IAAILAST
LAST: Looped Audio Spectrogram Transformer
The Looped Audio Spectrogram Transformer processes all tokens once and then reuses the same blocks to refine only the class token over fixed audio features, making later passes cheap.
Inside the method
[ Blocks ] on all tokens → [ Same blocks ] × passes on class token only
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Core recurrence
- Depth control
- Fixed number of passes
- KV / state strategy
- Audio features fixed after the first pass
Reading note
On AudioSet, ten-pass LAST reaches 0.345 mAP, 2.1% relative above a twelve-layer transformer, with 49.4% fewer parameters, 42% fewer MACs and 9.8% higher throughput. Going from two to ten passes adds only 1.2% computation.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{last2026,
title = {LAST: Looped Audio Spectrogram Transformer},
author = {Haider Al-Tahan and Sean O'Brien and Anastasia Razdaibiedina and N. Apurva Ratan Murty},
year = {2026},
eprint = {2610.01926},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.01926}
}