AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

LatentLLM: Activation-Aware Transform to Multi-Head Latent Attention

Toshiaki Koike-Akino, Xiangyu Chen, Jing Liu, Ye Wang, Pu (Perry) Wang, Matthew Brand

AAAI 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 0 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
use root-covariance pre-conditioner instead of identity/Hessian/L1/L2/covariance in activation-aware SVD (ablation)ASVD · WikiText2, 20% compression, OPT-1.3Bperplexity: 28.7 -> 21.9Improved
replace local per-layer SVD with joint attention-aware SVD (LatentLLM) (ablation)ASVD (RootCov) · WikiText2, 20% compression, OPT-1.3Bperplexity: 21.9 -> 15.8Improved
joint SVD vs RootCov (ablation)ASVD (RootCov) · WikiText2, 20% compression, Qwen3-8Bperplexity: 26.0 -> 14.2Improved
joint SVD vs RootCov at 40% compression (ablation)ASVD (RootCov) · WikiText2, 40% compression, Qwen3-8Bperplexity: 119.2 -> 53.9Improved
joint SVD vs RootCov (ablation)ASVD (RootCov) · ScienceQA avg, 10% compression, LLaVA-7Baccuracy: 60.67 -> 65.76Improved
joint SVD vs RootCov (ablation)ASVD (RootCov) · ScienceQA avg, 50% compression, LLaVA-7Baccuracy: 49.30 -> 52.25Improved
joint SVD vs RootCov (ablation)ASVD (RootCov) · TextVQA, 30% compression, Qwen2.5-VL-7Baccuracy: 66.31 -> 73.90Improved
use covariance (not root) pre-conditioner (ablation)ASVD · ScienceQA avg, 10% compression, LLaVA-7Baccuracy: 60.67 -> 37.42Hurt
use block-identity junction matrixSVD low-rank layer · parameter countparameters: r(d'+d) -> r(d'+d) - r^2Improved
compress model with LatentLLMOPT-6.7B · speed and KV cacheKV cache bytes: 5.37G -> 0.57G (0% -> 80% compression)Improved

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