AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

MOAI: Module-Optimizing Architecture for Non-Interactive Secure Transformer Inference

Linru Zhang, Xiangning Wang, Jun Jie Sim, Zhicong Huang, Jiahao Zhong, HUAXIONG WANG, Pu Duan, Kwok-Yan Lam

ICLR 2026 · Architectures and efficiency › RNNs and normalization · approach: Distributed LLM inference systems · 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 consistent column/diagonal packing with rotation-free Softmax/LayerNorm and interleaved batching (MOAI) instead of THORTHOR · BERT-base on A100total time (s): 602.26 -> 283.95 (52.8% reduction)Improved
apply MOAI evaluation flow to Powerformer's FHE-friendly modelPowerformer · BERT-base on A100total time (s): 344.52 -> 152.4 (55.7% reduction)Improved
use rotation-free Softmax and LayerNorm (ablation)MOAI vs NEXUS/THOR · BERT-baseHE rotations: 2448 -> 0Improved
add column packing and interleaved batching for matrix multiplicationTHOR · BERT-baseHE rotations in matmuls: 22224 -> 9648Improved
MOAI vs Powerformer matmul rotationsPowerformer · BERT-baseHE rotations in matmuls: 16740 -> 9648Improved
run MOAI under FHE encryption instead of plaintextBERT-base · SST-2accuracy: 93.9 -> 93.8No clear effect
run MOAI under FHE encryption instead of plaintextBERT-base · QNLIaccuracy: 90.7 -> 90.5No clear effect
run MOAI under FHE encryption instead of plaintextBERT-base · RTEaccuracy: 65.7 -> 65.5No clear effect
bootstrap ctsum after the sum step in Softmax to lower the Softmax depth (ablation)MOAI Softmax · BERT-baseSoftmax depth: 20 -> 10 layersImproved
keep Powerformer pooler/classification (not optimized by MOAI)MOAI on Powerformer · BERT-base on A100Pooler & Classification time (s): 0.2 -> 0.7Hurt

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