About Me

I’m Jasraj Singh, a machine learning researcher with a passion for understanding how neural networks learn and how we can make them better. My work spans graph neural networks, large language models, and neural active learning, with a focus on theoretical insights, mechanisitic interpretations, and efficiency in deep learning.

I recently graduated with a Master’s in Machine Learning from University College London (UCL), earning first-class honors and a place on the Dean’s List. Advised by Prof. Laura Toni and Prof. Brooks Paige, my thesis research explored the limitations of edge-dropping algorithms like DropEdge in Graph Neural Networks (GNNs). I extended this theoretical analysis to 6 dropout-like algorithms commonly used for training deep GNNs, empirically demonstrating their negative effects on long-range learning tasks. Read more.

Before that, I earned a Bachelor’s in Mathematical and Computer Sciences from Nanyang Technological University (NTU), where I worked on my thesis research with Prof. Bryan Low at the National University of Singapore. We introduced Expected Variance with Gaussian Processes (EV-GP) – an active learning criterion with initialization robustness guarantees, leveraging Neural Tangent Kernel (NTK) theory. Read more.

In between, I explored language models and cognitive science at the Visual and Cognitive Neuroscience Lab. Collaborating with Prof. Xu Hong and Liu Fang, I helped develop LingML, an efficient way to incorporate linguistic knowledge into LLMs, achieving competitive performance against state-of-the-art methods at a fraction of the data and compute cost. Read more.

Beyond research, I have hands-on machine learning engineering experience at Shopee and Indeed, where I worked on real-time recommendation models, predictive analytics, and A/B testing. Through these roles, I developed a strong foundation in SQL, Python, PyTorch, TensorFlow, and large-scale data processing pipelines.