Two Oden Institute Researchers Win Jean-Pierre Le Cadre Best Paper Award

Two University of Texas at Austin researchers have been honored with a Best Paper Award for their innovative research into non-Gaussian kernels for ensemble mixture model filtering. Renato Zanetti, associate professor of aerospace engineering & engineering mechanics, and Andrey Popov, postdoctoral fellow, both with the Oden Institute For Computational Engineering and Sciences, received the 2024 Jean-Pierre Le Cadre Best Paper Award given by the International Society of Information Fusion (ISIF). The award is a renowned recognition within the field of data fusion and state estimation. 

Their paper, "Are Non-Gaussian Kernels Suitable for Ensemble Mixture Model Filtering?" focuses on addressing a fundamental challenge in computational sciences: accurately estimating the state of complex systems where traditional methods struggle, especially when dealing with high-dimensional data.

While spacecraft navigation was a key focus of their research, the implications extend far beyond. As Popov describes, particle filters have potential applications in power grid monitoring, where limited sensor data makes accurate state estimation critical, and numerical weather forecasting, which involves integrating large models of the atmosphere with a wealth of observational data from radar, planes, satellites, and ships. 

At the core of their research is a class of algorithms known as particle filters, which approximate the results of Bayes’ theorem to infer unknown states from data/particle filters, widely used in fields ranging from aerospace navigation to power systems and numerical weather forecasting. However, they often fall short in high-dimensional applications due to computational complexity and the reliance on assumptions that are rarely met in practice. 

“Particle filters approach the holy grail: true Bayesian inference, meaning that we can approximate the results of Bayes’ theorem for any distribution with arbitrary accuracy. Most state of the art methods rely on assumptions and approximations that are always violated in practice, at least for high dimensional problems," said Popov.

This limitation is precisely what Zanetti and Popov set out to overcome with their research, exploring the use of non-Gaussian kernels — in particular, the Epanechnikov kernel — to improve the performance of ensemble filters in high-dimensional systems. "The topic I care about are extremely high-dimensional," Popov elaborates. "People like Renato Zanetti, in aerospace engineering, care about applications for tracking space objects like satellites, navigation onboard a spacecraft, and for its usefulness in guidance. Particle filters for this type of application rarely even dream of approaching the state of the art. And yet, that is the dream that many like me chase." In addition to being an aerospace professor, Zanetti is also a core faculty member at the Oden Institute.

The pair’s Ensemble Gaussian Mixture Filter (EnGMF) uses a fraction of the computational resources required by traditional methods to achieve similar accuracy, offering new hope for scaling up particle filters to tackle these large, complex problems. 

Looking ahead, Zanetti and Popov aim to prove that their newly proposed method, the Ensemble Epanechnikov Mixture Filter (EnEMF), intersects to exact Bayesian inference in distribution, while also providing more robust results for high-dimensional filtering. 

Zanetti shared his thoughts on their research direction: “The Gaussian assumption is so convenient that became almost ubiquitous, even when working with non-Gaussian distributions we often end up representing them as a sum of Gaussians. I am pleased that we took a step back to look at alternatives that are more efficient and that we found an elegant solution that does not rely on local Gaussian representations.”

Popov believes their research could open the door to new applications that were previously thought impossible."We might be able to spend spacecraft into orbits where they are even more seldom detected than before, while still maintaining an excellent grasp as to our uncertainty about their location."

 This breakthrough could enhance not just space exploration, but also improve the precision of geoscience applications like climate modeling, where accurate data assimilation is critical.

With this award-winning paper, Zanetti and Popov have pushed the boundaries of what is possible in state estimation and particle filtering. Their research serves as hope for further exploration in non-Gaussian filtering techniques, shaping the future of many high-dimensional systems, from the skies above us to the weather systems that surround us. 

The Jean-Pierre Le Cadre Best Paper Award celebrates groundbreaking contributions to the advancement of technologies that are essential in systems where senior data must be integrated for tasks like navigation, tracking, and control. Zanetti and Popov’s paper took first place among three winners honored by ISIF.