Empirical Characterization of Rationale Stability Under Controlled Perturbations for Explainable Pattern Recognition
Abu Noman Md Sakib, Zhensen Wang, Merjulah Roby, Zijie Zhang · 2026
Read Empirical Characterization of Rationale Stability Under Controlled Perturbations for Explainable Pattern Recognition by Abu Noman Md Sakib, Zhensen Wang, Merjulah Roby, Zijie Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-04456.txt.
Description
Sakib and colleagues propose a metric for rationale stability in explainable pattern recognition: whether model reasons stay consistent across same-class samples and input shifts. The team builds the metric with a BERT model on the SST-2 dataset and toughness tests on RoBERTa, DistilBERT and IMDB, using SHAP scores. The metric uses cosine similarity of SHAP values for same-label inputs, flagging biased feature reliance or mixed reasoning for similar predictions. Tests against standard fidelity metrics show it spots off-track predictions and reason gaps, aiding verification of reason stability.
Catalog record
| CREATOR | Abu Noman Md Sakib, Zhensen Wang, Merjulah Roby, Zijie Zhang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32296 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.04456