AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Softmax Output Approximation for Activation Memory-Efficient Training of Attention-based Networks

Changhyeon Lee, Seulki Lee

NeurIPS 2023 · Architectures and efficiency › Attention and transformers · approach: Sparse and structured attention · 1 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
store only m=1 highest and lowest softmax outputs (11.9% memory) (ablation)softmax approximation · AG News XLNetf1-micro: 0.9323 -> 0.9175Hurt
increase stored elements to m=10 (20.9% memory) (ablation)softmax approximation · AG News XLNetf1-micro: 0.9323 -> 0.9335No clear effect
store m=30 elements (ablation)softmax approximation · AG News XLNetf1-micro: 0.9323 -> 0.9371Improved
m=1 on DBpedia (ablation)softmax approximation · DBpedia XLNetf1-micro: 0.9889 -> 0.9743Hurt
m=5 on DBpedia (15.9% memory) (ablation)softmax approximation · DBpedia XLNetf1-micro: 0.9889 -> 0.9869No clear effect
m=1 on SST-2 (11.9% memory) (ablation)softmax approximation · SST-2 ALBERTf1-micro: 0.8326 -> 0.8268No clear effect
m=40 on SST-2 (ablation)softmax approximation · SST-2 ALBERTf1-micro: 0.8326 -> 0.8360No clear effect
m=1 on IMDb (11.9% memory) (ablation)softmax approximation · IMDb XLNetf1-micro: 0.8769 -> 0.8833Improved
m=1 (2.5% memory) on Multi30k De-En (ablation)softmax approximation · Multi30k De-EnBLEU: 36.52 -> 35.53Hurt
m=2 (14.4% memory) on De-En (ablation)softmax approximation · Multi30k De-EnBLEU: 36.52 -> 36.29No clear effect
m=4 or more (19.4% memory) on De-En (ablation)softmax approximation · Multi30k De-EnBLEU: 36.52 -> 36.72Improved
m=1 on Multi30k En-De (ablation)softmax approximation · Multi30k En-DeBLEU: 34.05 -> 32.76Hurt

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