Skip to main content Skip to main navigation

Experience‑Based Learning Systems

Experience‑Based Learning Systems (EBLS) are AI systems that learn from experience, retain it, and use it to solve new problems. They are inspired by human cognitive processes and couple memory and learning with adaptive problem solving. As hybrid AI approaches, they combine symbolic techniques with subsymbolic methods and generative AI. The focus is on:

  • Case‑based reasoning (CBR) as the methodological foundation for problem solving by remembering and adapting cases
  • Knowledge‑based AI using semantic technologies, ontologies, and planning methods to process and exploit explicit knowledge
  • Data‑driven methods such as machine learning and deep learning to analyze existing datasets
  • Generative AI and foundation models for analysis, synthesis, and explanation

EBLS act both as autonomous systems and as components of human‑AI teams. They analyze processes, support decisions, adaptively control workflows, and improve through continuous learning from feedback. We study these systems in intelligent process management and in knowledge and experience management in complex application domains.

Research Topics:

Predictive Process Analytics and Optimization

We develop AI methods that predict, explain, optimize, and control complex workflows using process mining, deep learning, and reinforcement learning. The goals include data‑ and knowledge‑driven event prediction, retrieval of semantically similar process trajectories, and resource‑efficient control and decision support. Typical effects are shorter throughput times, improved on‑time performance, lower energy consumption, and deeper process understanding.

Topic Lead:  Dr. Joscha Grüger

Experience‑Based Agentic Reasoning

We investigate agentic AI systems that plan and steer their own hybrid reasoning processes. They orchestrate symbolic and subsymbolic reasoning agents and adapt their strategies based on experience to enable adaptive and explainable problem solving. The reasoning agents take on tasks such as knowledge and experience retrieval, solution adaptation and reuse, search and generative solution construction, validation, and interaction with the environment.

Topic Lead: Dr. Lukas Malburg

Generative AI for Knowledge‑Intensive Human‑Centered Processes

We support knowledge‑intensive, human‑centered processes through flexible, goal‑oriented process control and context‑sensitive assistance for complex process steps. The aim is to optimally combine the specific strengths of human and artificial actors and to learn from experience as a team in order to achieve better results together. Knowledge graphs and semantic search over heterogeneous knowledge bases, in combination with generative AI and innovative RAG (retrieval‑augmented generation) architectures, provide the foundation.

Topic Lead: Dr. Eric Rietzke

Experience‑Based Design with Explainable, User‑Centered AI

We research hybrid AI methods to support design processes based on case‑based reasoning and knowledge‑based representations. The goal is the systematic reuse and adaptation of experiential knowledge while accounting for functional, contextual, and user‑related aspects. Explainable and transparent decision mechanisms derive new solution ideas in a traceable manner and integrate them interactively into development processes.

Topic Lead:  Dr. Lisa Grewenig

Software Frameworks and Application Areas

For prototypical implementations of EBLS we develop modular software frameworks such as ProCAKE and CBRkit, which serve as the technological basis for demonstrating and testing experience‑based approaches.

Our application‑oriented research is conducted in close collaboration with partners from industry and society. EBLS are applied, among others, in the following areas:

Industrial Production

EBLS support planning and decision making in cyber‑physical systems, from process design and real‑time adaptation of production flows through flexible execution to predictive maintenance. We employ experience‑based agentic reasoning, symbolic AI planning, case‑based planning, large language models (LLMs), and reinforcement learning to realize transparent and safe process control.

Water and Energy Management

For resource‑ and energy‑efficient set‑point control we combine robust AI models for forecasting consumption, load, and feed‑in with experience‑based optimization. Digital twins provide the data basis on which the models are continually retrained and adapted in 24/7 operation via continuous learning. EBLS detect anomalies (e.g., leaks), recommend countermeasures, and make decisions with a high degree of automation. They adapt dynamically to changing operating and boundary conditions (e.g., demand, energy prices, weather) to balance cost, CO₂ footprint, and security of supply.

Public Administration & Skilled Trades

EBLS enable intelligent, flexible workflows together with semantic search across heterogeneous document and knowledge bases. Intelligent assistant systems accelerate case handling, increase quality, make decisions traceable, and provide efficient chat‑based interfaces to customers.

Recipe Design

Using case‑based and generative design, recipes including ingredients and processing steps are created or adapted to meet requirements. Existing experiential knowledge is leveraged and combined with deep domain knowledge to develop innovative yet valid recipes.

Emergency and Crisis Management

EBLS semantically integrate heterogeneous situational information from sensors, communication systems, and historical deployment data into a consistent common operating picture. They support control centers in time‑critical situations through explainable action recommendations, forecasts of resource needs, and coordination among responders in dynamic situations.

Healthcare

EBLS analyze treatment processes using process mining, find semantically similar patient cases, and predict next steps for personalized decision support. Guideline knowledge, process data, and explainable models are combined to foster transparency and trust in decisions.

Interested in a collaboration?

Your contact person for a joint AI project:

Dr. Eric Rietzke

Contact

Silke Kruft
Phone: +49 651 201 3875

Deutsches Forschungszentrum für
Künstliche Intelligenz GmbH (DFKI)
DFKI Branch Trier
Experience-based Learning Systems
Behringstraße 21
54296 Trier
Germany