Rob Romijnders

I'm a PhD student at the University of Amsterdam, where I focus on (federated) machine learning and differential privacy. Previously, I was an AI resident at Google, where I worked on representation learning, calibration, and robustness. My MSc degree is in Electrical Engineering.

Email  /  CV  /  Scholar  /  Twitter  /  Github

Details

I have done three internships: at Brave, focused on LLMs, at G-Research, focused on foundation models, and at Apple, focused on privacy-preserving machine learning. My PhD is sponsored by Qualcomm, and part of the ELLIS program. I received a Outstanding Reviewer award at ICCV 2021, and received an ELSA scholarship for a research visit with Prof. Antti Honkela. Below is an overview of my academic work, some are highlighted.

Studying knowledge transfer, modularity, and privacy in Large Language Models.

Improving the privacy-utility trade-off of my previous work on differential privacy in a decentralized setting.

Contact tracing with provable differential privacy guarantees in a decentralized setting.

A communication-efficient approach to decentralized, statistical contact tracing.

Examining robustness in out-of-distribution detection under dataset shift and training variations.

Introducing co-finetuning for more effective video understanding and localization.

Analyzing how aliasing affects generalization in convolutional networks.

A deeper look into calibration issues in deep neural networks.

Studying the generalization of CNNs under robustness and domain shifts.

A synthetic benchmark dataset to evaluate object-centric robustness.

Object-centric unsupervised representation learning from real-world video.

Enhancing semantic segmentation with selective and weakly supervised data.

Predicting basketball trajectories with sequential deep learning models.

A study of normalization layers and inter-domain dependencies in robust synthetic-to-real domain adaptation.

Miscellanea

Credits for the original code of this page.