Atri Sharma

Ai Research Scientist @Safe Intelligence

London, GB

Leading the robustness verification and learning initiatives for decision tree ensembles (e.g, XGBoost). • Developed a novel gradient boosting algorithm that displays improved robustness and stability to adversarial evasion attacks (minor imperceptible input variations), relevant for applications in safety critical domains such as fraud detection, loan approvals, and trading. • Developed a novel method to evaluate robustness of regression models using a Mixed-Integer Linear Programming solver (Gurobi). • Defined robustness specifications for tabular financial data, ensuring reliability and compliance for automated loan approval models.

Imperial College London

Master of Engineering - MEng, Aeronautical Engineering

Imperial College London

Doctor of Philosophy - PhD, Computer Science