AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

CSRv2: Unlocking Ultra-Sparse Embeddings

Lixuan Guo, Yifei Wang, Tiansheng Wen, Yifan Wang, Aosong Feng, Bo Chen, Stefanie Jegelka, Chenyu You

ICLR 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Sparse training from scratch · 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
add k-annealing (ablation)CSR · MTEB (6 task types, avg), k=2, e5-Mistral-7Baverage score: 44.17 -> 45.33Improved
add supervised contrastive loss (ablation)CSR · MTEB (6 task types, avg), k=2, e5-Mistral-7Baverage score: 44.17 -> 49.05Improved
combine k-annealing and supervision (ablation)CSR · MTEB (6 task types, avg), k=2, e5-Mistral-7Baverage score: 49.05 -> 53.25Improved
add full backbone finetuning (ablation)CSRv2-linear · MTEB (6 task types, avg), k=2, e5-Mistral-7Baverage score: 53.25 -> 58.34Improved
add k-annealing (ablation)CSR · MTEB avg, k=64, e5-Mistral-7Baverage score: 66.68 -> 67.35Improved
add full finetuning (ablation)CSRv2-linear · MTEB avg, k=64, e5-Mistral-7Baverage score: 67.58 -> 68.08Improved
replace CSR with CSRv2CSR · MTEB avg, k=2, Qwen3-Embedding-4Baverage score: 36.29 -> 58.53Improved
replace CSR with CSRv2 at full dimensionCSR · MTEB avg, 4096 active dims, e5-Mistral-7Baverage score: 69.70 -> 69.80No clear effect
use CSRv2 instead of SPLADEv3 at same low budgetSPLADEv3 · MTEB Retrieval (8 datasets), k=2nDCG@10: 11.34 -> 37.48Improved
use CSRv2 instead of MRL at k=2MRL · MTEB avg, k=2, e5-Mistral-7Baverage score: 33.81 -> 58.38Improved
k-annealing reduces dead neurons (ablation)CSR · dead neuron ratio, k=2dead neuron ratio: 90% -> about 20% (with supervision)Improved
finetuning backbone effect on dead neurons (ablation)CSRv2-linear · dead neuron ratiodead neuron ratioNo 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