IAAIRecursiveVLM
Looping Back to Move Forward: Recursive Transformers for Efficient and Flexible Large Multimodal Models
A recursive Transformer for large multimodal models with a Recursive Connector that fuses intermediate hidden states with modality-specific projections across steps, and a Monotonic Recursion Loss that makes quality improve with recursion depth.
Inside the method
[ LMM backbone ] × r via Recursive Connector (vision / language projections)
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Full-stack recurrence
- Depth control
- On-demand recursion depth
- KV / state strategy
- See paper
Reading note
Reports +3% over standard Transformers and +7% over vanilla recursive baselines, with few loops on constrained devices and more loops when compute allows.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{recursivevlm2026,
title = {Looping Back to Move Forward: Recursive Transformers for Efficient and Flexible Large Multimodal Models},
author = {Ruihan Xu and Yuting Gao and Lan Wang and Jianing Li and Weihao Chen and Qingpei Guo and Ming Yang and Shiliang Zhang},
year = {2026},
eprint = {2602.09080},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2602.09080}
}