Dear all,
The computational mathematics and statistics spring seminar will continue Wednesday, February 25th, at 14:00 – 16:00 (utc+2) on ZOOM with two presentations:
* Hamza Djelouat on “Exploiting Sparsity and Structure in Inverse Problems: ''Bayesian Methods for Massive Wireless Connectivity” * Reijo Leinonen on “Hierarchical Bayesian Joint Activity and Channel Estimation under Spatially Correlated Priors”
If you have questions about the seminar or would like to give a talk, please send an e-mail to markku.kuismin(at)oulu.fi
The recurring ZOOM link is: https://oulu.zoom.us/j/65514799333?pwd=fIviyQRrSNNGtWCkStvS8IxM8mdHhu.1 passcode: 948147
Link to the Continuously Updated Program: [https://res.public.onecdn.static.microsoft/assets/fluentui-resources/1.1.0/a... computational mathematics and data science seminar.docxhttps://unioulu-my.sharepoint.com/:w:/g/personal/kuismima_univ_yo_oulu_fi/EQq_4cxLAs1HvtERoIz0tz0Bw6xaulBfhgBXI6Z9kIRX3A?e=SB6kgI
Wednesday 25.02.2026 14:00 – 15:00 Hamza Djelouat (Communications Engineering)
Title: Exploiting Sparsity and Structure in Inverse Problems: ''Bayesian Methods for Massive Wireless Connectivity''
Abstract: Many modern data acquisition systems give rise to large-scale inverse problems in which the unknown quantity is high-dimensional, structured, and only partially observed through noisy linear measurements. A central challenge in such problems is to exploit a priori information—such as sparsity, correlation, and group structure—while preserving computational tractability and robustness to model mismatch.
This talk presents recent work on structured sparse recovery problems in which the unknown parameters exhibit row-wise sparsity, correlated and hierarchical support patterns, and unknown spatial correlation within their nonzero components. Such models naturally lead to a hierarchy of coupled inverse problems involving sparse signal recovery, covariance estimation, and latent variable inference.
We develop a Bayesian framework that unifies several approaches to structured inverse problems, ranging from point estimation via maximum a posteriori (MAP) formulations to full Bayesian inference methods, including group sparse Bayesian learning and structured spike-and-slab models. These models explicitly encode hierarchical sparsity and correlated structure. We also discuss total-variation–inspired priors as a mechanism for promoting smoothness and coherence in the recovered support. While the presentation is motivated by a wireless communication applications, the proposed methodology and the underlying insights are broadly applicable to inverse problems in signal processing, imaging, and high-dimensional statistics.
Selected References:
* H. Djelouat, R. Leinonen, M. J. Sillanpää, B. D. Rao and M. Juntti, "Adaptive and Self-Tuning SBL With Total Variation Priors for Block-Sparse Signal Recovery," in IEEE Signal Processing Letters, vol. 32, pp. 1555-1559, 2025 * H. Djelouat, M. J. Sillanpää, M. Leinonen and M. Juntti, "Hierarchical MTC User Activity Detection and Channel Estimation With Unknown Spatial Covariance," in IEEE Transactions on Wireless Communications, vol. 23, no. 10, pp. 14620-14636, Oct. 2024 * H. Djelouat, M. Leinonen and M. Juntti, "Spatial Correlation Aware Compressed Sensing for User Activity Detection and Channel Estimation in Massive MTC," in IEEE Transactions on Wireless Communications, vol. 21, no. 8, pp. 6402-6416, Aug. 2022
Wednesday 25.02.2026 15:00 – 16:00 Reijo Leinonen (Research Unit of Mathematical Sciences)
Title: Hierarchical Bayesian Joint Activity and Channel Estimation under Spatially Correlated Priors
Abstract: The presentation addresses the Joint Activity Detection and Channel Estimation (JADCE) problem within the framework of high-dimensional sparse recovery. We consider a system where device activity is non-independent, exhibiting spatial correlations typical of event-triggered processes. To model this structured sparsity, we propose a hierarchical Bayesian framework that couples a first-order Markov chain with a spike-and-slab prior. This formulation enables the simultaneous inference of discrete activation states and continuous channel coefficients.
We derive a posterior inference strategy using Markov Chain Monte Carlo (MCMC) sampling. The discussion will detail the construction of the hierarchical model, the convergence properties of the sampler, and performance gains in MSE and F1-scores. Finally, we provide a comparative analysis against Sparse Bayesian Learning (SBL) benchmarks to demonstrate the advantages of incorporating structured priors in correlated environments. For more details on the underlying model and derivations, you can refer to our conference paper: "Grouped Device Detection and Channel Estimation in MTC Using a Full Bayesian Approach".
Best,
Markku Kuismin
Tutkijatohtori| Postdoctoral researcher
Oulun yliopisto | University of Oulu
Matemaattisten tieteiden tutkimusyksikkö| Research unit of mathematical sciences
PO Box 8000, FI-90014 OULUN YLIOPISTO
www.oulu.fihttps://www.oulu.fi
[cid:6f9b6ac5-808a-45bb-ab17-a88e5c0f5e65]
Follow us on Facebookhttps://www.facebook.com/unioulu/, Instagramhttps://www.instagram.com/unioulu/?hl=fi, Linkedinhttps://www.linkedin.com/school/university-of-oulu/, Bluesky,https://bsky.app/profile/oulu.fi and YouTubehttps://www.youtube.com/user/UniversityOfOulu.
