Welcome
I am a PhD student in the Explainable Machine Learning group, part of the Computer Vision and Machine Learning Department (D2) at the Max Planck Institute for Informatics on the Saarland Informatics Campus. I am supervised by Dr. Jonas Fischer and Prof. Bernt Schiele.
I work on foundation models for vision, where I developed the initial concept for CFM. I am currently working on grounding the chain-of-thoughts in large vision language models.
News
- 09/2026: Presenting CFM at ECCV 2026 in Malmö as a spotlight presentation, and in the nectar tracks of the eXCV and FAILED workshops.
- 06/2026: Our paper on a language-aligned concept foundation model for vision was accepted at ECCV 2026 as a Spotlight.
- 02/2026: Joined the Explainable Machine Learning group in D2 at the Max Planck Institute for Informatics as a PhD student.
- 01/2026: CFM is on arXiv, with code and models on GitHub.
- 10/2025: Finished my master’s thesis, Family-CBM: Automatic Discovery of Hierarchies in Concept Bottleneck Models, supervised by Sukrut Rao and Jonas Fischer.
Publications
Language-aligned vision foundation models perform strongly across diverse downstream tasks. Yet, their learned representations remain opaque, making interpreting their decision-making difficult. Recent work decompose these representations into human-interpretable concepts, but provide poor spatial grounding and are limited to image classification tasks. In this work, we propose CFM, a language-aligned concept foundation model for vision that provides fine-grained concepts, which are human-interpretable and spatially grounded in the input image. When paired with a foundation model with strong semantic representations, we get explanations for any of its downstream tasks. Examining local co-occurrence dependencies of concepts allows us to define concept relationships through which we improve concept naming and obtain richer explanations. On benchmark data, we show that CFM provides performance on classification, segmentation, and captioning that is competitive with opaque foundation models while providing fine-grained, high quality concept-based explanations. Code at https://github.com/kawi19/CFM.
CFM: Language-aligned Concept Foundation Model for Vision
Vision foundation models perform well but are opaque. CFM maps every image patch to a sparse set of named concepts, which gives concept-based explanations and steering for classification, segmentation and captioning without giving up the original model’s performance. Also presented in the nectar tracks of the eXCV and FAILED workshops at ECCV 2026.
Talks
CFM: Language-aligned Concept Foundation Model for Vision
September 2026
ECCV 2026, spotlight presentation (Malmö, Sweden)
Academic Activities
Open Source
Multilingual Language Model Fine-Tuning
A study of what full fine-tuning, BitFit, LoRA and IA3 do to a multilingual model’s hidden states, with a focus on under-represented languages.