Search

Supplemental material, sj-pdf-1-inv-10.1177_15569845231164134 for Assessment of Large Animal Vascular Dimensions for Intra-Aortic Device Research and Development: A Systematic Review by Gabriel Georges, Thomas Couture and Pierre Voisine in Innovations: Technology and Techniques in Cardiothoracic and Vascular Surgery

Additional file 11.

Over the past decade, computational power has become more accessible, as hardware previously only found in research or military computers has become mainstream. These improvements have led to a vastly increased speed in repetitive calculations, allowing for the widespread adoption of machine learning and, especially, of deep neural networks. Whereas previously large neural networks were impossible to use due to limited memory and computer power, they have now become commonplace in many areas. But, with the increase in popularity of deep neural networks came a need to get a better grasp on their internal behavior. As such, research has deepened the understanding of layers that comprise these neural networks. Researchers now not only rely on fully connected layers but also locally connected layers or temporally connected layers. Further, they solved the mystery of deeper models becoming less performant, with residual layers. However, even though our understanding of neural network layers has considerably improved, our understanding of the components of these layers, the nodes, is still lacking. In this thesis, we focus on understanding how individual nodes contribute to a neural network's performance. To this end, we try to classify them using a metric that connects the individual nodes to the loss of the model. Furthermore, we also analyze how removing nodes can affect the neural network. Our underlying assumption is that not all nodes contribute to the neural network's performance. We believe that some nodes are redundant and that these nodes both do not actively contribute to the performance of the model and increase its needed resources. By removing them, we reduce wasted resources and speed up the processing time of the model at the training and inference stage. Our results show that not all nodes contribute to a model's performance. More specifically, some nodes are redundant as they extract the same features as others. Others negatively impact the performance as a whole, while some initially contribute positively to the network, but actually, only correct errors initiated by other nodes. Another important result is the increased stability between models from which we remove nodes. In other words, models of the same architecture trained on the same dataset, but with different initial weights, have a noticeably higher variation in performance before and after node pruning. This can be seen with the pruned convolutional neural network having a 5% reduction in its loss on average and a 19% lower loss variation. While this study makes no claim to provide a full understanding of how nodes impact neural networks, it does suggest promising paths for future works that could enhance our understanding of neural networks.

High Energy Density Physics (HEDP) flows are multi-material flows characterizedby strong shock waves and large changes in the domain shape due to rarefactionwaves. Numerical schemes based on the Lagrangian formalism are good candidatesto model this kind of flows since the computational grid follows the fluid motion.This provides accurate results around the shocks as well as a natural tracking ofmulti-material interfaces and free-surfaces. In particular, cell-centered Finite VolumeLagrangian schemes such as GLACE (Godunov-type LAgrangian scheme Conservativefor total Energy) and EUCCLHYD (Explicit Unstructured Cell-CenteredLagrangian HYDrodynamics) provide good results on both the modeling of gas dynamicsand elastic-plastic equations. The work produced during this PhD thesisis in continuity with the work of Maire and Nkonga [JCP, 2009] for the hydrodynamicpart and the work of Kluth and Després [JCP, 2010] for the hyperelasticitypart. More precisely, the aim of this thesis is to develop robust and accurate methodsfor the 3D extension of the EUCCLHYD scheme with a second-order extensionbased on MUSCL (Monotonic Upstream-centered Scheme for Conservation Laws)and GRP (Generalized Riemann Problem) procedures. A particular care is taken onthe preservation of symmetries and the monotonicity of the solutions. The schemerobustness and accuracy are assessed on numerous Lagrangian test cases for whichthe 3D extensions are very challenging.; La Physique des Hautes Densités d’Énergies (HEDP) est caractérisée par desécoulements multi-matériaux fortement compressibles. Le domaine contenant l’écoulementsubit de grandes variations de taille et est le siège d’ondes de chocs et dedétente intenses. La représentation Lagrangienne est bien adaptée à la descriptionde ce type d’écoulements. Elle permet en effet une très bonne description deschocs ainsi qu’un suivit naturel des interfaces multi-matériaux et des surfaces libres.En particulier, les schémas Volumes Finis centrés Lagrangiens GLACE (GodunovtypeLAgrangian scheme Conservative for total Energy) et EUCCLHYD (ExplicitUnstructured Cell-Centered Lagrangian HYDrodynamics) ont prouvé leur efficacitépour la modélisation des équations de la dynamique des gaz ainsi que de l’élastoplasticité.Le travail de cette thèse s’inscrit dans la continuité des travaux de Maireet Nkonga [JCP, 2009] pour la modélisation de l’hydrodynamique et des travauxde Kluth et Després [JCP, 2010] pour l’hyperelasticité. Plus précisément, cettethèse propose le développement de méthodes robustes et précises pour l’extension3D du schéma EUCCLHYD avec une extension d’ordre deux basée sur les méthodesMUSCL (Monotonic Upstream-centered Scheme for Conservation Laws) et GRP(Generalized Riemann Problem). Une attention particulière est portée sur la préservationdes symétries et la monotonie des solutions. La robustesse et la précision duschéma seront validées sur de nombreux cas tests Lagrangiens dont l’extension 3Dest particulièrement difficile.

Autosomal dominant cerebellar ataxias (ADCAs) are a group of neurodegenerative disorders that are clinically and genetically heterogeneous. We report here a genetic linkage study, with five chromosome 12q markers, of three Martinican families with ADCA type 1, for which the spinocerebellar ataxia 1 (SCA1) locus was excluded. Linkage to the SCA2 locus was demonstrated with a maximal lead score of 6.64 at theta = 0.00 with marker D12S354. Recombinational events observed by haplotype reconstruction demonstrated that the SCA2 locus is located in an approximately 7-cM interval flanked by D12S105 and D12S79. Using the z(max)-1 method, multipoint analysis further reduced the candidate interval for SCA2 to a region of 5 cM. Two families shared a common haplotype at loci spanning 7 cM, which suggests a founder effect, whereas a different haplotype segregated with the disease in the third family. Finally, a mean anticipation of 12+/-14 years was found in parent-child couples, with no parental sex effect, suggesting that the disease might be caused by an expanded and unstable triplet repeat.

Autosomal dominant cerebellar ataxia type I was diagnosed in three unrelated families from Martinique (French West Indies), and linkage to the locus for spinocerebellar ataxia 2 (SCA2) was established. Neuropathological findings in two patients were those of olivopontocerebellar atrophy without oligodendroglial cytoplasmic inclusions. Cerebellar ataxia was associated with hyporeflexia in 68% of 31 examined patients, with slowed and/or limited eye movements in 65% and with dementia in 29%. No patients had optic atrophy, pigmentary retinal degeneration, spasticity or parkinsonism. Mean age at onset was 33 +/- 16 years, and onset before the age of 20 years was correlated with a more rapid and severe course of the disease. Movement disorders, oculomotor disturbances, sphincter disturbances and cognitive impairment were significantly more frequent in early than in late onset patients. This explains why the phenotype was strikingly different in one family, in which mean age at onset was much earlier. Comparison with previously described SCA2 families indicated similarities, such as reduced saccade velocity, supranuclear ophthalmoplegia and decreased reflexes, although phenotypic heterogeneity remains the outstanding feature of this disorder.