What I Work On
Research Areas
Eleven interconnected areas spanning security, healthcare, efficient deep learning, and multimodal AI — each grounded in reproduced foundational papers and applied projects.
tree ./categories/
AI for Cyber Security
This area focuses on applying deep learning to network intrusion detection, malware classification, and phishing detection, with a strong emphasis on adversarial robustness — since security models face an adaptive adversary, not just a static test set.
AI for Healthcare
Work here targets diagnostic support systems — medical image classification, EHR-based risk prediction — designed to operate under strict privacy constraints and across heterogeneous hospital data distributions.
Computer Vision
Core computer vision work: convolutional and transformer-based backbones, self-supervised representation learning, and their transfer to specialized domains like security and medical imaging.
Machine Learning
General machine learning foundations underpinning all applied work: evaluation methodology, statistical rigor, feature engineering, and reproducibility tooling for experiments.
Deep Learning
Deep dives into architecture design and optimization dynamics — from residual connections to attention mechanisms — grounded in hands-on reproduction of the papers that introduced them.
Federated Learning
Federated learning enables model training across hospitals, banks, or devices without centralizing raw data. My focus is on non-IID data handling, communication efficiency, and privacy guarantees under FedAvg-style aggregation.
Continual Learning
Continual learning studies how models can adapt to new attack patterns or disease cohorts over time without forgetting previously learned knowledge — critical for security systems and clinical models deployed for years.
Related Projects
LLM
Practical work on adapting large language models efficiently — parameter-efficient fine-tuning, quantization, and evaluation of instruction-following behavior on domain-specific tasks.
Related Papers
Related Projects
Vision Language Model
Vision-language models bridge perception and language. My reproductions and projects here explore contrastive pretraining and its downstream transfer to zero-shot classification and retrieval.
Related Projects
Edge AI
Edge AI research focuses on getting models from notebook to device: quantization, pruning, and runtime selection for inference under tight latency, memory, and power constraints.
Related Papers
Related Projects
TinyML
TinyML pushes inference onto microcontroller-class hardware. I work on compressing intrusion-detection and anomaly-detection models to run within kilobytes of RAM on ARM Cortex-M devices.