Publications by authors named "Mario Geiger"
Leveraging SO(3)-steerable convolutions for pose-robust semantic segmentation in 3D medical data.
J Mach Learn Biomed Imaging
May 2024
Article Synopsis
- Convolutional neural networks (CNNs) enhance parameter sharing and translational equivariance using convolutional kernels, and adjusting these to be SO(3)-steerable improves their efficacy.
- These rotationally-equivariant convolutional layers offer benefits like increased robustness to unseen poses, reduced network size, and better sample efficiency compared to standard convolutional layers.
- The authors introduce a new family of segmentation networks utilizing equivariant voxel convolutions based on spherical harmonics, achieving improved performance in MRI brain tumor segmentation without needing rotation-based data augmentation.
Neurological conditions are the leading cause of disability and mortality combined, demanding innovative, scalable, and sustainable solutions. Brain health has become a global priority with adoption of the World Health Organization's Intersectoral Global Action Plan in 2022. Simultaneously, rapid advancements in artificial intelligence (AI) are revolutionizing neurological research and practice.
View Article and Find Full Text PDFArticle Synopsis
- - NequIP is a new neural network model that uses E(3)-equivariant convolutions to learn interatomic potentials from quantum mechanical calculations, improving how atomic environments are represented.
- - This approach achieves top-tier accuracy across various molecules and materials, showing the ability to work effectively with much less training data than traditional models.
- - By requiring significantly fewer training samples, NequIP challenges the notion that deep learning networks need large datasets, enabling accurate simulations of molecular dynamics over extended time periods.
Deep learning has been immensely successful at a variety of tasks, ranging from classification to artificial intelligence. Learning corresponds to fitting training data, which is implemented by descending a very high-dimensional loss function. Understanding under which conditions neural networks do not get stuck in poor minima of the loss, and how the landscape of that loss evolves as depth is increased, remains a challenge.
View Article and Find Full Text PDFAdvanced optical ray tracing software, CFSpro, was developed for the study and optimization of complex fenestration systems (CFSs). Using an algorithm mixing 2D and 3D approaches, accurate computation of large numbers of rays in extruded geometries can be performed and visualized in real time. A thin film model was included to assess the spectral control provided by coatings.
View Article and Find Full Text PDF