Amir Mohammad Karimi Mamaghan

I’m a WASP and ELLIS Ph.D. student at the Department of Decision and Control Systems (DCS) of KTH Royal Institute of Technology under the supervision of Karl Henrik Johansson and Stefan Bauer. I completed my master’s degree in Computer Engineering (Artificial Intelligence and Robotics) at the University of Tehran. During my master’s studies, I was a research assistant at Data Analytics Lab under the supervision of Behnam Bahrak.

Research

My main research interests lie in representation learning, inductive biases such as causal and object-centric learning, and training dynamics of deep transformer models. In particular, I am interested in understanding how high-level structure emerges in learned representations and how inductive biases influence this process. More recently, I have been exploring how training strategies such as progressive depth growth shape representations in self-supervised vision transformers.

I am also interested in applied ML in protein design, particularly in settings where experimental data is scarce. I am especially interested in using advances in ML to design high-affinity protein binders.

Supervision

• “Handbook of Growing: A Practical Guide to Depth Growing for Language Models,” TUM Data Innovation Lab, 2026. Co-supervised with Ferdinand Kapl, Vincent Pauline, Tobias Höppe, and Stefan Bauer.

• "Causal Adaptive Reinforcement for Effective Hospital Admission Interventions" by Nasir Alizade, 2026.

• "Causally Informed Reinforcement Learning for Hospital Readmission Reduction After ICU Care" by Valter Strandler, 2026.

• "Causal Reasoning for Predictive Health Modeling on EHR data" by Yuyang Tao, 2025.

Teaching

• Teaching assistant, EL2810 Machine Learning Theory, KTH Royal Institute of Technology (2024-2025)