Dans cette thèse, nous étudions les approches calculatoires dédiées des réseaux de neurones profonds et plus particulièrement des réseaux de neurones convolutionnels (CNN). En effet, l'efficacité des réseaux de neurones convolutionnels en font des structures calculatoires intéressantes dans de nombreuses applications. Nous étudions les différentes possibilités d'implémentation de ce type de réseaux pour en déduire leur complexité calculatoire. Nous montrons que la complexité calculatoire de ce type de structure peut rapidement devenir incompatible avec les ressources de l'embarqué. Pour résoudre cette problématique, nous avons fait une exploration des différents modèles de neurones et architectures susceptibles de minimiser les ressources nécessaires à l'application. Dans un premier temps, notre approche a consisté à explorer les possibles gains par changement de modèle de neurones. Nous montrons que les modèles dits impulsionnels permettent en théorie de réduire la complexité calculatoire tout en offrant des propriétés dynamiques intéressantes, mais nécessitent de repenser entièrement l'architecture matériel de calcul. Nous avons alors proposé notre approche impulsionnelle du calcul des réseaux de neurones convolutionnels avec une architecture associée. Nous avons mis en place une chaîne logicielle et de simulation matérielle dans le but d'explorer les différents paradigmes de calcul et implémentation matérielle et évaluer leur adéquation avec les environnements embarqués. Cette chaîne nous permet de valider les aspects calculatoires mais aussi d'évaluer la pertinence de nos choix architecturaux. Notre approche théorique a été validée par notre chaîne et notre architecture a fait l'objet d'une simulation en FDSOI 28 nm. Ainsi nous avons montré que cette approche est relativement efficace avec des propriétés intéressantes un terme de passage à l'échelle, de précision dynamique et de performance calculatoire. Au final, l'implémentation des réseaux de neurones convolutionnels en utilisant des modèles impulsionnels semble être prometteuse pour améliorer l'efficacité des réseaux. De plus, cela permet d'envisager des améliorations par l'ajout d'un apprentissage non supervisé type STDP, l'amélioration du codage impulsionnel ou encore l'intégration efficace de mémoire de type RRAM. / In this thesis, we study the dedicated computational approaches of deep neural networks and more particularly the convolutional neural networks (CNN).We highlight the convolutional neural networks efficiency make them interesting choice for many applications. We study the different implementation possibilities of this type of networks in order to deduce their computational complexity. We show that the computational complexity of this type of structure can quickly become incompatible with embedded resources. To address this issue, we explored differents models of neurons and architectures that could minimize the resources required for the application. In a first step, our approach consisted in exploring the possible gains by changing the model of neurons. We show that the so-called spiking models theoretically reduce the computational complexity while offering interesting dynamic properties but require a complete rethinking of the hardware architecture. We then proposed our spiking approach to the computation of convolutional neural networks with an associated architecture. We have set up a software and hardware simulation chain in order to explore the different paradigms of computation and hardware implementation and evaluate their suitability with embedded environments. This chain allows us to validate the computational aspects but also to evaluate the relevance of our architectural choices. Our theoretical approach has been validated by our chain and our architecture has been simulated in 28 nm FDSOI. Thus we have shown that this approach is relatively efficient with interesting properties of scaling, dynamic precision and computational performance. In the end, the implementation of convolutional neural networks using spiking models seems to be promising for improving the networks efficiency. Moreover, it allows improvements by the addition of a non-supervised learning type STDP, the improvement of the spike coding or the efficient integration of RRAM memory.

BACKGROUND: Pleiotropy is a biological phenomenon of a single gene exhibiting influence over several different seemingly disparate phenotypes. This phenomenon poses significant challenges to fully understanding the etiologies of many different Mendelian diseases. Two such Mendelian diseases are Idiopathic Hypogonadotropic Hypogonadism (IHH) and Craniosynostosis (CS). IHH results from the failure of differentiation, migration, secretion, or action of the GnRH neurons resulting in absent puberty and infertility. CS is characterized by premature fusion of one or more of the cranial sutures resulting in dysmorphic shape of the skull that can lead to life-threatening raised intercranial pressure requiring surgical intervention. Thus far, 77 genes have been implicated in IHH and 128 genes have been implicated CS, both representing ~50% of the cases in their respective diseases. Recent research has suggested a shared molecular landscape in CS and IHH but the full ensemble of this overlap is not known. OBJECTIVE: This study will attempt to utilize human genetics, bioinformatics, statistics, phenotype data of IHH patients, and the prior literature in order to ascertain the full extent of the shared biology of IHH and CS. METHODS: The gene sets of both IHH and CS were used in gene overlap statistical analysis to investigate shared genetics. Whole exome sequencing data from 1,395 patients from the IHH cohort of the Massachusetts General Hospital were used for gene-variant burden analysis to determine genetic overlap with CS. Detailed physician notes from this cohort were used to determine phenotypic presence of CS in IHH. Conversely, evidence of reproductive phenotypes in genetically characterized CS patients was gathered from the reported CS gene literature. The CS and IHH gene sets were also bioinformatically analyzed using both the Metascape and DAVID bioinformatic platforms for pathway annotation, protein-protein interaction (PPI), and functional interactions to provide evidence for the mechanism of shared biology. RESULTS: Of the 128 CS genes and 77 IHH genes, 4 were determined to be causal for both diseases with a further 3 considered as potentially causal candidates for both diseases. The 4 overlapping causal genes were tested using three different methods and this overlap was determined to be of statistical significance (p<0.05). Furthermore, the phenotypic review revealed that while there was not a significant enrichment for CS phenotypes in the IHH cohort, the literature review yielded 49 of 128 CS genes that were reported with phenotypic evidence of failure of the hypothalamic-pituitary portion of the HPG axis. Gene-variant burden analysis yielded nominal (p<0.05) enrichment in the IHH cohort for 17 CS genes, of which 3 were significant after Bonferroni multiple testing correction (p<0.00039). The CS/IHH gene sets were both enriched in 44 shared pathways according to Metascape and 17 shared pathways according to DAVID. PPI analysis yielded 3 shared communities between the two disorders with enrichment in fibroblast signaling, ossification, and cardiac chamber development. CONCLUSIONS: The shared biology between IHH and CS was significantly greater than what was previously appreciated. Shared pathways of the two gene sets point toward the neural crest origin of subpopulations of the GnRH neuron and cranial suture osteoblast as a possible foundation for this shared biology, as well as the migratory nature of these two cells and the role that many genes in both gene sets play in cellular motility. Several CS genes emerge as candidates for IHH and must be individually evaluated. Functional studies should be used to confirm and further unravel the underlying mechanisms for the biological overlap between these two diseases. This study may provide a model for preemptive in silico work prior to more expensive in vitro or in vivo studies of pleiotropy.

