AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Approximate Search with Quantized Sparse Representations

Himalaya Jain, Patrick Pérez, Rémi Gribonval, Joaquin Zepeda, Hervé Jégou

ECCV 2016 · Architectures and efficiency › Quantization · approach: Post-training weight quantization · 20 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 quantized scalar coefficients to residual quantization (Qα-RVQ vs RVQ)RVQ · SIFT1M 64 bits (un-normalized)squared reconstruction error: 22170 -> 22053Improved
add quantized scalar coefficients to residual quantization (Qα-RVQ vs RVQ)RVQ · GIST1M 64 bits (un-normalized)squared reconstruction error: 0.6986 -> 0.6537Improved
add quantized coefficients to PQ (Qα-PQ vs PQ) at equal code sizePQ · SIFT1M 64 bits (un-normalized)squared reconstruction error: 23515 -> 25859Hurt
add quantized coefficients to PQ (Qα-PQ vs PQ) at equal code sizePQ · GIST1M 64 bits (un-normalized)squared reconstruction error: 0.7121 -> 0.7224Hurt
use OPQ instead of PQ inside Qα-PQ (Qα-OPQ) (ablation)Qα-PQ · SIFT1M 64 bits (Euclidean aNN)recall R@1: 0.204 -> 0.227Improved
use OPQ instead of PQ inside Qα-PQ (Qα-OPQ) (ablation)Qα-PQ · SIFT1M 64 bits (Euclidean aNN)recall R@100: 0.900 -> 0.927Improved
apply OPQ to plain PQ (ablation)PQ · SIFT1M 64 bits (Euclidean aNN)recall R@1: 0.228 -> 0.243Improved
Qα-RVQ instead of RVQ with IMI T=100KRVQ · BIGANN 1B IMI T=100Krecall R@1: 0.181 -> 0.227Improved
Qα-PQ instead of PQ-64 with IMI T=100KPQ-64 · BIGANN 1B IMI T=100Krecall R@1: 0.170 -> 0.200Improved
Qα-PQ instead of PQ for encode time (half the codewords, inner products)PQ · SIFT1M 64 bits relative encoding timerelative encode time: 1 -> 0.206Improved
Qα-PQ instead of PQ for search timePQ · SIFT1M 64 bits relative search timerelative search time: 1 -> 1.867Hurt
add candidate pruning W'c=128 to Qα-RVQ in IVF (ablation)Qα-RVQ · BIGANN 1B IVFsearch time and recall: 10% faster than PQ with no performance lossImproved

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