Search

Deep Ensembles (DE) are a prominent approach for achieving excellent performance on key metrics such as accuracy, calibration, uncertainty estimation, and out-of-distribution detection. However, hardware limitations of real-world systems constrain to smaller ensembles and lower-capacity networks, significantly deteriorating their performance and properties. We introduce Packed-Ensembles (PE), a strategy to design and train lightweight structured ensembles by carefully modulating the dimension of their encoding space. We leverage grouped convolutions to parallelize the ensemble into a single shared backbone and forward pass to improve training and inference speeds. PE is designed to operate within the memory limits of a standard neural network. Our extensive research indicates that PE accurately preserves the properties of DE, such as diversity, and performs equally well in terms of accuracy, calibration, out-of-distribution detection, and robustness to distribution shift. We make our code available at https://github.com/ENSTA-U2IS/torch-uncertainty., Comment: Published as a conference paper at ICLR 2023 (notable 25%)

Lignocellulose biomass has a tremendous potential as renewable biomaterials for fostering the “bio-based society” and circular bioeconomy paradigm. It requires efficient use and breakdown of fiber cell walls containing mainly cellulose, hemicellulose and lignin biopolymers. Despite their great importance, there is an extensive debate on the true structure of fiber walls and knowledge on the macromolecular nano-organization is limited and remains elusive in 3D. We employed dual-axis electron tomography that allows visualization of previously unseen 3D macromolecular organization/biopolymeric nano-architecture of the secondary S2 layer of Norway spruce fiber wall. Unprecedented 3D nano-structural details with novel insights into cellulose microfibrils (~ 2 nm diameter), macrofibrils, nano-pore network and cell wall chemistry (volume %) across the S2 were explored and quantified including simulation of structure related permeability. Matrix polymer association with cellulose varied between microfibrils and macrofibrils with lignin directly associated with MFs. Simulated bio-nano-mechanical properties revealed stress distribution within the S2 and showed similar properties between the idealized 3D model and the native S2 (actual tomogram). Present work has great potential for significant advancements in lignocellulose research on nano-scale understanding of cell wall assembly/disassembly processes leading to more efficient industrial processes of functionalization, valorization and target modification technologies.

Bioenergy will be one of the most important renewable energy sources in the conversion from fossil fuels to bio-based products. Short rotation coppice Salix could be a key player in this conversion since Salix has rapid growth, positive energy balance, easy to manage cultivation system with vegetative propagation of plant material and multiple harvests from the same plantation. The aim of the present paper is to provide an overview of the main challenges and key issues in willow genetic improvement toward sustainable biofuel value chains. Primarily based on results from the research project "Optimized Utilization of Salix" (OPTUS), the influence of Salix wood quality on the potential for biofuel use is discussed, followed by issues related to the conversion of Salix biomass into liquid and gaseous transportation fuels. Thereafter, the studies address genotypic influence on soil carbon sequestration in Salix plantations, as well as on soil carbon dynamics and climate change impacts. Finally, the opportunities for plant breeding are discussed using willow as a resource for sustainable biofuel production. Substantial phenotypic and genotypic variation was reported for different wood quality traits important in biological (i.e., enzymatic and anaerobic) and thermochemical conversion processes, which is a prerequisite for plant breeding. Furthermore, different Salix genotypes can affect soil carbon sequestration variably, and life cycle assessment illustrates that these differences can result in different climate mitigation potential depending on genotype. Thus, the potential of Salix plantations for sustainable biomass production and its conversion into biofuels is shown. Large genetic variation in various wood and biomass traits, important for different conversion processes and carbon sequestration, provides opportunities to enhance the sustainability of the production system via plant breeding. This includes new breeding targets in addition to traditional targets for high yield to improve biomass quality and carbon sequestration potential.

Cerebral cavernous malformation (CCM) is a neurovascular disease that results in various neurological symptoms. Thrombi have been reported in surgically resected CCM patient biopsies, but the molecular signatures of these thrombi remain elusive. Here, we investigated the kinetics of thrombi formation in CCM and how thrombi affect the vasculature and contribute to cerebral hypoxia. We used RNA sequencing to investigate the transcriptome of mouse brain endothelial cells with an inducible endothelial-specific Ccm3 knock-out (Ccm3-iECKO). We found that Ccm3-deficient brain endothelial cells had a higher expression of genes related to the coagulation cascade and hypoxia when compared with wild-type brain endothelial cells. Immunofluorescent assays identified key molecular signatures of thrombi such as fibrin, von Willebrand factor, and activated platelets in Ccm3-iECKO mice and human CCM biopsies. Notably, we identified polyhedrocytes in Ccm3-iECKO mice and human CCM biopsies and report it for the first time. We also found that the parenchyma surrounding CCM lesions is hypoxic and that more thrombi correlate with higher levels of hypoxia. We created an in vitro model to study CCM pathology and found that human brain endothelial cells deficient for CCM3 expressed elevated levels of plasminogen activator inhibitor-1 and had a redistribution of von Willebrand factor. With transcriptomics, comprehensive imaging, and an in vitro CCM preclinical model, this study provides experimental evidence that genes and proteins related to the coagulation cascade affect the brain vasculature and promote neurological side effects such as hypoxia in CCMs. This study supports the concept that antithrombotic therapy may be beneficial for patients with CCM.

A host of physical and environmental factors may influence fungal decay including the wood substrate, temperature, moisture, oxygen, light, pH, and nitrogen. Understanding the effects of these factors on fungal decay is important for the effective utilization of wood decay fungi in biotechnological processes and for understanding the role of these organisms in global carbon cycling. The ascomycete Xylaria hypoxylon causes white rot of hardwoods, but remains relatively under-characterized. In this investigation, the decay capability of this fungus was studied using beech, hornbeam, oak and pine. Although Xylaria species are considered as causing white rot decay, Type II soft rot erosion was observed on hornbeam, Type I soft rot cavities were noted on beech, simultaneous rot was found on oak and selective rot on pine. Results indicated that both wood species and cell wall chemistry affected morphological decay patterns illustrating the relationships between fungal enzymatic capacity and wood cell wall structure/chemistry.