,

    S. Kar and J. M. Moura, “Consensus + innovations distributed inference over networks: cooperation and sensing in networked systems,” IEEE Signal Processing Magazine , vol. 30, no. 3, pp. 99–109, 2013

    work page 2013

  5. [5]

    Distributed fusion architectures and algorithms for target tracking,

    M. E. Liggins, C.-Y . Chong, I. Kadar, M. G. Alford, V . Vannicola, and S. Thomopoulos, “Distributed fusion architectures and algorithms for target tracking,” Proc. of the IEEE , vol. 85, no. 1, pp. 95–107, 1997

    work page 1997

  6. [6]

    Hall, C.-Y

    D. Hall, C.-Y . Chong, J. Llinas, and M. Liggins II, Distributed data fusion for network-centric operations . Crc Press, 2017

    work page 2017

  7. [7]

    Distributed fusion of PHD filters via exponential mixture densities,

    M. ¨Uney, D. E. Clark, and S. J. Julier, “Distributed fusion of PHD filters via exponential mixture densities,” IEEE Journal of Selected Topics in Signal Processing, vol. 7, no. 3, pp. 521–531, 2013

    work page 2013

  8. [8]

    On arithmetic average fusion and its application for distributed multi-Bernoulli multitarget tracking,

    T. Li, X. Wang, Y . Liang, and Q. Pan, “On arithmetic average fusion and its application for distributed multi-Bernoulli multitarget tracking,” IEEE Transactions on Signal Processing, vol. 68, pp. 2883–2896, 2020

    work page 2020

  9. [9]

    Elements of asymptotic the- ory with outer probability measures,

    J. Houssineau, N. Chada, and E. Delande, “Elements of asymptotic the- ory with outer probability measures,” arXiv preprint arXiv:1908.04331, 2019

    work page arXiv 1908

  10. [10]

    Parameter estimation with a class of outer probability measures,

    J. Houssineau, “Parameter estimation with a class of outer probability measures,” arXiv preprint arXiv:1801.00569 , 2018

    work page arXiv 2018

  11. [11]

    Possibility theory and its applications: Where do we stand?,

    D. Dubois and H. Prade, “Possibility theory and its applications: Where do we stand?,” in Springer Handbook of Computational Intelligence , pp. 31–60, Springer, 2015

    work page 2015

  12. [12]

    Target tracking in the framework of possibility theory: The possibilistic Bernoulli filter,

    B. Ristic, J. Houssineau, and S. Arulampalam, “Target tracking in the framework of possibility theory: The possibilistic Bernoulli filter,” Information Fusion, vol. 62, pp. 81–88, 2020

    work page 2020

  13. [13]

    A linear algorithm for multi-target tracking in the context of possibility theory,

    J. Houssineau, “A linear algorithm for multi-target tracking in the context of possibility theory,” IEEE Transactions on Signal Processing , vol. 69, pp. 2740–2751, 2021

    work page 2021

  14. [14]

    Possibility generalized labeled multi-bernoulli filter for multi-target tracking under epistemic uncertainty,

    H. Cai, J. Houssineau, B. A. Jones, M. Jah, and J. Zhang, “Possibility generalized labeled multi-bernoulli filter for multi-target tracking under epistemic uncertainty,” IEEE Transactions on Aerospace and Electronic Systems (Early Access) , 2022

    work page 2022

  15. [15]

    Possibility theory and data fusion in poorly informed environments,

    D. Dubois and H. Prade, “Possibility theory and data fusion in poorly informed environments,” Control Engineering Practice , vol. 2, no. 5, pp. 811–823, 1994

    work page 1994

  16. [16]

    Merging fuzzy information,

    D. Dubois, H. Prade, and R. Yager, “Merging fuzzy information,” in Fuzzy sets in approximate reasoning and information systems , pp. 335– 401, Springer, 1999

    work page 1999

  17. [17]

    Tracking and data association,

    Y . Bar-Shalom, T. E. Fortmann, and P. G. Cable, “Tracking and data association,” Journal of the Acoustical Society of America, vol. 87, no. 2, pp. 918–919, 1990

    work page 1990

  18. [18]

    R. P. S. Mahler, Statistical Multisource-Multitarget Information Fusion . Artech House, 2007

    work page 2007

  19. [19]

    A tutorial on Bernoulli filters: theory, implementation and applications,

    B. Ristic, B.-T. V o, B.-N. V o, and A. Farina, “A tutorial on Bernoulli filters: theory, implementation and applications,” IEEE Transactions on Signal Processing, vol. 61, no. 13, pp. 3406–3430, 2013

    work page 2013

  20. [20]

    Distributed Kalman filtering and control through embedded average consensus information fusion,

