analiza aiqbench
LAST introduces a novel architecture that reuses transformer blocks in loops to refine class tokens, significantly reducing parameter count and computational cost compared to deeper sequential models. This approach achieves superior accuracy on the AudioSet benchmark while maintaining high throughput, demonstrating a practical path toward more efficient audio recognition systems. The method also shows enhanced robustness against temporal masking and various auditory augmentations.
LAST: Looped Audio Spectrogram Transformer
arXiv:2610.01926v1 Announce Type: cross Abstract: Increasing depth of transformer models improves recognition, but it comes at a substantial cost. Each additional layer requires more parameters, which makes the process computationally inefficient. We ask whether additional processing can focus on integrating features already computed. Looped Audio Spectrogram Transformer (LAST) first processes all tokens, then reuses the same blocks to refine only the class token over fixed audio features, thereby making later passes inexpensive. On AudioSet, ten-pass LAST achieves 0.345 mean average precision, exceeding a twelve-layer sequential transformer by 2.1% relative with 49.4% fewer parameters, 42% fewer multiply-accumulate operations, and 9.8% higher measured throughput. Across separately trained models, increasing the pass count from two to ten improves accuracy while adding only 1.2% computation. Further evaluations show improved robustness to temporal masking and various other auditory augmentations, with better generalization on classification tasks with music, environmental, and event sounds.
Abstract:Increasing depth of transformer models improves recognition, but it comes at a substantial cost. Each additional layer requires more parameters, which makes the process computationally inefficient. We ask whether additional processing can focus on integrating features already computed. Looped Audio Spectrogram Transformer (LAST) first processes all tokens, then reuses the same blocks to refine only the class token over fixed audio features, thereby making later passes inexpensive. On AudioSet, ten-pass LAST achieves 0.345 mean average precision, exceeding a twelve-layer sequential transformer by 2.1% relative with 49.4% fewer parameters, 42% fewer multiply-accumulate operations, and 9.8% higher measured throughput. Across separately trained models, increasing the pass count from two to ten improves accuracy while adding only 1.2% computation. Further evaluations show improved robustness to temporal masking and various other auditory augmentations, with better generalization on classification tasks with music, environmental, and event sounds.
Sound (cs.SD); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
I.2.6; I.5.4; H.5.5
arXiv:2610.01926 [cs.SD]
(or arXiv:2610.01926v1 [cs.SD] for this version)
https://doi.org/10.48550/arXiv.2610.01926 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 245 cuvinte
Subiecte conexe
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · 25 sept. 2026
[AINews] The Future of Latent Space
Latent Space · 25 sept. 2026
[AINews] Opus 5.5 is good at explainer videos
Latent Space · 29 sept. 2026
Build a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod
AWS · 4 sept. 2026