Efficiency Calibration of Implicit Regularization in Deep Networks via Self-Paced Curriculum-Driven Singular Value Selection
Abstract
We address the context of Single-Audience Value-Based Abstract Argumentation Framework (AVAF), where the arguments are labeled with the social values that they promote and the activation/deactivation of the attacks depends on the audience profile (expressed as a set of preferences between the social values). Herein, we introduce a new notion of robustness for measuring the sensitivity of the outcome of the reasoning to the extent of changes in the audience profile. In particular, for a set of arguments S or a single argument a, we define the robustness degree of the status of S or a as the maximum number k* of deletions/insertions of preferences from/into the audience profile that are tolerable, in the sense that S remains an extension (or a non-extension) or a accepted (or unaccepted) after performing at most k* deletions/insertions. We introduce the decision problems related to the computation of the robustness degree and focus on thoroughly investigating their computational complexity.
Cite
Text
Li et al. "Efficiency Calibration of Implicit Regularization in Deep Networks via Self-Paced Curriculum-Driven Singular Value Selection." International Joint Conference on Artificial Intelligence, 2024. doi:10.24963/ijcai.2024/499Markdown
[Li et al. "Efficiency Calibration of Implicit Regularization in Deep Networks via Self-Paced Curriculum-Driven Singular Value Selection." International Joint Conference on Artificial Intelligence, 2024.](https://mlanthology.org/ijcai/2024/li2024ijcai-efficiency/) doi:10.24963/ijcai.2024/499BibTeX
@inproceedings{li2024ijcai-efficiency,
title = {{Efficiency Calibration of Implicit Regularization in Deep Networks via Self-Paced Curriculum-Driven Singular Value Selection}},
author = {Li, Zhe and Chen, Shuo and Yang, Jian and Luo, Lei},
booktitle = {International Joint Conference on Artificial Intelligence},
year = {2024},
pages = {4515-4523},
doi = {10.24963/ijcai.2024/499},
url = {https://mlanthology.org/ijcai/2024/li2024ijcai-efficiency/}
}