Proximal methods for point source localisation

Tuomo Valkonen
DOI: https://doi.org/10.46298/jnsao-2023-10433
2023-09-20
Abstract:Point source localisation is generally modelled as a Lasso-type problem on measures. However, optimisation methods in non-Hilbert spaces, such as the space of Radon measures, are much less developed than in Hilbert spaces. Most numerical algorithms for point source localisation are based on the Frank-Wolfe conditional gradient method, for which ad hoc convergence theory is developed. We develop extensions of proximal-type methods to spaces of measures. This includes forward-backward splitting, its inertial version, and primal-dual proximal splitting. Their convergence proofs follow standard patterns. We demonstrate their numerical efficacy.
Optimization and Control,Computer Vision and Pattern Recognition
What problem does this paper attempt to address?
### Problems the paper attempts to solve This paper attempts to solve the point - source localization problem. Specifically, the point - source localization problem is usually modeled as a Lasso - type problem in the measure space, in the following form: \[ \min_{0 \leq \mu \in \mathcal{M}(\Omega)} \frac{1}{2} \| A\mu - b \|^2 + \alpha \| \mu \|_{\mathcal{M}(\Omega)} \] where: - \(\alpha > 0\) is the regularization parameter, - \(b \in \mathbb{R}^n\) is the measurement data, - \(A \in \mathcal{L}(\mathcal{M}(\Omega); \mathbb{R}^n)\) is the forward operator, defined on the Radon measure space \(\mathcal{M}(\Omega)\) on \(\Omega \subset \mathbb{R}^m\). ### Main research contents 1. **Limitations of existing methods**: - Most of the current numerical algorithms for point - source localization are based on the Frank - Wolfe conditional gradient method, and this method has developed a convergence theory for specific problems. - Forward - backward splitting and other proximal methods are less applied in non - Hilbert spaces, especially in the Radon measure space. 2. **Proposed new methods**: - The author has developed extensions of proximal methods in the measure space, including forward - backward splitting, its inertial version, and the primal - dual proximal splitting method. - The convergence proofs of these methods follow a standard pattern, and their effectiveness has been verified through numerical experiments. 3. **Key techniques and innovation points**: - A semi - inner product \(\langle \mu, \nu \rangle_{\mathcal{D}} := \langle \mathcal{D}\mu | \nu \rangle\) is constructed, where \(\mathcal{D} \in \mathcal{L}(\mathcal{M}(\Omega); C_0(\Omega))\) is a "particle - to - wave" operator. - Based on this semi - inner product, a new proximal penalty term is defined, making the forward - backward splitting method practical in non - Hilbert spaces. - \(\mathcal{D}\) is constructed through the convolution operator, and its relationship with weak* convergence is studied. 4. **Theoretical analysis**: - The convergence of these new methods is proved, and the inexact computation of each step method is considered. - Through the method of semi - infinite programming, multiple points can be inserted to support the update of \(\mu\), which is different from the traditional conditional gradient method. 5. **Numerical experiments**: - The effectiveness of the proposed methods is verified through numerical experiments, showing their potential in practical applications. ### Conclusion This paper solves the optimization problem of point - source localization in non - Hilbert spaces by developing new proximal methods, providing a more effective solution. These methods are not only strictly proved theoretically, but also show good performance in numerical experiments.