GSoC 2026 DeepLense – Guidance on improving lensing model

[Sasisundar2211]

@sachdevkartik

Hi Kartik,

I applied for GSoC 2026 with the DeepLense project and have been working on the Genie tasks.

I implemented the autoencoder and a GNN pipeline by converting jet images into graph representations using non zero pixels and k nearest neighbor edges.

I also ran a comparison across CNN, ResNet, and GNN models with evaluation metrics to understand performance differences.

Here is my updated work:
https://github.com/Sasisundar2211/ml4sci-gsoc

I am now focusing on improving the graph representation and model performance. From your experience, which direction would add the most value at this stage?

Thanks,
Sasi