The whiteboard: 3 papers

Pick a paper: the desk shows its run, its results and the acceptance email, and the printer prints it.

Partially Encrypted Federated Vision Transformers for Remote Sensing Scene Classification

N. Alam, et al. 4th International Conference on Computing Advancements (ICCA 2026), ACM, Oct 2026. First author, presented.

~95% About 95% accuracy on NWPU-RESISC45 with encryption on, on par with unencrypted training.

ViT-Base trained with FedAvg, with CKKS homomorphic encryption applied to the aggregation of selected layers.

ScaleRAG. Retrieval for frozen time-series foundation models: 5.49% lower RMSSE than Chronos-2 on the M5 panel. In preparation. Code

Done, not published: my undergraduate thesis on action recognition in video, in a team of three. See the thesis