2025 IEEE Statistical Signal Processing Workshop

These tutorials will run in parallel on Sunday 8th June at Room G05 and G06, 50 George Square, Edinburgh.

More information on the tutorials can be found below:

1. Inverse problems: toward high performance, speed and uncertainty quantification. Prof. Pierre Chainais, University of Lille and Dr. Pierre-Antoine Thouvenin, University of Lille.

Abstract: Bayesian methods for inverse problems in signal and image processing have the advantage of giving access to the a posteriori distribution of the parameters of interest. Thus, one not only accesses a solution to the problem, but also valuable credibility intervals. For example, in astrophysics or medicine, there is generally no ground truth. Providing predictions with confidence intervals is essential: reading the reconstructed image is done with a controlled level of confidence. Nevertheless, Monte Carlo simulations of a posteriori distributions are reputed to be computationally intensive, with limited scalability for high-dimensional inverse problems based on large datasets. We will present a family of approaches called “Asymptotically Exact Data Augmentation” (AXDA). This approach, inspired by splitting in optimization, makes it possible to systematically construct an approximate distribution that is less expensive to sample than the target distribution of the initial model thanks to the Split-Gibbs-sampler (SGS). This framework offers a trade-off between numerical efficiency and quality of the approximation. Moreover these methods pave the way to many variations, in particular Plug-and-Play (PnP) Langevin based sampling such as the the PnP-Split-Gibbs-sampler (PnP-SGS) using diffusion models. We will discuss and illustrate by applications to the resolution of inverse problems. Finally, addressing very high dimensional problems will lead to the question of distributed MCMC based inference.

Biography: Pierre Chainais received the Ph.D. degree in physics from the Ecole Normale Superieure de Lyon, France, in 2001. He joined the University Blaise Pascal of Clermont-Ferrand as an Associate Professor in signal processing in 2002. He moved to Centrale Lille Institute in 2011, where he currently is a Professor in signal processing and machine learning with the CRIStAL laboratory of informatics in Lille (France). His research interests include statistical signal processing and machine learning with applications to image processing and physical sciences (astrophysics, cosmology, gravitational waves). Bayesian approaches and stochastic sampling methods to solve high dimensional inverse problems as well as the specific study of bivariate signals are among his present hot topics.

Biography: Pierre-Antoine Thouvenin received the Ph.D. degree in signal processing from the National Polytechnic Institute of Toulouse, France, in 2017. From 2017 to 2019, he was a post-doctoral research associate with the Institute of Sensors, Signals an Sytems (ISSS), Heriot-Watt University, Edinburgh, UK. Since sept. 2019, he is assistant professor at Centrale Lille Institute, France. His research interests include statistical signal processing and distributed computing, with a particular interest in high dimensional inverse problems and applications to astronomy and remote sensing.

2. Unlimited Sensing: Signal Processing with Quantization Noise. Dr Ayush Bhandari, Imperial College London.

Abstract. The ability to digitize information has profoundly transformed human lives. At the heart of digital acquisition lies quantization—the rounding-off of time and amplitudes. These foundational concepts are taught in undergraduate engineering curricula worldwide. While the Shannon-Nyquist sampling theorem proves that time-quantization (or sampling) is lossless, amplitude-quantization introduces quantization noise, making the analog-to-digital conversion lossy. 

But can we recover a signal from quantization noise alone? Remarkably, the answer is “Yes!” Even more surprising, this recovery follows the well-known Nyquist criterion. This approach, which redefines the quantization noise (QN) as the digital signal itself, is at the core of the Unlimited Sensing Framework (USF). Rooted in an intriguing mathematical principle, the USF exploits the fact that, for smooth functions, their fractional parts (the QN) encode their integer parts (the digital signal). By leveraging this insight, the USF simultaneously achieves high-dynamic-range and high-digital-resolution within a given bit-budget; this is not possible with conventional methods, including compressed sensing. This tutorial has two primary goals. First, since the USF spans theory, algorithms, hardware, and experiments, we aim to provide a holistic introduction to this emerging topic from first principles. Second, we aim to introduce the new class of inverse problems posed by the USF, exploring signal processing challenges—such as sparse recovery, sub-Nyquist sampling, radar/comms and computational imaging—through the lens of modulo non-linearities. Case studies will feature novel hardware and lab experiments, highlighting the performance breakthroughs enabled by hardware-software co-design. 

