27.03.2025
Beyond the Black Box: Choosing the Right Feature Importance Method
MCML Research Insight - With Fiona Katharina Ewald, Ludwig Bothmann, Giuseppe Casalicchio and Bernd Bischl
Machine learning models make powerful predictions, but can we really trust them if we don’t understand how they work? Global feature importance methods help us discover which factors really matter - but choosing the wrong method can lead to misleading conclusions. To see why this is important, …
©TUM
24.03.2025
Azade Farshad Receives BVM Award for Top Medical Imaging Research
Our junior member Azade Farshad was awarded the BVM Award for the best thesis in medical image processing. Her work, supervised by Nassir Navab, was recognized for its innovation and excellence. The award was presented at the German Conference on Medical Image Computing (BVM), held at OTH Regensburg. …
©aiforgood
21.03.2025
Explainable Multimodal Agents With Symbolic Representations & Can AI Be Less Biased?
Our Junior Member Ruotong Liao at United Nations AI for Good
More than 170 audiences visited the online lecture of our Junior Member Ruotong Liao on Monday, 17. March 2025, as an invited speaker at the United Nations “AI for Good”. With her talk “Perceive, Remember, and Predict: Explainable Multimodal Agents with Symbolic Representations,” Ruotong Liao took …
©TUM
18.03.2025
New Method Significantly Reduces AI Energy Consumption
TUM News Article
Researchers at the Technical University of Munich have developed an innovative method that drastically lowers the energy consumption of artificial intelligence systems. The approach optimizes computational efficiency, making AI applications more sustainable and cost-effective. Our Associate Felix Dietrich emphasized the importance of …
13.03.2025
ReNO: A Smarter Way to Enhance AI-Generated Images
MCML Research Insight - With Luca Eyring, Shyamgopal Karthik, Karsten Roth and Zeynep Akata
Despite their impressive capabilities, Text-to-Image (T2I) models frequently misinterpret detailed prompts, leading to errors in object positioning, attribute accuracy, and color fidelity. Traditional improvements rely on extensive dataset training, which is not only computationally expensive but also may not generalize well to unseen prompts. To …
10.03.2025
Björn Schuller Was Featured on ARTE TV
Therapy via AI
Our PI Björn Schuller was featured in the new ARTE TV documentary on “Therapy via AI." In the documentary, he briefly outlined the focus of his research. The AI program he presented is capable of recognizing the emotion behind spoken words, demonstrating how artificial intelligence can analyze vocal expressions to assess emotional states. Together …