AI Paper Atlas

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

An Importance Weighted Feature Selection Stability Measure

Victor Hamer, Pierre Dupont

JMLR 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Feature attribution methods · 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
weight feature selection stability by importance (phi_iw instead of subset phi)stability measure · toy map of Figure 4stability value: 0.72 -> 0.21Improved
use phi_iw instead of phi_pears when half the importance is stablestability measure · toy map of Figure 6stability value: 0.25 -> 0.5Improved
optimize phi_iw instead of phi for choosing the accuracy/stability compromiseHybrid-RFE · singhaccuracy: 0.889 -> 0.934Improved
optimize phi_iw instead of phi_pears for the chosen compromiseHybrid-RFE · singhaccuracy: 0.930 -> 0.934Improved
increase correlated group size c (phi, group lasso)phi · group lasso with correlated groupsstability: phi increases when c growsHurt
increase correlated group size c (phi_pears)phi_pears · group lasso with correlated groupsstability: phi_pears starts higher and decreasesHurt
use phi_iw under growing group size cphi_iw · group lasso with correlated groupsstability: phi_iw remains approximately constant with cImproved
force selection of 5 highest-variance features with small differential shrinkage alpha=0.1Hybrid-RFE · singhfeature stability mapImproved
decrease lambda2 from 1 (lasso) toward elastic netlogistic regression · micro-array data setsstability: all three stability measures increase when departing from lasso (lambda2 = 1)Improved
increase final regularization lambda_flogistic RFE · lsvtphi_iw and phi_pears: phi unchangedImproved
increase final regularization lambda_f stronglylogistic RFE · lsvtaccuracy: accuracy first increases then dropsMixed
grow forest size from 3 trees to 3000 treesrandom forest selection · all data setsstability and accuracy: converges for T ~ 1000Improved

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