AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

Efficient Training of Retrieval Models using Negative Cache

Erik Lindgren, Sashank J. Reddi, Ruiqi Guo, Sanjiv Kumar

NeurIPS 2021 · Architectures and efficiency › RNNs and normalization · approach: Dataset distillation and data-efficient training · 3 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
Use negative cache with 8k negatives instead of 256 sampled negativesDual encoder · MS MARCO devMRR@10: 0.273 -> 0.310Improved
Increase cache size from 8k to 2M negativesNegative Cache · MS MARCO devMRR@10: 0.310 -> 0.331Improved
Increase cache size from 8k to 512k negativesNegative Cache · TREC 2019NDCG@10: 0.600 -> 0.649Improved
Increase cache size from 512k to 2M negativesNegative Cache · TREC 2019NDCG@10: 0.649 -> 0.630Hurt
Increase sampled negatives from 16 to 256Sample Cross-Entropy · MS MARCO devMRR@10: 0.235 -> 0.273Improved
Negative cache (2M) instead of DPR BM25 hard negativesDual encoder · MS MARCO devMRR@10: 0.311 -> 0.331Improved
Negative cache (2M) instead of ANCE 8M-negative trainingDual encoder · MS MARCO devMRR@10: 0.330 -> 0.331No clear effect
Negative cache with 2M negatives instead of DPRDual encoder · Natural QuestionsRecall@20: 0.784 -> 0.784No clear effect
Negative cache with 2M negatives instead of DPRDual encoder · Natural QuestionsRecall@100: 0.854 -> 0.856Improved
ANCE (20M negatives) versus negative cache (2M)Dual encoder · Natural QuestionsRecall@20: 0.819 -> 0.784Hurt
Increase cached negatives from 8k to 512kNegative Cache · Runtimesteps/sec: 3.38 -> 3.32No clear effect
Use 2M cache with 512 updates instead of 512k cache with 256 updatesNegative Cache · Runtimesteps/sec: 3.32 -> 2.04Hurt

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