Data-Efficient Agentic Learning
for Data Mining (DEAL-DM)

Background

Modern data mining increasingly demands intelligent systems that can operate under realistic constraints, including limited labeled data, costly feedback, noisy environments, and high-risk decision making. Agentic learning systems, which perceive, reason, plan, use tools, and improve from feedback, offer a promising pathway. However, many agentic approaches remain data-hungry, requiring extensive supervision or interaction to perform reliably.

This special session addresses the critical need for data-efficient agentic learning that maximizes the informational yield from every unit of data, feedback, or real-world interaction. We focus on techniques such as experience augmentation, memory and retrieval-augmented agents, budget-efficient reinforcement learning, in-context adaptation, and human-in-the-loop mining. By bringing together advances in agentic AI and data-efficient learning, this session aims to enable robust, deployable data mining systems in domains where data and interaction are scarce, expensive, or risky.

Aims and Scope

This special session, Data-Efficient Agentic Learning for Data Mining (DEAL-DM), provides a focused forum on data-efficient agentic learning for data mining. We interpret agentic data mining systems broadly as systems that can perceive data or environments, reason and plan, retrieve and reuse knowledge, interact with tools or users, and improve from feedback.

The session is especially concerned with realistic settings in which supervision and interaction are limited, noisy, delayed, costly, or high-risk. We welcome original research on methods, systems, evaluation, and applications that improve the information yield obtained per unit data, feedback, or real-world interaction.

Topics of Interest

This special session invites authors to submit original manuscripts that demonstrate and explore current advances in all related areas mentioned above. Topics of interest include, but are not limited to:

• Agentic data mining and autonomous data analysis workflows
• Experience augmentation for data-efficient learning, including data generation, experience transformation, and simulated interaction
• Memory, retrieval, knowledge reuse, and external knowledge integration for agentic data mining
• Planning, tool use, verification, reflection, modular agent design, and structured execution protocols
• In-context adaptation, prompt optimization, parameter-efficient fine-tuning, and test-time adaptation
• Budget-efficient reinforcement learning and learning from sparse, delayed, implicit, or preference-based feedback
• Human-in-the-loop, interactive, and feedback-driven data mining
• Personalization and recommender systems with limited user-specific data
• Agentic systems for scientific discovery, healthcare, eScience, and decision support
• Trustworthy evaluation, robustness, safety, privacy, and benchmarking of agentic data mining systems

Formatting Guidelines

We welcome original and unpublished research contributions that align with the themes of Data-Efficient Agentic Learning for Data Mining.

• Paper Format: All submissions must strictly follow the same specifications, requirements, and policies as the main track submissions in terms of paper length, formatting, and important policies. Please refer to the Submission Guidelines in the main track Call for Papers.
• Page Limit: The maximum length is 15 pages.
• Double-blind Review: All submissions will undergo a double-blind peer review process, ensuring that author identities and affiliations remain anonymous to the reviewers, authors submit blinded manuscripts without identifying information, and reviewers and authors avoid any de-anonymization attempts.

Submission Guidelines

Authors are invited to submit original research papers, case studies, and technical reports via the CMT online submission system. Each submission will undergo a rigorous peer-review process by at least three reviewers. Accepted papers will be included in the conference proceedings and published in the Springer LNCS series.

• Submission Site: https://cmt3.research.microsoft.com/ADMA2026
• When submitting your manuscript, please select the Special Session Track option and choose the area: Special Session: Data-Efficient Agentic Learning for Data Mining.

Important Dates (AoE Time)

Organizers and Committee

Technical Program Committee

CMT Acknowledgment

The Microsoft CMT service was used for managing the peer-reviewing process for this conference. This service was provided for free by Microsoft and they bore all expenses, including costs for Azure cloud services as well as for software development and support.