The No-IDLE project marks the successful launch of the first strategic corridor initiative by DFKI-IML, funded by the Federal Ministry for Research, Technology, and Space (BMFTR). Completed on March 31, 2026, this three-year endeavor focused on making complex deep learning (DL) methodologies accessible to non-experts through interactive deep learning frameworks. We extend our gratitude to the BMFTR for their support in advancing the methodological foundations of interactive machine learning, ensuring Germany remains at the forefront of human-centered AI innovation.
Key project details at a glance:
IML is an emerging paradigm that integrates human feedback into machine learning. This requires parallel progress along two essential dimensions: human-centered interaction design and technical innovation.
No-IDLE adopts the binocular view from the field of Intelligent User Interfaces (IUI), emphasizing the co-design of algorithms and interfaces rather than treating them as isolated components. The overall goal is to enable users to perform application tasks with improved usability and user experience, while ensuring that human-AI teams using IML technology achieve greater efficiency and effectiveness than humans or AI systems working alone.
No-IDLE’s example application is interactive photobook creation, a highly creative, individual, and iterative multimedia creation task. Further, No-IDLE considered technology transfer to the healthcare domain, where explainability and accountability are vital, as well as learning under low-resource conditions.
Key outcomes:
- Development of NLP components for generating personalized and context-sensitive captions (and refining them through natural-language feedback) and for modeling textual explanations in accordance with the user expertise
- Development of ML components for improving active learning processes and XAI methods for improving explainability of ML models
- Elaboration of user requirements and interaction designs in the areas of Human-Computer Interaction (HCI) and Multimodal-Multisensor Interaction (MMI)
- Development and evaluation of interactive DL-based user interfaces for the annotation of multimodal data streams and of decision support systems in the transfer domain Medicine
- Development of a demonstrator in the subject area of interactive photobook creation
- Publication of 50 peer-reviewed scientific reports, including 22 full papers
- Completion of 17 student theses and 2 dissertations
sponsored by:
