Building instance imblance invariant networks, losses and metrics for Semantic, Instance, Panoptic and Part Aware Segmentation.

Engineering Triton kernels specifically to enhance segmentation pipelines.

With Yang Li, we are developing end-to-end hardware + software stack for FL on the Edge and work on the hyper-optimisation of neural networks for edge deployment.

Investigate all sorts of parallelism strategies to massively accelerate large model training and inference.

Orchestration of scalable pipelines for acquisition, filtering, and curation of high-quality, multi-lingual code datasets to train and test large code models.