👋 Hi. I’m Can, a machine learning engineer based in 🇩🇪 Hamburg, Germany with a focus on explainable AI.

can.rager[at]posteo.de

Scholar / CV / LinkedIn

Articles & Publications

Publications cSite

Experience

AI Explainability Researcher

in collaboration with Northeastern University | Remote | Oct 2022 – Present

I am delivering ML engineering projects which may serve to assess models for EU AI Act compliance. Currently, I develop AI explainability methods supervised by Samuel Marks and Prof. David Bau at Northeastern University in Boston. We specifically examine automatic circuit dicovery using sparse dictionaries in transformer models. Browse through our resuls in our interactive app: https://circuit-clustering.streamlit.app/ Previously, I published:

• Automatic Circuit Dicovery with Aaquib Syed and Arthur Conmy
• Neuromorphic Hardware Safety with Dr. Kyle Webster
• Maze-Solving Transformer Models at AI Safety Camp

Research Engineering Trainee

ARENA | London, UK | Jun 2023 – Oct 2023

• Performed mechanistic interpretability and distributed model training
• Published a novel method to examine Memory Management in Transformer Models
• Developed an automated interpretability technique offering substantial speed-up while at least retaining SOTA performance of 0.96 AUC