AI Paper Atlas

Research map / Architectures and efficiency / Quantization

LSQ++: Lower running time and higher recall in multi-codebook quantization

Julieta Martinez, Shobhit Zakhmi, Holger H. Hoos, James J. Little

ECCV 2018 · Architectures and efficiency › Quantization · approach: Ultra-low-bit LLM quantization · 37 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
replace conjugate gradient codebook update with direct regularized Cholesky solution (ablation)LSQ (Julia, C++) · SIFT1M 64 bitstime per iteration: 14.2 s -> 5.6 sImproved
replace conjugate gradient codebook update with direct regularized Cholesky solution (ablation)LSQ (Julia, CUDA) · SIFT1M 128 bitstime per iteration: 20.3 s -> 4.3 sImproved
replace conjugate gradient codebook update with direct regularized Cholesky solution (ablation)LSQ (Julia, CUDA) · Deep1M 64 bitstime per iteration: 3.3 s -> 1.0 sImproved
add stochastic relaxation SR-D (noise in codebooks) to LSQLSQ · SIFT1M 64 bitsRecall@1: 0.340 -> 0.346Improved
add SR-C (noise added to data) to LSQ at 128 bitsLSQ · MNIST, LabelMe22K, SIFT1M 128 bitsRecall@1Hurt
add SR-C to LSQ at 64 bitsLSQ · 64-bit datasetsRecall@1Mixed
double LSQ++ iterations from 25 to 50 (ablation)LSQ++ · SIFT1M 64 bitsRecall@1: 0.346 -> 0.348Improved
increase LSQ++ to 100 iterations with 128 encoding iterations (ablation)LSQ++ · SIFT1M 64 bitsRecall@1: 0.348 -> 0.351Improved
double base encoding ILS iterations from 128 to 256 (ablation)LSQ++ · SIFT1M 64 bitsRecall@1: 0.351 -> 0.353Improved
train CQ on the learn set instead of the base set (ablation)CQ · SIFT1M 64 bitsRecall@1: 0.290 -> 0.162Hurt
use norm codebook so RVQ and ERVQ account for query timeRVQ, ERVQ · MNIST, LabelMe22K, SIFT1MRecall@1Hurt

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