Protein representation learning using graph neural networks and Alphafold generated 3D structures.

Custom multi-modal CLIP-based embedding and captioning models fine-tuned on remote sensing data using an additional self-distillation loss to mitigate intra-class similarities

Physarum Powered Differentiable Linear Programming Layers