AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Sparse Modular Activation for Efficient Sequence Modeling

Liliang Ren, Yang Liu, Shuohang Wang, Yichong Xu, Chenguang Zhu, ChengXiang Zhai

NeurIPS 2023 · Architectures and efficiency › Attention and transformers · approach: State space and recurrent sequence models · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Add sparse GAU activation to SSM sequence modelingSeqBoat versus S5 · LRAaverage accuracy: 87.46 -> 87.62Improved
Add sparse GAU activation to SSM sequence modelingSeqBoat versus MEGA-chunk · LRAaverage accuracy: 85.66 -> 87.62Improved
Add sparse GAU activation to SSM sequence modelingSeqBoat versus S4 · LRAtraining speedup: 4.8$\times$ -> 10.4$\times$Improved
Add sparse GAU activation to SSM sequence modelingSeqBoat versus S4 · Speech Commands SC10accuracy: 97.50 -> 97.35Hurt
Add sparse GAU activation to SSM sequence modelingSeqBoat versus MEGA-chunk · Speech Commands SC10training speedup: 1.00$\times$ -> 1.32$\times$Improved
Add sparse GAU activation to SSM sequence modelingSeqBoat versus MEGA-chunk · enwik8bits per character: 1.02 -> 1.02No clear effect
Add working memory with local attention to the compressed sequenceSeqBoat versus SeqBoat-full · LRA Pathfinderaccuracy: 93.00 -> 96.35Improved
Always activate GAU instead of using SMA (ablation)SeqBoat-full · LRA Imagevalidation accuracy: 90.16 -> 90.70Improved
Always activate GAU instead of using SMA (ablation)SeqBoat-full · LRA ListOpstraining speedup: 1.79$\times$ -> 1.11$\times$Hurt
Replace Tempered Softmax with Gumbel Softmax (ablation)SeqBoat-full · LRA Imagevalidation accuracy: 90.16 -> 89.94Hurt
Replace Tempered Softmax with Gumbel Softmax (ablation)SeqBoat-full · LRA ListOpsvalidation accuracy: 60.95 -> 59.75Hurt
Replace MD-EMA kernel with S4D (ablation)SeqBoat-full · LRA Imagevalidation accuracy: 90.16 -> 82.84Hurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy