MemAgents Workshop at ICLR 2026

The workshop Memory for LLM-Based Agentic Systems (MemAgents), co-located with ICLR 2026 in Rio de Janeiro, focused on the role of memory in long-horizon AI agents.

As LLMs evolve into agentic systems, memory becomes crucial for continuity, adaptation, and reasoning over past experience. The workshop discussed how agents can store and use memory across tasks, connecting ideas from NLP, reinforcement learning, and neuroscience.

It was co-organized by MCML PI Hinrich Schütze and our Junior Members Yunpu Ma and Ercong Nie, together with international collaborators.

Keynote speakers included MCML PI Volker Tresp, Chelsea Finn, Jeff Clune, Mengye Ren, Aditi Raghunathan, Weiwen Liu, Fred Sala, and Jeff Pan, covering topics such as long-term memory, self-evolving agents, and data-centric approaches.

The event featured oral and poster sessions with over 110 submissions, highlighting strong interest in memory-centric AI research and its growing importance for future agent systems.