Convolutional neural networks for analysis of cortical anatomy; experiments with different network architectures and input formats + data augmentation
-
Data augmentation:
- angle jitter
-- TO DO: -- - label noise
- training via Contrastive Learning
- other curricula / scheduled learning
- angle jitter
-
Architectures:
- transfer learning with ResNet
- self-attention layers in the style of SAGAN
- VAE-like image-reconstruction error
-- TO DO: -- - GAN?
- Spline n.n. final layers?
- Geometric deep learning (e.g., ConvNet with discrete Riemannian metric)