SAKCL: a deep neural network test data selection method based on self-attention and K-means clustering
Tingting Huo, Qiang Sun, Rui Ding · 2026
Read SAKCL: a deep neural network test data selection method based on self-attention and K-means clustering by Tingting Huo, Qiang Sun, Rui Ding (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/PMC13529774.txt.
Description
Tingting Huo, Qiang Sun, and Rui Ding pick better test data for neural networks in a 2026 paper. Weak test sets let faulty models slip through, which is risky in safety fields. Their method, SAKCL, first weighs feature parts by self-attention. That step lifts the informative patterns in each sample. Clustering then groups the polished samples by shape.
Picks from the clusters balance fault-finding power with variety. Four standard image sets and three network designs hosted the trials. The method raised the share of error-revealing cases by over twelve percent on average. Diversity scores rose more than six percent. Retraining on the picked data gained about three points of accuracy. Math checks backed the gains as real. The authors offer the method as a practical quality gate for model testing.
Catalog record
| CREATOR | Tingting Huo, Qiang Sun, Rui Ding |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 81694 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |