Florian Blume

• Dissertation: “Knowledge-Augmented and Context-Sensitive Face Perception” within interdisciplinary excellence cluster Science of Intelligence
• Goal: Develop multi-modal deep-learning models for context-sensiGve facial expression recognition, integrating vision, audio, and language.
• Developed a theoretical framework for the topic from a joint psychological – ML viewpoint, combined with a literature analysis (Consciousness and Cognition ‘22).
• Developed a multi-task multi-modal self-supervised learning (SSL) approach to shape representation space. Combined vision models with an audio model and an LLM, improved accuracy by more than 5% over other SSL approaches that use the same modalities (CVPR’24).
• Fused generative AI models with a self-attention and a classification module to classify and simultaneously generate context-augmented facial expressions. Generations were verified as approximations of mental representations in a study with human participants (GCPR’24).

• Master’s thesis: Implemented and compared four reinforcement learning approaches for improving denoising biological microscopy images of cells
• Capstone project: Developed a 6D pose annota6on tool and CNNs for surgical instrument localizaGon in laparoscopic videos

Development of an application to create mobile machine control apps.

Assisting in running the computer lab.

Java-based backend development.

Development of iOS radio apps.