    S. P. Talebi and S. Werner, “Distributed Kalman filtering and control through embedded average consensus information fusion,” IEEE Trans- actions on Automatic Control , vol. 64, no. 10, pp. 4396–4403, 2019

    work page 2019

  21. [21]

    Consensus Bernoulli filter for distributed detection and tracking using multi-static doppler shifts,

    M. B. Guldogan, “Consensus Bernoulli filter for distributed detection and tracking using multi-static doppler shifts,” IEEE Signal Processing Letters, vol. 21, no. 6, pp. 672–676, 2014

    work page 2014

  22. [22]

    Distributed Bernoulli filtering for target detection and tracking based on arithmetic average fusion,

    T. Li, Z. Liu, and Q. Pan, “Distributed Bernoulli filtering for target detection and tracking based on arithmetic average fusion,” IEEE Signal Processing Letters, vol. 26, no. 12, pp. 1812–1816, 2019

    work page 2019

  23. [23]

    Conditioning in possibility theory with strict order norms,

    B. De Baets, E. Tsiporkova, and R. Mesiar, “Conditioning in possibility theory with strict order norms,” Fuzzy Sets and Systems , vol. 106, no. 2, pp. 221–229, 1999

    work page 1999

  24. [24]

    Possibility theory, probability and fuzzy sets misunderstandings, bridges and gaps,

    D. Dubois, H. T. Nguyen, and H. Prade, “Possibility theory, probability and fuzzy sets misunderstandings, bridges and gaps,” in Fundamentals of fuzzy sets , pp. 343–438, Springer, 2000

    work page 2000

  25. [25]

    Smoothing and filtering with a class of outer measures,

    J. Houssineau and A. Bishop, “Smoothing and filtering with a class of outer measures,” SIAM/ASA Journal on Uncertainty Quantification , vol. 6, no. 2, pp. 845–866, 2018

    work page 2018

  26. [26]

    E. T. Jaynes, Probability theory: The logic of science . Cambridge university press, 2003

    work page 2003

  27. [27]

    A non-divergent estimation algorithm in the presence of unknown correlations,

    S. J. Julier and J. K. Uhlmann, “A non-divergent estimation algorithm in the presence of unknown correlations,” in Proceedings of the 1997 American Control Conference, vol. 4, pp. 2369–2373, IEEE, 1997

    work page 1997

  28. [28]

    Estimation under unknown correlation: Covariance intersection revisited,

    L. Chen, P. O. Arambel, and R. K. Mehra, “Estimation under unknown correlation: Covariance intersection revisited,” IEEE Transactions on Automatic Control, vol. 47, no. 11, pp. 1879–1882, 2002

    work page 2002

  29. [29]

    Possibilistic information fusion using maximal coherent subsets,

    S. Destercke, D. Dubois, and E. Chojnacki, “Possibilistic information fusion using maximal coherent subsets,” IEEE Transactions on Fuzzy Systems, vol. 17, no. 1, pp. 79–92, 2008

    work page 2008

  30. [30]

    A survey of distributed data aggregation algorithms,

    P. Jesus, C. Baquero, and P. S. Almeida, “A survey of distributed data aggregation algorithms,” IEEE Communications Surveys & Tutorials , vol. 17, no. 1, pp. 381–404, 2014

    work page 2014

  31. [31]

    Development of track to track fusion algorithms,

    K. H. Kim, “Development of track to track fusion algorithms,” in Proceedings of 1994 American Control Conference-ACC’94 , vol. 1, pp. 1037–1041, IEEE, 1994

    work page 1994

  32. [32]

    Distributed linear estimation over sensor networks,

    G. C. Calafiore and F. Abrate, “Distributed linear estimation over sensor networks,” International Journal of Control, vol. 82, no. 5, pp. 868–882, 2009. 12

    work page 2009

  33. [33]

    A consistent metric for performance evaluation of multi-object filters,

    D. Schuhmacher, B.-T. V o, and B.-N. V o, “A consistent metric for performance evaluation of multi-object filters,” IEEE transactions on Signal Processing, vol. 56, no. 8, pp. 3447–3457, 2008

    work page 2008

  34. [34]

    Robust target motion analysis using the possibility particle filter,

    B. Ristic, J. Houssineau, and S. Arulampalam, “Robust target motion analysis using the possibility particle filter,” IET Radar, Sonar & Navigation, vol. 13, no. 1, pp. 18–22, 2019

    work page 2019

  35. [35]

    Robust bayesian inference in complex models with possibility theory,

    J. Houssineau and D. J. Nott, “Robust bayesian inference in complex models with possibility theory,” arXiv preprint arXiv:2204.06911, 2022. APPENDIX A. Likelihood for the possibilistic Bernoulli filter We first consider separately the probabilistic and possibilis- tic components of the likelihood. We model the uncertainty about the detection of the target...

    work page arXiv 2022