Biography: Ayush Bhandari received the Ph.D. degree from Massachusetts Institute of Technology (MIT) in 2018, for his work on computational sensing and imaging which, in part, led to the co-authored open-access book Computational Imaging in MIT Press (2022). He is currently a faculty member with the Engineering Department at Imperial College London. He was appointed the August–Wilhelm Scheer Visiting Professor (Dept. of Mathematics), in 2019 by the Technical University of Munich. He holds 10 US Patents. His work has received several distinctions including the 2020 UKRI Future Leaders Fellowship, the 2020 IEEE SPS Best PhD Dissertation Award, the 2021 President’s Medal for Outstanding Early Career Researcher, the 2023 Frontiers of Science Award, and the 2024 ERC Starting Grant.

3. Introduction to Multi-target Tracking with Stone Soup. Dr Steven Hiscocks, Dstl and Dr James Wright, Dstl

Abstract: This tutorial focuses on multi-target state estimation and tracking using the Stone Soup framework through hands-on practical sessions with simulated data. Participants will learn to apply multi-sensor, multi-object estimation algorithms, covering state estimation techniques such as Gaussian (Kalman) and Monte Carlo sampling (particle) filtering, with linear and non-linear models, data association, and track management. The course introduces single and multi-target tracking and practical tracking challenges, starting with basic examples to help attendees familiarise themselves with target tracking concepts and Stone Soup’s capabilities.

Biography: Steve Hiscocks is a principal data fusion scientist and lead developer of Stone Soup. His focus is on the design and architecture aspects of the framework, and application of data fusion with real-world applications.

Biography: James Wright is a senior data fusion scientist focussing on developing the algorithmic offerings within the Stone soup framework and applying the framework through simulation modelling and analysis of real-world data in the maritime and air domains.

4. Space-Time Covariance Matrix Factorisation and Estimation for Broadband Multichannel Problems. Prof. Stephan Weiss, University of Strathclyde and Prof. Ian K. Proudler, University of Strathclyde.

Abstract: This tutorial addresses recent developments in formulating and solving broadband multichannel problems through matrices of functions and their factorisations, such as the analytic eigenvalue decomposition of the space-time covariance, or an analytic singular value decomposition applied to a data model. This can generalise well known formulations of narrowband problems using covariance matrices, and of narrowband solutions via their diagonalisation, to the broadband case. We present theoretical background on the factorisation of matrices of functions, and show how the estimation of the statistical parameters impacts on the perturbation of the ground truth factors of a decomposition. We review a number of algorithms, and discuss some sample applications such as direction of arrival estimation, beamforming, weak transient signal and subspace detection, MIMO communications, speech enhancement, or source separation.

Biography: Stephan Weiss is a professor of signal processing at the University of Strathclyde, Glasgow, Scotland. He specialises in multirate, array, and adaptive signal processing, and is a Senior Member of IEEE and an elected member of both the IEEE Signal Processing Society’s technical committee on Signal Processing Theory and Methods and the EURASIP Technical Area of Signal Processing for Multisensor Systems. For Elsevier Signal Processing he serves as a subject editor. He has presented tutorials on the topic of polynomial matrix algebra at IEEE SAM’16 and EUSIPCO’23. 

Biography: Ian K. Proudler received the graduation degree in physics from Oxford University, and a PhD in signal processing from Cambridge University. He is currently a Professor of signal processing at the University of Strathclyde. From 1986 until 2011, he was in the defence sector looking into various adaptive digital signal processing issues such as: numerical stability and efficient computation; antenna algorithm for HF communications; signal separation for ESM purposes; magnetic detection for maritime surveillance; and GPS antijam systems. He has published some 60 research papers, contributed to three textbooks and holds a patent on an adaptive filtering architecture. Dr. Proudler received the John Benjamin Memorial Prize, in 1992 and 2001, and the IEE J. J. Thomson Medal, in 2002, for his work on signal processing algorithms. He was an Honorary Editor of IEE Proceedings: Radar, Sonar and Navigation for ten years. He has been on the organizing committee of several international conferences.

5. Estimation of Graph Signals. Prof. Tirza Routtenberg, Ben-Gurion University of the Negev.

Abstract:

Biography: