Benchmarking MRI Reconstruction Neural Networks on Large Public Datasets

Deep learning is starting to offer promising results for reconstruction in Magnetic Resonance Imaging (MRI). A lot of networks are being developed, but the comparisons remain hard because the frameworks used are not the same among studies, the networks are not properly re-trained, and the datasets used are not the same among comparisons. The recent release of a public dataset, fastMRI, consisting of raw k-space data, encouraged us to write a consistent benchmark of several deep neural networks for MR image reconstruction. This paper shows the results obtained for this benchmark, allowing to compare the networks, and links the open source implementation of all these networks in Keras. The main finding of this benchmark is that it is beneficial to perform more iterations between the image and the measurement spaces compared to having a deeper per-space network.

Reference:  Z. Ramzi, P. Ciuciu and J.-L. Starck. “Benchmarking MRI reconstruction neural networks on large public datasetsApplied Sciences, 10, 1816, 2020.  doi:10.3390/app10051816

CosmosClub: Florent Mertens (10/03/20)

Date: March 18th 2020, 10.30am

Speaker: Florent Mertens (LERMA / Kapteyn Astronomical Institute)

Title: The challenges of observing the Epoch of Reionization and Cosmic Dawn

Room: Cassini 


Low-frequency observations of the redshifted 21cm line promise to open a new window onto the first billion years of cosmic history, allowing us to directly study the astrophysical processes occurring during the Epoch of Reionization (EoR) and the Cosmic Dawn (CD). This exciting goal is challenged by the difficulty of extracting the feeble 21-cm signal buried under astrophysical foregrounds orders of magnitude brighter and contaminated by numerous instrumental systematics. Several experiments such as LOFAR, MWA, HERA, and NenuFAR are currently underway aiming at statistically detecting the 21-cm brightness temperature fluctuations from the EoR and CD. While no detection is yet in sight, considerable progress has been made recently. In this talk, I will review the many challenges faced by these difficult experiments and I will share the latest development of the LOFAR Epoch of Reionization and NenuFAR Cosmic Dawn key science projects.


The first Deep Learning reconstruction of dark matter maps from weak lensing observational data

DeepMass: The first Deep Learning reconstruction of dark matter maps from weak lensing observational data (DES SV weak lensing data)


 This is the first reconstruction of dark matter maps from weak lensing observational data using deep learning. We train a convolution neural network (CNN) with a Unet based architecture on over 3.6 x 10^5 simulated data realisations with non-Gaussian shape noise and with cosmological parameters varying over a broad prior distribution.  Our DeepMass method is substantially more accurate than existing mass-mapping methods. With a validation set of 8000 simulated DES SV data realisations, compared to Wiener filtering with a fixed power spectrum, the DeepMass method improved the mean-square-error (MSE) by 11 per cent. With N-body simulated MICE mock data, we show that Wiener filtering with the optimal known power spectrum still gives a worse MSE than our generalised method with no input cosmological parameters; we show that the improvement is driven by the non-linear structures in the convergence. With higher galaxy density in future weak lensing data unveiling more non-linear scales, it is likely that deep learning will be a leading approach for mass mapping with Euclid and LSST.

Reference 1:  N. Jeffrey, F.  Lanusse, O. Lahav, J.-L. Starck,  "Learning dark matter map reconstructions from DES SV weak lensing data", Monthly Notices of the Royal Astronomical Society, in press, 2019.


CosmosClub: Irène Waldspurger (04/12/19)

Date: December 4rd 2019, 10.30am

Speaker: Irène Waldspurger (CEREMADE,  Université Paris-Dauphine)

Title: Convex and non-convex algorithms for phase retrieval

Room: Cassini 


Phase retrieval problems consist in recovering elements of a complex vector space from the modulus of their scalar product with a fixed family of measurement vectors. Traditional reconstruction algorithms rely on simple local optimization heuristics. Although they can in principle, because of the non-convexity of the problem, get stuck in local optima, they are observed to work well in many situations.

In this talk, we will see which theoretical correctness guarantees one can establish, in a particular setting, for the most well-known such algorithm. We will also present a different family of algorithms, based on so-called convexification techniques, describe its advantages and limitations.



Radio Astronomical Images Restoration with Shape Constraint


Journal: Proceedings of SPIE
Year: 2019
Download: DOI




Weak gravitational lensing is a very promising probe for cosmology that relies on highly precise shape measurements. Several new instruments are being deployed and will allow for weak lensing studies on unprecedented scales, and at new frequencies. In particular, some of these new instruments should allow for the blooming of radio-weak lensing, specially the SKA with many Petabits per second of raw data. Hence, great challenges will be waiting at the turn. In addition, processing methods should be able to extract the highest precision possible and ideally, be applicable to radio-astronomy. For the moment, the two methods that already exist do not satisfy both conditions. In this paper, we present a new plug-and-play solution where we add a shape constraint to deconvolution algorithms and results show measurements improvement of at least 20%.