Carsten T. Lüth

Heidelberg, Germany

I am a deep learning researcher currently a PhD Student in the Interactive Machine Learning Group at the DKFZ / Helmholtz Imaging headed by Dr. Paul Jäger(DeepMind) and Heidelberg University. Currently, my main area of research is Deep Active Learning, which aims at reducing the annotation effort for supervised training by selecting the most informative samples for annotation. To find these most informative samples, I am also highly interested in uncertainty estimation, generative modeling. Further, I believe that the combination of Active Learning with alternate techniques to reduce the amount of annotated data for training algorithms (Semi- and Self-Supervised Learning and Foundation Models) will allow to use the power of deep learning for many small scale projects and research questions.

On the side, I am one of the organizers of heidelberg.ai, a community with over 2700 members on meetup, where we host events to encourage scientific discussions about artificial intelligence.

Further, I have past working experiences and am still interested in the fields of Deep Generative Methods and Anomaly Detection. Until recently, I studied physics at the University of Heidelberg, focusing on Computational Physics.

Generally, I like to read widely because: “You never know what you don’t know”. But Deep Learning, Physics, Psychology, Programming, and Statistics have captured my mind for a long time.

Also, during my free time, I never let a chance pass by to entertain myself with some fiction, just to let my imagination run wild. Whenever I am neither reading nor working, I do something like wakeboarding, bouldering, playing piano, cooking, meeting friends, or programming.

I also maintain an occasional blog—feel free to check it out and explore my thoughts and insights on various subjects. Thank you for visiting my page, and I look forward to connecting and sharing ideas with you.

selected publications

ValUES: A Framework for Systematic Validation of Uncertainty Estimation in Semantic Segmentation

Carsten T. Lüth^*, Kim C Kahl^*, Till J Bungert, and 3 more authors

ICLR 2024 (Oral Top 5%), Jan 2024

HTML Code
Navigating the pitfalls of active learning evaluation: A systematic framework for meaningful performance assessment

Carsten T. Lüth, Till J Bungert, Lukas Klein, and 1 more author

NeurIPS 2023, Dez 2023

HTML Code
A Call to Reflect on Evaluation Practices for Failure Detection in Image Classification

Paul F Jaeger, Carsten T. Lüth, Lukas Klein, and 1 more author

ICLR 2023 (Oral Top 1%), Jan 2023

HTML Code
cOOpD: Reformulating COPD Classification on Chest CT Scans as Anomaly Detection Using Contrastive Representations

Carsten T. Lüth^*, Silvia D Almeida^*, Tobias Norajitra, and 8 more authors

In MICCAI 2023 (STAR-Award), Oct 2023

HTML Code
Guided Image Generation with Conditional Invertible Neural Networks

Lynton Ardizzone, Carsten T. Lüth, Jakob Kruse, and 2 more authors

arXiv:1907.02392 [cs], Jul 2019

Abs HTML

In this work, we address the task of natural image generation guided by a conditioning input. We introduce a new architecture called conditional invertible neural network (cINN). The cINN combines the purely generative INN model with an unconstrained feed-forward network, which efficiently preprocesses the conditioning input into useful features. All parameters of the cINN are jointly optimized with a stable, maximum likelihood-based training procedure. By construction, the cINN does not experience mode collapse and generates diverse samples, in contrast to e.g. cGANs. At the same time our model produces sharp images since no reconstruction loss is required, in contrast to e.g. VAEs. We demonstrate these properties for the tasks of MNIST digit generation and image colorization. Furthermore, we take advantage of our bi-directional cINN architecture to explore and manipulate emergent properties of the latent space, such as changing the image style in an intuitive way.