Developed a deep learning framework using ResNeXt50 to classify benign versus malignant breast tumors from histopathology images, achieving reliable performance through hybrid loss design and robust multi-scale feature learning.

Proposed a hybrid spiking–transformer architecture for EEG-based motor imagery decoding, integrating adaptive LIF neurons with spiking multi-head attention to enable temporally sparse, energy-efficient, and biologically inspired BCI inference.