Applications of Optimization II

WD-04: Applications of Optimization II
Stream: Applications of Optimization
Room: Lagrange
Chair(s): Sebastien Bourguignon

Continuous optimization of Fourier sampling schemes for MRI
Alban Gossard, Frédéric de Gournay, Pierre Weiss
Fourier sampling is a critical issue in image processing. While Shannon and compressed sensing theories dominated the field for decades, a recent trend is to optimize the sampling scheme for a specific dataset. In this work, we’ll focus on methods that continuously optimize the sampling points with respect to a reconstruction algorithm and to a database. Most attempts in this direction report optimization issues. We show that this phenomenon cannot be avoided since the problem is intrinsically combinatorial and has a huge number of spurious minimizers. We propose ideas to leverage this issue.

GPU-Accelerated Plan Optimization for Intensity-Modulated Radiotherapy
Juan José Moreno Riado, Janusz Miroforidis, Dmitry Podkopaev, Ernestas Filatovas, Ignacy Kaliszewski, Ester M Garzon
Intensity-Modulated Radiotherapy (IMRT) is a technique for cancer treatment that allows precise control over the geometry and intensity profile of radiation beams. Since the dose deposited in the patient’s body by a given IMRT plan is modeled by a sparse matrix, most of the computation time during the plan optimization stage is spent in sparse matrix multiplications. Therefore, an adequate use of HPC techniques can drastically reduce planning time. In this work, we describe a GPU-accelerated gradient-based optimization technique capable of generating adequate plans in a limited time span.

Optimization of molecular descriptors using memetic algorithms
Savíns Puertas Martín, Juana Lopez Redondo, Horacio Pérez-Sánchez, Pilar M. Ortigosa
One of the aims of Drug Discovery is to find compounds similar to a reference. For this purpose, different methods are appearing, among which we highlight Virtual Screening (VS). Usually, the VS techniques are ad-hoc methods that optimize a particular descriptor used as a scoring function. In this work, we propose a generic memetic optimization algorithm able to deal with any descriptor. Results show that our proposal outperforms the solutions provided by the state-of-the-art algorithms, previously proposed in the literature.

Theme: Overlay by Kaira