Fraunhofer Institute for Open Communication Systems
Fraunhofer Institute for Open Communication Systems
Humanoider Roboterarm mit Stab in der Hand vor blauen Hintergrund mit graphischen Elementen wie Wellenlinien aus Punkten und gelben Kreisen
© Fraunhofer FOKUS

© Fraunhofer FOKUS

Understanding and Shaping Artificial Intelligence

❯❯❯

Artificial intelligence opens new possibilities for digital innovation but also presents organizations with challenges related to transparency, traceability, and reliable operation. Fraunhofer FOKUS researches and develops AI-based systems to make their functionality understandable and to ensure their deployment is technically and organizationally manageable. The focus is on integrating AI into complex IT environments, on robust architectures, and on secure operation under real-world conditions. This results in AI that does not operate in isolation but is responsibly and sustainably embedded in productive systems.

The following technologies and areas of expertise constitute key building blocks used in our fields of application across design, development, integration, and operation.

Get in contact

Technologies and Expertise for Research-Based Consulting and Development

Machine Learning & Deep Learning

ML and deep learning methods enable the analysis of large volumes of data and precise pattern recognition. 

  • Development of innovative ML models and deep learning architectures
  • Creation and operation of scalable ML pipelines (including MLOps and quality management)
  • Optimization, integration, and consulting for AI-powered systems
  • AI-based video and media analytics, QoE analysis, and pattern recognition in media streams

Generative AI (GenAI) & Language Models

Generative models and LLMs generate and edit text, images, and other media. 

  • Language Processing and Language Analysis
  • LLM-powered text and media generation
  • Integration of Retrieval-Augmented Generation (RAG) 
  • Consulting and Prototype Development
  • Generative media processing, AI-powered content analysis and generation

Explainable AI (XAI) & Trustworthy AI 

Explainable AI creates transparency, builds trust in complex models, and ensures content authenticity and provenance. Traceability methods support the responsible and compliant use of AI systems.

  • Development of interactive explanation tools
  • Promotion of transparent decision-making processes
  • Consulting on AI ethics and standards
  • Certification processes for trustworthy AI

Hybrid AI & Neurosymbolic Approaches

Combining symbolic methods with machine learning increases the flexibility and traceability of AI systems. This results in robust and explainable applications for challenging scenarios.

  • Development of neurosymbolic AI models
  • Integration of ontological knowledge into ML systems
  • Combining Rules and Data-Driven Learning

Data Management & Platform Technologies

Efficient data management forms the foundation for AI platforms and analytics environments. Interoperable solutions support data quality, integration, and the sustainable use of large datasets. 

  • Building data platforms, data pipelines, and AI ecosystems
  • Ensuring data quality, data protection, and integration into existing IT landscapes
  • Consulting on best practices in data management and the integration of AI solutions

Sensor Data Fusion & Real-Time Processing

Heterogeneous sensor data is combined and analyzed in real time. This enables more precise decision-making in mobility, security, and IoT applications.

  • Development of sensor data fusion systems with AI-powered pattern recognition
  • Real-time analysis of data streams using AI algorithms
  • IoT integration with AI-based detection and prediction systems
  • Real-time anomaly detection using AI models

Standardization, Regulation & Certification

Standards and certification processes ensure the quality and reliability of AI systems. They enable the widespread use of interoperable, secure, and compliant technologies.

  • Standardization Strategies and Standard Development
  • Consulting on Certification Processes
  • Interoperability and conformity assessment
  • Integration of automated CI/CD testing processes into certification scenarios

Human-in-the-Loop & Interactive Systems

Interactive AI systems involve humans in decision-making processes. This increases acceptance and control, especially in safety-critical or complex application areas.

  • Development of User-Centric AI Systems
  • Promoting Human-Machine Collaboration
  • Speech dialogue systems & real-time speech recognition
  • Security by Design for safety-critical interactions
  • Human-Centric Media & Learning Technologies

Data-Driven Decision Support

Data analysis with AI methods transforms large, heterogeneous datasets into actionable insights and improves accuracy and transparency for well-informed decision-making.

  • Analytics-as-a-Service solutions
  • Semantic data integration and knowledge graphs
  • Real-time data processing and smart data analytics
  • Development of domain-specific AI algorithms
  • Streaming analytics, QoE and media monitoring

Multi-Agent Systems (MAS) & Agent-Based AI

MAS distribute decision-making processes among cooperating, controllable AI agents, enabling adaptive solutions for dynamic environments.

  • Decomposition into coordinated sub-decisions with clear task, control, and escalation logic
  • Standardized agent interactions with defined conflict resolution, verification, and escalation (e.g., A2A)
  • Emergent overall behavior based on local rules, shared goals, and monitored safety requirements

Synthetic Data & AI Validation

Synthetic data and digital twins support the systematic evaluation and validation of AI-driven decisions in complex systems.

  • Generation of synthetic data for rare and extreme scenarios, as well as edge cases and error scenarios
  • Digital twins for emulating system behavior and malfunctions under varying conditions
  • Continuous evaluation of AI decisions based on defined robustness and performance criteria prior to deployment

Our Experts

Adrian Paschke

Contact Press / Media

Prof. Dr. rer. nat. Adrian Paschke

Head of Data Analytics and AI

Expert in Federated Learning, explainable & secure AI, Quantum Machine Learning

Phone +49 30 3463-7228

Jürgen Großmann

Contact Press / Media

Dr.-Ing. Jürgen Großmann

Head of Development of Critical Systems

Expert in AI Safety, Standardization and Automotive Open System Architecture

Phone +49 30 3463-7390

Christopher Krauß

Contact Press / Media

Dr.-Ing. Christopher Krauß

Media & Data Science Lead

Expert in AI for Media and Chatbots

Phone +49 30 3463-7236

Ilja Radusch

Contact Press / Media

Dr.-Ing. Ilja Radusch

Director Business Unit Smart Mobility

Expert in Digital Twins, AI Perception, Smart Mobility, and Indoor Navigation

Phone +49 30 3463-7474

Background Knowledge

Expand all Close all

  • AI learns from large amounts of structured or unstructured data, such as text, images, audio, or sensor data. Using machine learning and deep learning techniques, it recognizes patterns and can use them to make predictions or decisions. The quality of the results depends largely on the data used, the algorithms, and the continuous validation of the models.

  • AI can automate repetitive tasks, speed up processes, and support data-driven decision-making. Businesses benefit from more efficient workflows, higher productivity, and better customer service. Especially in areas such as marketing, sales, production, or customer service, AI creates measurable added value when used in a targeted and meaningful way.

  • AI can produce erroneous or biased results if the data or models are inadequate. Generative AI can also generate false information or reinforce biases. Therefore, results should always be reviewed, and data protection, IT security, and legal requirements must be taken into account. Human oversight remains indispensable for important decisions.

  • Responsible AI stands for the transparent, fair, and secure use of artificial intelligence. This includes data protection, traceability, the prevention of discrimination, and compliance with legal requirements. Especially in sensitive applications, humans should be able to monitor decisions and intervene when necessary (“Human in the Loop”).

  • Successful AI implementation begins with a clearly defined use case and a high-quality dataset. Equally important are data protection, IT security, compliance, and regular quality checks of the models. Companies should also establish clear guidelines for the use of AI and train their employees