AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Beyond Hamming: Query-Aware Decoding of Binary Cosine Sketches

DaeHun Nyang

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Learned embeddings for retrieval

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
decode with query-aware probit MLE instead of Hamming agreementSimHash · ArguAna HARD0.1 (K=100)Hit@K: 0.000 -> 0.454Improved
decode with query-aware probit MLE instead of Hamming agreementSimHash · SciFact HARD0.3 (K=100)Hit@K: 0.233 -> 0.467Improved
decode with query-aware probit MLE instead of Hamming agreementSimHash · FiQA HARD0.3 (K=200)Hit@K: 0.087 -> 0.357Improved
decode with query-aware probit MLE instead of Hamming agreementSimHash · ArguAna HARD0.1 (recall 0.8)candidates at fixed recall: 766 -> 385Improved
decode with query-aware probit MLE instead of Hamming agreementSimHash · FiQA HARD0.2 (recall 0.8)candidates at fixed recall: 5,831 -> 3,786Improved
replace search-time scorer in hnswlib with QA-COS (128 bits)SimHash scorer · Native hnswlib on BEIRRecall@10: .770 -> .853Improved
replace search-time scorer in hnswlib with QA-COS (64 bits)SimHash scorer · Native hnswlib on BEIRRecall@100: .258 -> .335Improved
replace search-time scorer in hnswlib with QA-COS (128 bits)SimHash scorer · Native hnswlib on BEIRvisited nodes: 1364.5 -> 1228.5Improved
use gated T=1 QA-COS decoder in storage-aware pipelineSimHash scorer · FiQA matched recallmean latency (ms): 1.175 -> .997Improved
use QA-COS instead of SimHash for angle estimationSimHash · 32 (d,m) settingsMAE: 22.3-28.9% reductionImproved
use Li's practical reduced-statistic estimator instead of SimHashSimHash · 32 (d,m) settingsMAE: 11.8-18.3% reductionImproved
use full query-aware likelihood (QA-COS) instead of Li's reduced statisticLi practical estimator · 32 (d,m) settingsMAE reduction: 11.8-18.3% -> 22.3-28.9%Improved

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