AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Simple strategies for recovering inner products from coarsely quantized random projections

Ping Li, Martin Slawski

NeurIPS 2017 · Architectures and efficiency › Quantization · approach: Post-training weight quantization · 8 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
normalize projected vectors (normalized estimator instead of linear estimator)full-precision random projection cosine estimation · rho=0.99asymptotic MSE ratio: 5000Improved
normalize projected vectors at low similarityfull-precision random projection cosine estimation · rho=0.5asymptotic MSE ratio: 2.2Improved
use MLE instead of normalized estimatorfull-precision cosine estimation · generalMSE: at most by a factor of 2Improved
raise bit depth from 3 to 4 in linear code-based estimatorlinear estimator · rho=1 squared bias boundsquared bias bound: 7.2 · 10−3 -> 4.5 · 10−4Improved
raise bit depth from 2 to 3linear estimator · rho=1 squared bias boundsquared bias bound: 1.2 · 10−1 -> 7.2 · 10−3Improved
quantize to 4 bits instead of full precision with k=1000linear estimator · MSE vs full precisionMSE: at most a factor of 1.45Hurt
add normalization to quantized linear estimator (ablation)quantized code-based estimator · bias reductionsquared biasImproved
collision-based (b=2, doubled k) instead of normalized (b=4)quantized cosine estimation at equal bits · high similarity rhoMSE: crossover rho from about 0.55 to 0.95Mixed
increase b for collision-based estimatorcollision-based estimator · MSE vs rhoMSEMixed
increase b for code-based estimatorsnormalized estimator · MSE over whole rho rangeMSEImproved
increase k beyond order of squared biasquantized linear estimator · MSE, large rhoMSENo clear effect
use normalized b=4 instead of collision b=2 with doubled kSVM on arcene, Dexter, farm, PEMS · UCI classificationtest accuracyImproved

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