AI Paper Atlas

Research map / Learning theory and algorithms / Prediction and decision losses

LAVA: Data Valuation without Pre-Specified Learning Algorithms

Hoang Anh Just, Feiyang Kang, Tianhao Wang, Yi Zeng, Myeongseob Ko, Ming Jin, Ruoxi Jia

ICLR 2023 · Learning theory and algorithms › Prediction and decision losses · approach: Data attribution and recourse · 6 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 LAVA's class-wise Wasserstein data ranking to remove low-value pointsCIFAR-10 Trojan Square backdoor detection · CIFAR-10, Trojan Square backdoordetection rate: at least 80% of poisoned data in the first 20% of removed pointsImproved
Use LAVA's data ranking to remove low-value pointsCIFAR-10 feature-collision poisoning detection · CIFAR-10, feature-collision poisoningdetection rate: 80% detection rate by removing only 11k samplesImproved
Use LAVA's data ranking to remove low-value pointsCIFAR-10 noisy feature detection · CIFAR-10, noisy featuresdetection rate: all 12,500 noisy samples by inspecting less than 15,000 samplesImproved
Use LAVA's data ranking to remove low-value pointsCIFAR-10 mislabeled data detection · CIFAR-10, mislabeled datadetection rateImproved
Use 2K rather than 10K validation data (ablation)LAVA · CIFAR-10, mislabeled data, removal budget 25Kdetection rate: 80% detection rate at data removal budget of 25K; best baseline achieves such a performance with 5 times bigger validatiImproved
Use 500 validation examples (ablation)LAVA · CIFAR-10, mislabeled data detectiondetection rate: 500Improved
Use LAVA's data ranking to remove low-value pointsCIFAR-10 irrelevant data detection · CIFAR-10, irrelevant Deer images in Truck class, first 500 inspecteddetection rate: 94%Improved
Use LAVA's data ranking to remove low-value pointsCIFAR-10 irrelevant data detection · CIFAR-10, irrelevant Cat images in Dog class, first 500 inspecteddetection rate: 45% within the top 500Mixed
Use LAVA's data ranking to remove low-value pointsLAVA on GTSRB Hello Kitty blending attack · GTSRB, Blend backdoorpoisoned examples detected: 5000 points ... contain all poisoned data; ... KNN-SV ... around 11,000 samplesImproved
Use LAVA's data ranking to remove low-value pointsLAVA on MNIST noisy feature detection · MNIST, noisy featuresnoisy samples detected: all noisy data within first 14,000 samples, which is 5,000 less than the best baselineImproved

Later papers that compared with this method

2 later accepted papers report LAVA as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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