• Publications
  • Influence
Local Spectral Graph Convolution for Point Set Feature Learning
TLDR
This article replaces the standard max pooling step with a recursive clustering and pooling strategy, devised to aggregate information from within clusters of nodes that are close to one another in their spectral coordinates, leading to richer overall feature descriptors. Expand
Affinity Graph Supervision for Visual Recognition
TLDR
It is shown that affinity learning between objects boosts scene categorization performance and that the supervision of affinity can also be applied to graphs built from mini-batches, for neural network training. Expand
Robust environment mapping using flux skeletons
TLDR
The online construction of a topological map and the implementation of a control law for guiding the robot to the nearest unexplored area are presented and the results demonstrate complete mapping of different environments with smooth topological abstraction without spurious edges. Expand
Cartan Frames for Heart Wall Fiber Motion
TLDR
A moving frame method is developed that shows promise in providing Cartan connection parameters to describe spatio-temporal rotations of fibers, which in turn could benefit subsequent analyses or be used for diagnostic purposes. Expand
FAN: Focused Attention Networks
TLDR
This work demonstrates that the focused supervision leads to improved attention distribution across meaningful entities, and that it enhances the representation by aggregating features from them, and introduces a focused attention backbone to learn attention weights for general tasks. Expand
Minimizing Non-holonomicity: Finding Sheets in Fibrous Structures
TLDR
This work proposes an algorithm which, given an input vector field, finds a second one with which it best spans a sheet-like structure locally, by minimizing non-holonomicity, and provides high quality sheet reconstructions from both heart wall DTI data and labeled tracts in the human brain. Expand
Denoising Moving Heart Wall Fibers Using Cartan Frames
TLDR
A geometric approach, with a spatio-temporal Cartan frame field to model spatial and temporal rotations within a single consistent mathematical framework, and the promise of this approach for outlier detection and denoising via inpainting is demonstrated. Expand
Estimating Sheets in the Heart Wall
TLDR
An alternate method to estimate sheets, which uses only the principal eigenvector, which is more consistent with the geometry of myofibers than those obtained using all three eigenvectors, particularly when they curve or fan. Expand