Workshop on Memory-Based Reasoning for Responsible GenAI
The 1st Workshop on Memory-Based Reasoning for Responsible GenAI is going to take place at the 34th International Conference on Case-Based Reasoning (ICCBR-26) on 15 August, 2026, Bremen, Germany, co-located with IJCAI/ECAI. This workshop will provide a gathering space for the CBR, memory-based learning, and LLM communities to discuss synergies, share current research, and explore potential collaborations on mutual interests and objectives.
Important Dates
Submission
Notification
Camera-Ready Copy & Author Registration
Workshop Date
Topics of Interest
Generative Artificial Intelligence (GenAI) is a groundbreaking technology changing our daily tasks and lives. Instrumental to this is the high performance that Large Language Models (LLMs) currently achieve. However, LLMs still have weaknesses; for example, hallucinations and lack of transparency limit their application and pose new challenges. Recent work has argued that augmenting LLMs with reasoning from episodic memory can alleviate some of these challenges. This workshop focuses on building more responsible GenAI through memory augmented approaches. In our context, responsibility means having properties such as explainability, trustworthiness, privacy, and contestability, bias mitigation, and safety. The workshop examines how integrating memory-based methods with LLMs can advance responsible GenAI through capabilities such as provenance, traceability, and how short- and long-term memory models can support privacy preserving and safer, more controllable behaviour. The workshop also welcomes submissions on specific integrations of memory-based methods with LLMs, and of memory-based methods for responsible AI.
Memory-based learning directly uses records of prior episodes to make decisions regarding a new problem to solve, aligning naturally with memory-augmented GenAI. An example of this type of learning is Case-Based Reasoning (CBR), a longstanding memory-based learning approach that uses experiential knowledge to solve new problems by retrieving and adapting records of similar prior episodes stored as cases.
As discussed in the paper “Case-Based Reasoning Meets Large Language Models: A Research Manifesto For Open Challenges and Research Directions” (https://hal.science/hal-05006761v1/file/main.pdf), LLMs and CBR can enhance each other. CBR can enhance LLM performance by providing a persistent, auditable external memory via principled retrieval and adaptation of prior cases. Conversely, LLMs can enhance CBR, for example through case acquisition and modeling from heterogeneous sources, by performing case adaptation, and by enabling fluent interactive interfaces for applications such as case-based decision-making systems. Integrations might solve shared and individual challenges and take advantage of opportunities for growth.
This workshop will provide a gathering space for the CBR, memory-based learning, and LLM communities to discuss synergies, share current research, and explore potential collaborations on mutual interests and objectives.
- Memory-based learning to improve LLM performance or output
- Bias and fairness in memory retrieval and adaptation
- Memory safety (guardrails on what can be stored, retrieved, or used, and safe forgetting)
- Role of CBR revision for Lifecycle governance (maintenance, deprecations, versioning, and compliance)
- LLMs to improve memory-based learning performance or output
- CBR for provenance and traceability for generated outputs, and provenance methods for CBR
- Provenance and traceability for generated outputs (case level citations, audit trails)
- Guided generation for LLMs using memory-based learning
- Retrieval Augmented Generation (RAG) & CBR
- Retrieval and similarity metrics to enhance GenAI
- CBR-based adaptation methods to enhance GenAI
- The role of knowledge representation methods (e.g., ontologies) for LLM fine-tuning and prompt engineering
- Improving accuracy and mitigating hallucination in LLMs using knowledge-based systems or expert systems
- Explainability & auditability in GenAI using memory-based learning
- Explainability & auditability using memory-based learning and enhanced through GenAI
- LLM-based knowledge acquisition using memory-based learning
- LLM-based maintainability using memory-based learning
- LLMs & memory-based learning as Explainable Artificial Intelligence (xAI) methods
- Personalization in memory-based learning XAI systems using LLMs
- Operational efficiency in GenAI using memory-based learning
- Evaluation of GenAI enhanced through memory-based learning
- Evaluation protocols and benchmarks specifically for memory augmented GenAI (task suites, ablations, leakage tests)
- Other Case-Based Reasoning (CBR) & GenAI synergies
Submission Procedure
Papers must be submitted in PDF format in the CEURART article style (http://ceur-ws.org/Vol-XXX/CEURART.zip).
We accept both long papers (10-14 pages + 2 pages of references) and short papers (5-7 pages + 2 pages of references). Each paper must include a statement describing any use of generative AI in the paper.
Long papers can be novel approaches to introduce to the community, presentations of relevant research results, or survey papers about the workshop topics. Short papers can be early-stage works, position papers, or demonstrations to show a specific tool to the community.
Submissions will be made through the ICCBR EasyChair site:
https://easychair.org/conferences/?conf=iccbr26
Multiple Submission Policy
Papers submitted to other conferences must state this fact as a footnote on page 1, and the organizing committee, must be notified by email. If a paper appears in another conference or journal, it must be withdrawn from the workshop.
Author Registration and Participation Policy
At least one author of each accepted paper must register for the workshop by the workshop deadline for camera-ready copy. Workshop registration is included with ICCBR registration, and ICCBR will also offer a special 1-day registration for the workshop only. To appear in the workshop proceedings, papers must be presented in person at the workshop by one of the authors. Video presentations are not permitted.
Organizing Committee
Ralph Bergmann, Trier University & German Research Center for Artificial Intelligence, Germany
Marta Caro-Martínez, Complutense University of Madrid, Spain (primary contact)
David Leake, Indiana University, USA
Nirmalie Wiratunga, Robert Gordon University, UK
Program Committee
David Aha, Naval Research Laboratory, USA
Kerstin Bach, Norwegian University of Science and Technology, Norway
Belén Díaz-Agudo, Universidad Complutense de Madrid, Spain
Viktor Eisenstadt, University of Hildesheim, Germany
Michael Floyd, Knexus Research, USA
Mark Keane, UCD Dublin, Ireland
Mirko Lenz, Trier University, Germany
Jean Lieber, LORIA – Inria Lorraine, France
Lukas Malburg, University of Trier / German Research Center for Artificial Intelligence (DFKI), Germany
Kyle Martin, Robert Gordon University, UK
David Ménager, Institute for the Study of Learning and Expertise, USA
Mirjam Minor, Goethe University Frankfurt, Germany
Hector Munoz-Avila, Lehigh University, USA
Juan Antonio Recio García, Complutense University of Madrid, Spain
Rosina Weber, Drexel University, USA
Anjana Wijekoon, BT Group, UK
Kaitlynne Wilkerson, Indiana University, USA