EUVIMEDEUVIMED entdecken
Crossref

On the Impact of Weight Quantization on Deep Neural Network Uncertainty

Shuang Liang, Xun Lu, Zi-Ang Liu, Ming-Liang Wang, Yan Lyu, Shao-Qun Zhang

Proceedings of the AAAI Conference on Artificial Intelligence · 14.03.2026 · 40 (28) · 23425-23432

DOI: 10.1609/aaai.v40i28.39513 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 14.03.2026. Verfasst von Shuang Liang und 5 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Weight Quantization (WQ) is a key technique for lightweight Deep Neural Network (DNN) computations. While existing algorithms often pursue memory compression and inference acceleration with accuracy comparable to full-precision models, the effect of WQ on DNN uncertainty remains largely unexplored. In this paper, we quantify the impact of WQ on DNN uncertainty through the novel Exact Moment Propagation (EMP) uncertainty estimator. It is observed that WQ significantly increases DNN uncertainty. Based on the EMP estimator, we propose the MOMent Alignment (MOMA) to reduce WQ-induced uncertainty and preserve the accuracy of weight-quantized DNNs. Empirical results across various DNN architectures and datasets validate the effectiveness of both EMP and MOMA methods.

Zitieren

APA

Liang, S., Lu, X., Liu, Z., Wang, M., Lyu, Y., & Zhang, S. (2026). On the Impact of Weight Quantization on Deep Neural Network Uncertainty. Proceedings of the AAAI Conference on Artificial Intelligence, 40(28), 23425-23432. https://doi.org/10.1609/aaai.v40i28.39513

Vancouver

Liang S, Lu X, Liu Z, Wang M, Lyu Y, Zhang S. On the Impact of Weight Quantization on Deep Neural Network Uncertainty. Proceedings of the AAAI Conference on Artificial Intelligence. 2026;40(28):23425-23432. doi:10.1609/aaai.v40i28.39513

BibTeX
@article{Liang2026,
  author = {Liang, Shuang and Lu, Xun and Liu, Zi-Ang and Wang, Ming-Liang and Lyu, Yan and Zhang, Shao-Qun},
  title = {On the Impact of Weight Quantization on Deep Neural Network Uncertainty},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  year = {2026},
  volume = {40},
  number = {28},
  pages = {23425--23432},
  doi = {10.1609/aaai.v40i28.39513}
}
Kostenlos starten →