cv

PhD Candidate

University of Amsterdam

Pioneering transformer model interpretability through monosemanticity engineering for enhanced AI safety. Focused on AI Alignment strategies to ensure long-term value preservation.

• AI Safety
• Mechanistic Interpretability
• Transformer Models
• Monosemanticity Engineering

Research Assistant

University of Heidelberg

Infused dendritic computation principles into neural networks. Explored novel optimization criteria for dynamical systems.

• Dendritic Computation
• Neural Networks
• Dynamical Systems

Research Intern

Central Institute of Mental Health

Investigated initialization schemes for a piecewise-linear recurrent neural network using expectation-maximization.

• Recurrent Neural Networks
• Initialization Schemes
• Expectation-Maximization

Co-founder and Core Team Member

AI Safety Initiative Amsterdam

Co-founded and actively contribute to a group dedicated to promoting AI safety research and awareness in Amsterdam.

• Organized OpenAI Talk and Q&A on AI and Existential Risk
• Coordinated Panel Discussion on AI Risks: From Today to Doomsday
• Facilitated reading groups on AGI Safety Fundamentals

PhD

University of Amsterdam

Artificial Intelligence

• Continual Learning
• Mechanistic Interpretability

MSc

University of Heidelberg

Computational Physics

• Visual Learning and Computer Vision
• Machine Learning
• Artificial Intelligence
• Time Series Analysis

Exchange Student

National Taiwan University

Mandarin Chinese

• Advanced-level Mandarin Chinese

BSc

University of Heidelberg

Physics

• Analysis
• Linear Algebra
• Statistical Physics
• General Relativity

High School

Gymnasium Ernestinum

Abitur

• Physics, Mathematics, Chemistry
• Latin, History

Prized by German Mathematical, Physical, and Chemical Societies

German Mathematical, Physical, and Chemical Societies

Recognized for outstanding performance in mathematics, physics, and chemistry during Abitur.

2nd Place, AI Safety Hackathon

Entrepreneur First and Apart Research

Achieved second place in the AI Safety Hackathon held in Delft.

Mechanistic Interpretability for AI Safety - A Review

CoRR

Comprehensive review of mechanistic interpretability in AI, focusing on its applications in AI safety and potential to prevent catastrophic outcomes.

Taming Simulators: Challenges, Pathways and Vision for the Alignment of Large Language Models

AAAI Inaugural Summer Symposium Series

Explored challenges and strategies for aligning large language models, emphasizing the concept of simulators and simulacra in AI alignment.

Continual Learning of Dynamical Systems With Competitive Federated Reservoir Computing

Conference on Lifelong Learning Agents

Proposed a novel approach to continual learning of dynamical systems using competitive federated reservoir computing.

Tractable Dendritic RNNs for Reconstructing Nonlinear Dynamical Systems

International Conference on Machine Learning (ICML)

Developed an interpretable and tractable piecewise-linear RNN augmented with a linear spline basis expansion for inferring nonlinear dynamical systems.

Unsupervised Part-Based Disentangling of Object Shape and Appearance

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

Presented a novel approach for unsupervised disentanglement of object shape and appearance in computer vision (oral presentation, best paper finalist).