Physics-informed neural networks can make black-box models more reliable. The result is versatile surrogate models for complex applications, such as the precise control of soft robots.
Neural Attention Search (NAtS) enables AI to focus on what truly matters — significantly reducing energy and hardware costs while maintaining performance.
Fed-FUEL enables fair, powerful and privacy-compliant AI in federated learning – without adapting the underlying models and with a special focus on protecting disadvantaged groups.
With its 40 finely graded emotional states and balanced dataset, EmoNet Face bridges the gap between 'emotionally blind' AI and the emotional perception of human experts.
DeepCAVE provides insights into complex AutoML optimisations, helping users understand key parameters, performance dynamics, and opportunities for improvement.
HyperSHAP reveals which hyperparameters truly matter and how they interact, bringing new transparency and efficiency to the optimisation of modern AI systems.
A combination of a pulsating fluid jet, audio monitoring and AI enables the safe and minimally invasive removal of bone cement – an important component for gentle, automatable revision surgery.
At the 10th L3S Town Hall Meeting on 6 February 2026, outstanding research results took centre stage alongside current developments and new projects. Prof. Dr. […]
AI systems that understand both images and text play a key role in technologies like search engines and autonomous vehicles – but they remain vulnerable to manipulation and bias. Researchers at L3S have developed a new training method that improves the reliability and fairness of these systems.
Sensors can become inaccurate over time, which can lead to incorrect decisions being made. L3S's AutoML solution automatically detects this drift and corrects it in real time, ensuring reliable data.
