Philipp Becker
I am currently pursuing a PhD in Machine Learning at the Karlsruhe Institute of Technology (KIT), supervised by Prof. Gerhard Neumann. During my PhD studies, I was involved with the FZI Research Center for Information Technology, working towards establishing a new research group under Prof. Neumann's direction from spring 2023 to summer 2024. Before my time at KIT, I worked with the Bosch Center for Artificial Intelligence and the University of Tübingen for eight months. Furthermore, I have a Bachelor's degree in Computer Science and a Master's degree in Autonomous Systems (Computer Science) from the Technical University Darmstadt.
I am currently interning at the Samsung AI Center in Cambridge and aim to complete my PhD by Summer 2025.
Research
My research focuses on world models for Reinforcement Learning with multimodal and high-dimensional observations. Here, I combine probabilistic state space models with deep learning to learn dynamics models and concise representations as published in:
- Combining Reconstruction and Contrastive Methods for Multimodal Representations in RL (RLC 2024)
- On Uncertainty in Deep State Space Models for Model-Based Reinforcement Learning (TMLR 2022)
- Recurrent Kalman Networks: Factorized Inference in High-Dimensional Deep Feature Spaces (ICML 2019)
Beyond this, I worked on projects covering a wide range of topics such as
- Imitation Learning for Versatile Behavior
- Dynamics and Representation Learning for Robotics
- Model-Free RL with a Focus on Trust Region Methods
- Bayesian Deep Learning
Here is a list of all my publications.
Updates
Started Internship at Samsung AI Center Cambridge
Published:
I am excited to share that I started a research internship at the Samsung AI Center in Cambridge. During the internship, I’ll focus on efficient Text-to-Image generation by including recent structured state space approaches into diffusion models.
New Paper at Reinforcement Learning Conference
Published:
I am happy to announce that my paper Combining Reconstruction and Contrastive Methods for Multimodal Representations in RL got accepted at the new Reinforcment Learning Conference (RLC).