Paris, Île-de-France, France
• Self-study formal methods/formal verification: TLA+ (specification language with associated model checker), Coq, Lean (theorem provers), abstract interpretation, type theory, depend types, Hoare logic, algebra-driven design • Self-study functional languages: Haskell • Self-study more Category Theory, including string diagrams which are applicable to software and system design
• Empower my local community to form unlikely connections • Find and train mentors to support high school students outside the margins through the power of tutoring and kindness
• Read existing literature, performed original research on visualizing & interpreting Transformer vision models • Collaborated in team of 3 applying existing and original models, wrote paper and presented to 90+ people • Prior to our work, there had been no applications of SHAP to transformer-based image classifiers • Used PyTorch to train 3 powerful models on dataset of 60k images, all obtained high accuracies (90%+) • To interpret those models, we trained _surrogates_ on those models, also obtaining high accuracies (~90%) • Demonstrated FastSHAP's usefulness for the first time on transformer-based image classifiers