Razvan Stefanescu

Dr. Razvan Stefanescu

orcid

Research Focus & Metrics

Academic Career Length

13Years (2014-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 August 2014 in International Journal for Numerical Methods in Fluids
Verified
Journal Article
Closed access
PDF file
RSRazvan Stefanescu
Razvan Stefanescu
ASAdrian Sandu
Adrian Sandu
INIonel Michael Navon
Ionel Michael Navon

9 October 2017 in International Journal for Numerical Methods in Engineering
Verified
Journal Article
Open access
PDF file
AMAzam Moosavi
Azam Moosavi
RSRazvan Stefanescu
Razvan Stefanescu
ASAdrian Sandu
Adrian Sandu
Summary: This paper introduces multivariate input‐output models to predict the errors and bases dimensions of local parametric Proper Orthogonal Decomposition reduced‐order models. We refer to these mappings as the multivariate predictions of local reduced‐order model characteristics (MP‐LROM) models. We use Gaussian processes and artificial neural networks to construct approximations of these multivariate mappings. Numerical results with a viscous Burgers model illustrate the performance and potential of the machine learning‐based regression MP‐LROM models to approximate the characteristics of parametric local reduced‐order models. The predicted reduced‐order models errors are compared against the multifidelity correction and reduced‐order model error surrogates methods predictions, whereas the predicted reduced‐order dimensions are tested against the standard method based on the spectrum of snapshots matrix. Since the MP‐LROM models incorporate more features and elements to construct the probabilistic mappings, they achieve more accurate results. However, for high‐dimensional parametric spaces, the MP‐LROM models might suffer from the curse of dimensionality. Scalability challenges of MP‐LROM models and the feasible ways of addressing them are also discussed in this study.

6 February 2017 in International Journal for Numerical Methods in Engineering
Verified
Journal Article
Open access
PDF file
RSRazvan Stefanescu
Razvan Stefanescu
KSKathleen Schmidt
Kathleen Schmidt
JHJason Hite
Jason Hite
RSRalph C Smith
Ralph C Smith
JMJohn Mattingly
John Mattingly
We propose several algorithms to recover the location and intensity of a radiation source located in a simulated 250 × 180 m block of an urban center based on synthetic measurements. Radioactive decay and detection are Poisson random processes, so we employ likelihood functions based on this distribution. Owing to the domain geometry and the proposed response model, the negative logarithm of the likelihood is only piecewise continuous differentiable, and it has multiple local minima. To address these difficulties, we investigate three hybrid algorithms composed of mixed optimization techniques. For global optimization, we consider simulated annealing, particle swarm, and genetic algorithm, which rely solely on objective function evaluations; that is, they do not evaluate the gradient in the objective function. By employing early stopping criteria for the global optimization methods, a pseudo-optimum point is obtained. This is subsequently utilized as the initial value by the deterministic implicit filtering method, which is able to find local extrema in non-smooth functions, to finish the search in a narrow domain. These new hybrid techniques, combining global optimization and implicit filtering address, difficulties associated with the non-smooth response, and their performances, are shown to significantly decrease the computational time over the global optimization methods. To quantify uncertainties associated with the source location and intensity, we employ the delayed rejection adaptive Metropolis and DiffeRential Evolution Adaptive Metropolis algorithms. Marginal densities of the source properties are obtained, and the means of the chains compare accurately with the estimates produced by the hybrid algorithms. Copyright © 2016 John Wiley & Sons, Ltd.

4 December 2019 in Remote Sensing
Verified
Journal Article
Open access
PDF file
MGMichael Gorbunov
Michael Gorbunov
RSRazvan Stefanescu
Razvan Stefanescu
VIVladimir Irisov
Vladimir Irisov
DZDusanka Zupanski
Dusanka Zupanski
We review different approaches to the variational assimilation of radio occultation (RO) observations into models of global atmospheric circulation. We derive the general equation for the bending angle that reduces to the Abel integral for a spherically layered atmosphere. We review the full 3-D observation operator for bending angles, which provides the strictest solution, but is also most computationally expensive. Commonly used is the 2-D approximation that allows treating rays as plane curve. We discuss a simple 1-D approach to the assimilation of bending angles. The observation operator based on the standard form of the Abel integral has a disadvantage, because it cannot account for waveguides. Alternative approaches use 1-D ray-tracing. The most straightforward way is to use the same framework as for the 3-D observation operator, with the refractivity field reduced to a single profile independent from the horizontal coordinates. An alternative 1-D ray-tracing approach uses the form of ray equation in a spherically layered medium that uses an invariant. The assimilation of refractivity has also 1-D and 3-D options. We derive a new simple form of the refractivity-mapping operator. We present the results of numerical tests of different 3-D and 1-D observation operators, based on Spire data.

17 May 2023 in Proceedings of the ACM on Human-Computer Interaction
Verified
Journal Article
Open access
SASamantha Aziz
Samantha Aziz
DLDillon J. Lohr
Dillon J. Lohr
RSRazvan Stefanescu
Razvan Stefanescu
OKOleg Komogortsev
Oleg Komogortsev
This paper proposes a novel evaluation framework, termed "critical evaluation periods," for evaluating continuous gaze prediction models. This framework emphasizes prediction performance when it is most critical for gaze prediction to be accurate relative to user perception. Based on perceptual characteristics of the human visual system such as saccadic suppression, this framework provides a more practical assessment of gaze prediction performance for gaze-contingent rendering compared to the dominant sample-by-sample evaluation strategy employed in literature, which overemphasizes performance during easy-to-predict periods of fixation. Using a case study with a lightweight deep learning gaze prediction model, we observe a significant discrepancy in the reported prediction accuracy between the proposed critical evaluation periods and the dominant evaluation strategy employed in literature. Based on our findings, we suggest that the proposed framework is more suitable for evaluating the performance of continuous gaze prediction models intended for gaze-contingent rendering applications.

12 December 2019 in Algorithms
Verified
Journal Article
Open access
PDF file
JCJared A. Cook
Jared A. Cook
RSRalph C Smith
Ralph C Smith
JHJason M. Hite
Jason M. Hite
RSRazvan Stefanescu
Razvan Stefanescu
JMJohn Mattingly
John Mattingly
Surrogate models are increasingly required for applications in which first-principles simulation models are prohibitively expensive to employ for uncertainty analysis, design, or control. They can also be used to approximate models whose discontinuous derivatives preclude the use of gradient-based optimization or data assimilation algorithms. We consider the problem of inferring the 2D location and intensity of a radiation source in an urban environment using a ray-tracing model based on Boltzmann transport theory. Whereas the code implementing this model is relatively efficient, extension to 3D Monte Carlo transport simulations precludes subsequent Bayesian inference to infer source locations, which typically requires thousands to millions of simulations. Additionally, the resulting likelihood exhibits discontinuous derivatives due to the presence of buildings. To address these issues, we discuss the construction of surrogate models for optimization, Bayesian inference, and uncertainty propagation. Specifically, we consider surrogate models based on Legendre polynomials, multivariate adaptive regression splines, radial basis functions, Gaussian processes, and neural networks. We detail strategies for computing training points and discuss the merits and deficits of each method.