IAAIPAR
PAR: Progress-Aligned Recurrence for Length Generalization in State Tracking
Progress-Aligned Recurrence makes extra iterations useful for longer state-tracking problems by supervising the prediction at iteration k to match the state after k operations and anchoring recurrent state norms to the first iteration. A learned completion readout decides when to answer.
Inside the method
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Adaptive depth
- Depth control
- Iterations scale with problem length; learned stopping
- KV / state strategy
- See paper
Reading note
A Qwen2.5-0.5B model trained on at most 32 swap-S5 moves reaches 95.5% full-state accuracy at 512 moves. Removing norm anchoring or progress-aligned targets sharply reduces extrapolation. Results cover sequential state-tracking tasks.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{par2026,
title = {PAR: Progress-Aligned Recurrence for Length Generalization in State Tracking},
author = {Pan Wang and Yihao Hu and Guangyuan Dong and Yitong Lin and Hang Wang and Zimi Liu and Zhihao Wen and Jianshe Li and Jiang-Ming Yang and Wei Wu and Yongqi Tong},
year = {2026},
eprint = {2610.27628},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.27628}
}