Mohd Aamir

• Engineered state-of-the-art anomaly detection pipeline for multivariate time series data by adapting deep learning models enabling direct modeling of missing values and improving anomaly detection F1-score by 12% over Huawei's baseline models.

• Benchmarked 12+ architectures (e.g. LSTM/TCN/Transformer) on synthetic and confidential Huawei datasets; implemented algorithms to inject controlled missing timestamps for testing.

• Integrated 3 pre-trained diffusion models into MLflow, enabling real-time inference of missing 3D holographic views during teaching sessions for the EU Adroit6G project.

• Validated outputs through various KPIs such as SSIM/PSNR, sustaining over 95% SSIM on generated views.

• Collaborated with 20+ researchers and industry partners across Europe, representing EURECOM in bi-weekly consortium meetings and delivering technical papers and work packages.

• Generated a dataset of 50,000 noisy raw images by reversing Huawei's proprietary image processing pipeline, enabling supervised training for raw image denoising.

• Implemented a CNN-based raw image denoising model in PyTorch, pruned CNN to cut inference time by 21% with PSNR drop less than 3 dB.

• Optimized training pipelines/hyperparameters, reducing training time by 15% on Huawei's GPU clusters.