AI Paper Atlas

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

Evaluating Neuron Interpretation Methods of NLP Models

Yimin Fan, Fahim Dalvi, Nadir Durrani, Hassan Sajjad

NeurIPS 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Dense features and foundation priors · 2 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 Probeless instead of Random neuron selectionneuron ranking · POS BERTNeuronVote: 0.021 -> 0.532Improved
use Probeless instead of IoU (both corpus-based)neuron ranking · POS BERTNeuronVote: 0.365 -> 0.532Improved
use Probeless instead of LCAneuron ranking · POS BERT layer 12NeuronVote: 0.391 -> 0.515Improved
use Probeless instead of Lassoneuron ranking · POS XLM-R layer 12NeuronVote: 0.267 -> 0.497Improved
use Ridge instead of Lasso at last layer of RoBERTaneuron ranking · POS RoBERTa layer 12NeuronVote: 0.350 -> 0.510Improved
move from layer 1 or 6 to last layer 12LCA and Lasso on BERT · POSNeuronVote: 0.544 -> 0.391Hurt
use classifier accuracy as evaluation metric instead of compatibilityevaluation metric · POS BERT 30 neuronsaccuracy: 0.92 -> 0.95Hurt
use random neuron selection with more neurons (ablation)classifier accuracy metric · POS BERTaccuracy: 0.81 -> 0.92Improved
replace consensus of other methods with Lasso only when comparing Probeless vs LCA (ablation)Probeless · POS BERTNeuronVote: 0.449 vs 0.516Hurt
compare Probeless vs Lasso under all consensus subsets (ablation)Probeless · POS BERTNeuronVote: 0.382 vs 0.324Improved
compare Probeless vs IoU under all consensus subsets (ablation)Probeless · POS BERTNeuronVote: 0.426 vs 0.221Improved
use Gaussian instead of other methodsneuron ranking · POS BERTNeuronVote: 0.169 vs 0.532Hurt

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