WCSE 2022
ISBN: 978-981-18-3959-7 DOI: 10.18178/wcse.2022.06.039

Deeptab: A Deep Neural Network for Non-uniform Tabular Data

Xinyu Jiang, Hongyu Guo, Qian Li

Abstract— Over the last years, deep neural networks have been the optimal solution for most tasks in several fields. Moreover, various algorithms have been implemented in industrial applications, such as face recognition, language translation, object classification, and object detection. However, when we use deep neural networks for practical applications on tabular data, we find the problem of non-uniform training data and test data. The training data usually exhibit high-quality forms, while the actual prediction data are lower quality than the training data. We propose Deeptab, a deep neural network for non-uniform tabular data based on this problem. Specifically, we make the training data mimic the actual data situation as much as possible and improve the robustness and flexibility of the network by coding each feature separately. We eventually tested our approach on public datasets and our real datasets and achieved better results. We also tested the impact of training data quality and found a better treatment for non-uniform tabular data.

Index Terms—peak clustering, siphon effect, dichotomy, density clustering, unsupervised learning

Xinyu Jiang, Hongyu Guo, Qian Li
The 15th Research Institute of China Electronics Technology Group Corporation, CHINA

[Download]


Cite:Xinyu Jiang, Hongyu Guo, Qian Li, "Deeptab: A Deep Neural Network for Non-uniform Tabular Data, " Proceedings of 2022 the 12th International Workshop on Computer Science and Engineering (WCSE 2022), pp. 275-279, June 24-27, 2022.