AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Scalable Neural Data Server: A Data Recommender for Transfer Learning

Tianshi Cao, Sasha Doubov, David Acuna, Sanja Fidler

NeurIPS 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Dense features and foundation priors · 3 citations (OpenAlex)

Read the paper

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 SNDS-selected data instead of random sample, 2% budgetSNDS-50 · 9 fine-grained targets averageaccuracy: 55.75 -> 59.07Improved
use SNDS-selected data instead of random sample, 5% budgetSNDS-50 · 9 fine-grained targets averageaccuracy: 63.95 -> 64.96Improved
use SNDS-selected data instead of random sample, 10% budgetSNDS-50 · 9 fine-grained targets averageaccuracy: 67.10 -> 68.33Improved
SNDS (indexing on public data) versus NDS (experts on each source), 2% budgetSNDS-50 · 9 targets averageaccuracy: 59.53 -> 59.07Mixed
SNDS versus NDS at 10% budgetSNDS-50 · 9 targets averageaccuracy: 68.33 -> 68.33No clear effect
replace stochastic sampling with greedy selection, 2% budget (ablation)SNDS · 9 targets averageaccuracy: 59.07 -> 57.55Hurt
replace stochastic sampling with greedy selection, 5% budget (ablation)SNDS · 9 targets averageaccuracy: 64.96 -> 64.45Hurt
greedy selection lowers pretraining task accuracy, 2% samples (ablation)SNDS · 601-way OpenImages pretrainingaccuracy: 29.44 -> 25.02Hurt
greedy selection lowers pretraining task accuracy, 5% samples (ablation)SNDS · 601-way OpenImages pretrainingaccuracy: 34.40 -> 31.61Hurt
use 10 experts instead of 50, 2% budget (ablation)SNDS · 9 targets averageaccuracy: 59.07 -> 58.32Hurt
use 10 experts instead of 50, 5% budget (ablation)SNDS · 9 targets averageaccuracy: 64.96 -> 64.00Hurt
drop public clusters overlapping target classes (Pets, CUB200) (ablation)SNDS · Oxford Pets, CUB200accuracyNo clear effect

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