zain khan

I'm conducting deep learning research at LUMS's EmNet Lab and Knowledge & Data Engineering Lab, with three coauthored papers in preparation, including work under the UKRI-funded S2Cool project.

At EmNet Lab, I engineered sensor-data cleaning and profile-driven synthesis pipelines for indirect evaporative cooling, then trained BiLSTM and Transformer models for real-time supply-air temperature prediction. My best model reached an RMSE of 0.586°C and R² of 0.938, while smooth synthesis reduced error by about 25% versus random synthesis.

Previously, I benchmarked seven ML and deep learning models on an operational IEC prototype dataset, achieving R² of 0.9997 and RMSE of 0.046°C. I also build end-to-end PyTorch systems across low-light enhancement, industrial digital twins, diabetic retinopathy classification, and hematological diagnosis.

For my BioNAFNet final-year project, I designed a low-light image-enhancement architecture that surpassed Retinexformer on LOLv1 and compressed it 6.4x through knowledge distillation with no quality loss. I'm motivated by rigorous benchmarking, practical ML pipelines, and research that turns complex data into reliable models.