IAAIPonderNet
PonderNet: Learning to Ponder
Learns end-to-end how many computational steps to take, modeling halting as a probabilistic process and trading off prediction accuracy, computational cost and generalization.
Inside the method
[ Step function ] × n, halt with learned probability λₙ
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Adaptive depth
- Depth control
- Learned probabilistic halting
- KV / state strategy
- Not applicable
Reading note
Improves over earlier adaptive computation methods on a synthetic task, succeeds at extrapolation tests, and matches state of the art on a QA dataset with less compute.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{pondernet2021,
title = {PonderNet: Learning to Ponder},
author = {Andrea Banino and Jan Balaguer and Charles Blundell},
year = {2021},
eprint = {2107.05407},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2107.05407}
}