REVIEW 3 cited by
Probabilistic Stability Guarantees for Feature Attributions
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
read the original abstract
Stability guarantees have emerged as a principled way to evaluate feature attributions, but existing certification methods rely on heavily smoothed classifiers and often produce conservative guarantees. To address these limitations, we introduce soft stability and propose a simple, model-agnostic, sample-efficient stability certification algorithm (SCA) that yields non-trivial and interpretable guarantees for any attribution method. Moreover, we show that mild smoothing achieves a more favorable trade-off between accuracy and stability, avoiding the aggressive compromises made in prior certification methods. To explain this behavior, we use Boolean function analysis to derive a novel characterization of stability under smoothing. We evaluate SCA on vision and language tasks and demonstrate the effectiveness of soft stability in measuring the robustness of explanation methods.
Forward citations
Cited by 3 Pith papers
-
What makes an Ensemble (Un) Interpretable?
A complexity-theoretic analysis showing that the number, size, and type of base models determine whether ensemble explanations are tractable, with linear-model ensembles intractable even for two models.
-
Invisible Attributes, Visible Biases: Exploring Demographic Shortcuts in MRI-based Alzheimer's Disease Classification
Race and sex are readable from adult brain MRI with high accuracy, and demographic imbalance in training data leads to measurable shortcut learning and bias in MRI-based Alzheimer's disease classification.
-
Explaining, Fast and Slow: Abstraction and Refinement of Provable Explanations
Abstraction-refinement over neuron merging computes provably sufficient and minimal explanations of neural network predictions substantially faster than verifying on the full network.
Discussion (0). Sign in to comment.