However, this representation fails to identify collective patterns in neural activations. We instead argue that the focus should We introduce the Singular Value Representation be shifted to the successive linear maps which define the (SVR), a new method to represent the internal network. We propose to use Singular Value Decomposition state of neural networks using SVD factorization (SVD) of those maps to identify meaningful input and of the weights. This construction yields a new output directions corresponding to collective activation patterns weighted graph connecting what we call spectral of neurons. We further investigate the interaction of neurons, that correspond to specific activation those directions across deep layers, which yields a new patterns of classical neurons. We derive a precise graph representation of neural networks. This graph provides statistical framework to discriminate meaningful a high-level overview of the network that allows to connections between spectral neurons for fully witness the emergence of global phenomenons across deep connected and convolutional layers. To demonstrate layers as highlighted in the last section.

Deep learning models are state-of-the-art in compressive spectral imaging (CSI) recovery. These methods use a deep neural network (DNN) as an image generator to learn non-linear mapping from compressed measurements to the spectral image. For instance, the deep spectral prior approach uses a convolutional autoencoder network (CAE) in the optimization algorithm to recover the spectral image by using a non-linear representation. However, the CAE training is detached from the recovery problem, which does not guarantee optimal representation of the spectral images for the CSI problem. This work proposes a joint non-linear representation and recovery network (JR2net), linking the representation and recovery task into a single optimization problem. JR2net consists of an optimization-inspired network following an ADMM formulation that learns a non-linear low-dimensional representation and simultaneously performs the spectral image recovery, trained via the end-to-end approach. Experimental results show the superiority of the proposed method with improvements up to 2.57 dB in PSNR and performance around 2000 times faster than state-of-the-art methods.

Spectral histopathology is a diagnostic tool based on the numerical analysis of vibrational spectral images (Raman or infrared) for which the contribution of deep learning has been superficially studied. The purpose of this thesis is to explore the potential applications of deep learning in spectral histopathology for the characterization and quantification by infrared spectral imaging of the different types of fibrosis and inflammation on renal graft biopsies. The first objective will be to study the independence of deep learning from preprocessing of infrared spectra, as suggested in literature. In a second step, the impact of the spatial definition of the acquired spectral images on the performance of convolutional neural networks will be studied. The third objective will be to compare convolutional neural networks to traditional supervised classification methods such as large support vector machines (SVM) and random forests (RF). The fourth objective of this work will be to take advantage of the capacities of autoencoders to learn transfer functions of different infrared imagers to make the proposed methodology transferable in clinical routine. The last objective of this work will be to study whether deep learning is able to identify and quantify the subtypes of fibrosis and inflammation in order to refine the diagnosis and offer a better prognosis for renal grafts as well as to guide the clinician in the choice of a therapeutic treatment adapted to each renal graft recipient.

Schizophrenia is a complex psychiatric disorder involving changes in thought patterns, perception, mood, and behavior. The diagnosis of schizophrenia is challenging and requires that patients show two or more positive symptoms for at least one month. Delays in identifying this debilitating disorder can impede a patient ability to receive much needed treatment. Advances in neuroimaging and machine learning algorithms can facilitate the diagnosis of schizophrenia and help clinicians to provide an accurate diagnosis of the disease. This paper presents a methodology for analyzing spectral images of Electroencephalography collected from patients with schizophrenia using convolutional neural networks. It also explains how we have developed accurate classifiers employing Model-Agnostic Meta-Learning and prototypical networks. Such classifiers have the capacity to distinguish people with schizophrenia from healthy controls based on their brain activity.

Neural networks have helped scientists monitor the Sun's extreme ultraviolet outbursts after an instrument on NASA's Solar Dynamic Observatory suffered an electrical malfunction, making it difficult for scientists to monitor a portion of extreme ultraviolet energy (EUV) being spewed by our star. EUV rays ejected from solar flares are particularly worrisome. The surge of highly energetic particles bombarding Earth can cause radio communication blackouts, knock satellites out of place, and disturb GPS signals. Space agencies around the world keep a close eye on the Sun's activity in an attempt to study and predict these outbursts. NASA's SDO is just one of the many spacecrafts currently orbiting our planet's star.

The increasing access to brain signal data using electroencephalography creates new opportunities to study electrophysiological brain activity and perform ambulatory diagnoses of neuronal diseases. This work proposes a pairwise distance learning approach for Schizophrenia classification relying on the spectral properties of the signal. Given the limited number of observations (i.e. the case and/or control individuals) in clinical trials, we propose a Siamese neural network architecture to learn a discriminative feature space from pairwise combinations of observations per channel. In this way, the multivariate order of the signal is used as a form of data augmentation, further supporting the network generalization ability. Convolutional layers with parameters learned under a cosine contrastive loss are proposed to adequately explore spectral images derived from the brain signal. Results on a case-control population show that the features extracted using the proposed neural network lead to an improved Schizophrenia diagnosis (+10pp in accuracy and sensitivity) against spectral features, thus suggesting the existence of non-trivial, discriminative electrophysiological brain patterns.

Spectral images captured by satellites and radio-telescopes are analyzed to obtain information about geological compositions distributions, distant asters as well as undersea terrain. Spectral images usually contain tens to hundreds of continuous narrow spectral bands and are widely used in various fields. But the vast majority of those image signals are beyond the visible range, which calls for special visualization technique. The visualizations of spectral images shall convey as much information as possible from the original signal and facilitate image interpretation. However, most of the existing visualizatio methods display spectral images in false colors, which contradict with human's experience and expectation. In this paper, we present a novel visualization generative adversarial network (GAN) to display spectral images in natural colors. To achieve our goal, we propose a loss function which consists of an adversarial loss and a structure loss. The adversarial loss pushes our solution to the natural image distribution using a discriminator network that is trained to differentiate between false-color images and natural-color images. We also use a cycle loss as the structure constraint to guarantee structure consistency. Experimental results show that our method is able to generate structure-preserved and natural-looking visualizations.