Stochastic filtering under model ambiguity (Q6180475)

From MaRDI portal
scientific article; zbMATH DE number 7791632
Language Label Description Also known as
English
Stochastic filtering under model ambiguity
scientific article; zbMATH DE number 7791632

    Statements

    Stochastic filtering under model ambiguity (English)
    0 references
    19 January 2024
    0 references
    The filtering (estimation) problem for systems described by stochastic differential equations has been studied extensively since the principal works of \textit{R. L. Stratonovich} [Teor. Veroyatn. Primen. 5, 172--195 (1960; Zbl 0094.13101)] and \textit{H. J. Kushner} [J. Differ. Equations 3, 179--190 (1967; Zbl 0158.16801)]. The optimal filtering equation is a non-linear stochastic partial differential equation (SPDE), which is usually called the Kushner--Stratonovich equation or the Kushner--FKK equation. The developed classical filtering (estimation) methods are not directly applicable in practice since the exact spectral structure of both the signal and observation processes is not usually available. Therefore, it is reasonable to consider the estimates, called minimax-robust, which minimize the maximum of the errors for all spectral measures from a given set of admissible spectral measures simultaneously. The authors of this article investigate the filtering problem under model uncertainty for the model with real-valued signal and observation processes described by the equations \[ \begin{cases} dX_t=b(X_t)dt+\sigma(X_t)dW_t,\quad &X_0=x,\\ dY_t=h(X_t)dt+dB_t,\quad &Y_0=0, \end{cases} \] where the coefficients \(b,\sigma,h\) are continuous real functions, \((W_t,B_t)\) is a 2-dimensional standard Brownian motion under probability measure \(P\in\mathcal P\). The probability measure \(P\) serves as an evaluation criterion for the signal process observed by external observers. The probability measure set \(\mathcal P\) is considered to encompass all evaluation criteria with the ambiguity parameters \(\theta\in\Theta\). The estimate of the signal process is obtained by minimizing the squared error in the worst-case scenario \[\min_{\xi}\max_{P\in\mathcal P} \mathbb E^P[|X_t-\xi|^2],\] where \(\mathbb E^P\) is the expectation with respect to the probability measure \(P\), and \(\xi\) is over all \(\mathcal G\)-measurable random variables. Here \(\mathcal{G}\equiv\sigma (Y_s:s\leq t)\). The authors start with proving the existence of the optimal control that minimizes the error under the most unfavorable evaluation criteria, relying on partially observable information. Then, by utilizing the mini-max theorem, they interchanged the order of extremum problems and characterized the optimal control. The most favorable evaluation criteria, namely, the optimal probability measure is described. For more related results and references see [\textit{S. A. Kassam} and \textit{H. V. Poor}, Proc. IEEE 73, No. 3, 433--481 (1985; Zbl 0569.62084); \textit{M. Moklyachuk}, ``Minimax-robust estimation problems for stationary stochastic sequences'', Stat. Optim. Inf. Comput. 3, No. 4, 348--419 (2015); \textit{M. Luz} and \textit{M. Moklyachuk}, Estimation of stochastic processes with stationary increments and cointegrated sequences. London: ISTE; Hoboken, NJ: John Wiley \& Sons (2019; Zbl 1430.62007)].
    0 references
    stochastic control theory
    0 references
    nonlinear filtering
    0 references
    uncertainty
    0 references
    min-max theorem
    0 references
    weighted mean-field system
    0 references
    0 references
    0 references
    0 references

    Identifiers

    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references