No description available.

The work deals with the generation of ECG signals using generative adversarial networks (GAN). It examines in detail the basics of artificial neural networks and the principles of their operation. It theoretically describes the use and operation and the most common types of failures of generative adversarial networks. In this work, a general procedure of signal preprocessing suitable for GAN training was derived, which was used to compile a database. In this work, a total of 3 different GAN models were designed and implemented. The results of the models were visually displayed and analyzed in detail. Finally, the work comments on the achieved results and suggests further research direction of methods dealing with the generation of ECG signals.

Práca je zameraná na navrhnutie vhodnej metódy analýzy smerovosti neuritov. Využité boli snímky neurónov z fluorescenčnej mikroskopie. Pred samotnou segmentáciou bolo potrebné snímky predspracovať, pričom sa postupne využila úprava kontrastu, ostrenie a adaptívna filtrácia pomocou Weinerovského filtru. Jednotlivé návrhy metód segmentácie pozostávali z prostého prahovania, narastaním oblastí a využitím morfologických operácií. Následná analýza smerovosti využívala smer gradientov v obraze. Navrhnutá metóda bola využitá aj ako klasifikátor, ktorý dokázal rozdeliť jednotlivé snímky do skupín podľa smeru rastu.

The goal was to get acquainted with the problems of network elements to describe neural networks that can be used to manage such a feature. The theoretical part deals with the neural networks from their inception to the present. It focuses mainly on the network, witch can be used for management control. These are the two network: Hopfield network and Kohonen network. The practical part deals with the network element model and ist implementation. It contains a practical element model using a neural network, witch is controlled by a network element.

Goal of this thesis is to get familiar with problematics of neural networks and commonly used security protocols in cryptography. Theoretical part of the thesis is about neural networks theory and chooses best suitable type of neural network to use in cryptographic model. In practical part, a new type of security protocol is created, using chosen neural network.

This work deals with traveling salesman problem (TSP) and examines it’s possibilities to use in business. It is about the optimization of the travel cost, saving time and unnecessary mileage. Part of the work is a program with a GUI written in program MATLAB. Program uses neural networks to calculate the most effective path between places, where the trader has to reach. It’s possible to use the algorithm for many purposes, e.g. distribution of goods, store management, planning of PCBs or rescue services. Program communicates with the Google Maps API server, which provides the actual information of the path.

This master thesis is focused on designing and implementing a software system, that is able to trade autonomously at stock market. Neural networks are used to predict future price. Genetic algorithm was used to find good combination of input parameters.

In 2007, Modolo and colleagues derived a population density equation for a population of Izhekevich neurons. This population density equation can describe oscillations in the brain that occur in Parkinson’s disease. Numerical simulations of the population density equation showed bursting behaviour even though the individual neurons had parameters that put them in the tonic firing regime. The bursting comes from neuron interactions but the mechanism producing this behaviour was not clear. In this thesis we study numerical behaviour of the population density equation and then use a combination of analysis and numerical simulation to analyze the basic qualitative behaviour of the population model by means of a simplifying assumption: that the initial density is a Dirac function and all neurons are identical, including the number of inputs they receive, so they remain as a point mass over time. This leads to a new ODE model for the population. For the new ODE system, we define a Poincaré map and then to describe and analyze it under conditions on model parameters that are met by the typical values adopted by Modolo and colleagues. We show that there is a unique fixed point for this map and that under changes in a bifurcation parameter, the system transitions from fast tonic firing, through an interval where bursting occurs, the number of spikes decreasing as the bifurcation parameter increases, and finally to slow tonic firing. / Graduate