AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

CAMERA: Multi-Matrix Joint Compression for MoE Models via Micro-Expert Redundancy Analysis

Yuzhuang Xu, Xu Han, Yuanchi Zhang, Yixuan Wang, Yijun Liu, Shiyu Ji, Qingfu Zhu, Wanxiang Che

AAAI 2026 · Architectures and efficiency › Mixture of experts · approach: Efficient MoE language model systems · 4 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 CAMERA-P micro-expert pruning instead of D2-MoE/NAEEDeepseek-MoE-16B · 9 zero-shot tasks, 60% prunedavg accuracy: 46.72 (D2-MoE) -> 51.62Improved
use CAMERA-P instead of D2-MoE/NAEEQwen3-30B-A3B · 9 zero-shot tasks, 40% prunedavg accuracy: 65.77 (NAEE) -> 67.46Improved
use CAMERA-P instead of NAEE at low pruning ratioDeepseek-MoE-16B · 9 zero-shot tasks, 20% prunedavg accuracy: NAEE 60.51 vs CAMERA-P 61.03Mixed
use CAMERA-Q micro-expert mixed precision instead of GPTQDeepseek-MoE-16B · 2.25-bit quantization avg accuracyavg accuracy: 53.45 -> 56.56Improved
use CAMERA-Q instead of expert-level mixed precision MCQwen3-30B-A3B · 2.25-bit quantization avg accuracyavg accuracy: 52.79 -> 54.75Improved
slice matrices along micro-expert (row/column consistent) instead of column-wise (ablation)CAMERA-Q · 2.25-bit quantization Deepseek-MoE-16BWikitext2 perplexity: 11.28 -> 9.51Improved
slice matrices consistently per micro-expert (ablation)CAMERA-Q · 2.25-bit quantization Deepseek-MoE-16Bavg accuracy: 52.69 -> 56.56Improved
increase calibration set size from 8 to 256 sequences (ablation)CAMERA-P · Wikitext2 perplexity, 20% pruned Deepseek-MoE-16Bperplexity: 6.66 -> 6.56No clear effect
tune balance coefficient alpha (ablation)CAMERA-P · Deepseek-MoE-16B 20% prunedperplexity and avg accuracyNo clear effect
prune micro-experts instead of whole experts (decoding)CAMERA-P vs D2-MoE · Deepseek-MoE-16B MoE layer, 40% pruneddecoding speedup: 1.04x -> 1.42xImproved
apply CAMERA-P to dense model FFN (20%)openPangu-7B · 9 zero-shot tasks avgavg accuracy: 52.89 -> 51.00Hurt
add Wanda 20% unstructured pruning after CAMERA-P (ablation)openPangu-7B · 9 zero-shot tasks avgavg accuracy: 51.00 -> 51.11No clear effect

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