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

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.


GOLD : The Golden Cosmological Surveys Decade

This 10-week programme on the Golden Cosmological Surveys Decade will be held at the new Institut Pascal, in Paris Orsay, from 1st April 2020 to 5th June 2020. The Institut Pascal provides offices, seminar rooms, common areas and supports long-term scientific programmes. 
GOLD 2020 will include a summer school, three workshops (on Lensing, Galaxy Clustering, Theory and Interpretation of the Data). 
In-between, an active training programme will be run. We plan to host around 40 people for the whole programme, plus around 30 scientists during the workshops. 
Whether you are a PhD, a postdoc, a senior scientist and are interested in attending this programme, you can now apply. Deadline for applications: 1st October 2019.




The workshop on Computational Intelligence in Remote Sensing and Astrophysics (CIRSA) aims at bringing together researchers from the environmental sciences, astrophysics and computer science communities in an effort to understand the potential and pitfalls of novel computational intelligence paradigms including machine learning and large-scale data processing.



French-Chinese Days on Weak Lensing

Date: October 4-5, 2018

Organizer:  Jean-Luc Starck and Martin Kilbinger


Local information

CEA Saclay is around 23 km South of Paris. The astrophysics division (DAp) is located at the CEA site at Orme des Merisiers, which is around 1 km South of the main CEA campus. See for detailed information on how to arrive.

On 4 and 5 October, 2018, we are organizing the first French-Chinese weak-lensing meeting at DAp, CEA Saclay. 


All talks are taking place at DAp, Salle Kepler (Building 709)

Thursday, October 4

9:30 - 10:00h.  Café

10:00 - 10:15h.    Welcome & introductions

10:15 - 10:45h.   Hu Zhan,   Overview of CSS-OS

10:45 - 11:15h.   Martin Kilbinger, Overview of CFIS Weak Lensing

11:15 - 11:45h.  Jun Zhang, Fourier_Quad,  a shear measurement method in Fourier Space

11:45 - 14:00h.  Lunch at the Rotonde

14:00 - 14:30h. Morgan Schmitz,  PSF Modeling using a Graph Manifold

14:30 - 15:00h. Chengliang Wei, A full sky WL simulation with semi-analytic galaxy formation 

15:00 - 15:30h. Jean-Luc Starck,  WL Mass Mapping

15:30 - 16:00h. Zuhui Fan,  WL peak statistics

16:00 - 16:30h. Austin Peel,  Cosmology with Mass Maps

Friday, October 5

9:30 - 10:00h.      Café

10:00 - 10:30h.   Sam Farrens,   The CFIS pipeline

10:30 - 11:00h.  Ran Li,  Lensing studies of sub-structures

11:00 - 11:30h.  Axel  Guinot,  Preliminary CFIS results

11:30 - 12:00h.  Liping Fu, Shear measurement from VOICE deep survey

12:00 - 14:00h. Lunch at Les Algorithmes

14:00 - 14:30h. Jean-Charles Cuillandre, The Euclid mission and ground-based observations

14:30 - 15:00h.  Huanyuan Shan: KiDS WL studies (via skype)

15:00 - 15:30h.  Alexandre Bruckert, Machine learning for blended objects separation

15:30 - 16:00h.   Rebeca Araripe Furtado Cunha,  Optimal Transport and PSF Modeling

16:00 -  17:00h. Discussion