Main Track
Maximum limit of 7 pages, excluding references and appendix.
Machine Learning/CORE Unranked
SCALE@ICML2026 2026
ICML 2026 Workshop on Multimodal AI Agents
SCALE@ICML2026 is a workshop focused on scalable learning and optimization for efficient multimodal AI agents, addressing challenges in agentic memory, efficient system design, test-time scaling, planning, and evaluation. It brings together researchers to advance the capabilities of multimodal agents in real-world applications such as robotics, software automation, and web interaction, with an emphasis on memory, efficiency, and robustness under constraints.
Key deadlines
Verified September 9, 2026Abstract registration
April 29, 2026
AoE
Full paper
April 29, 2026
AoE
Workshop timeline
Submission and decisions
Abstract registrationKey deadline
April 29, 2026 · AoE
Full paperKey deadline
April 29, 2026 · AoE
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Benchmarking and dataset track
Maximum limit of 7 pages, excluding references and appendix.
Late-breaking track
Maximum limit of 3 pages, excluding references and appendix.
Research areas in scope
01
Agentic memory
Short-, long-term, and hierarchical multimodal memory designsMemory consolidation, retrieval, and forgetting for long-horizon tasksMemory-grounded reasoning and cross-task knowledge reuseTemporal, causal, and compositional memory abstractionsRobustness to noisy or missing multimodal inputs
02
Efficient agentic AI systems
Compute- and memory-efficient agent architecturesHardware-aware co-design for heterogeneous platformsSafety, verification, and controllability under constraintsCompression, distillation, caching, and adaptive executionModular pipelines with dynamic compute allocation
03
Scaling of multi-modal agents
Test-time scaling laws for reasoning and planningMixture-of-agents and expert routing strategiesMulti-agent coordination and parallel reasoningGeneralization and robustness at scaleTrade-offs among model size, agent count, and compute
04
Multi-modal agents for planning
Perception-memory-action integrationLong-horizon embodied planning and controlWorld models and simulation-based planningNeuro-symbolic and interpretable planning methodsPlanning under uncertainty and partial observability
05
Evaluation and benchmarking
Long-context reasoning and memory benchmarksMulti-step reasoning and tool-use evaluationRobustness to noise and distribution shiftEfficiency metrics: latency, energy, memoryReal-world, deployment-focused evaluation protocols
Policies worth checking twice
- All submissions must be in PDF format using the ICML 2026 LaTeX style file.
- The reviewing process will be double blind. Anonymize your submission.
- Papers under review elsewhere are allowed.
- The workshop is non-archival. Submissions can be submitted to other venues.
- All submissions must include a section describing the nature and extent of LLM/agent usage in their research and ideation. This section does not count toward the page limit.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026