• Rayala Kiran
• S. Satyanarayana
• Dodle. Sreenith Kumar
• Billakanti Sailaxmi
• Dumpala Akhil

This research paper presents SAINT (Self-Attention Integrated Neural Networks with Contrastive Pre-training), an innovative approach to improve neural networks’ performance on tabular data. SAINT incorporates row attention and contrastive pre- training techniques. Row attention selectively focuses on informative rows, capturing the hierarchical structure in tabular data. Contrastive pre-training utilizes large-scale unlabeled tabular datasets to enhance representation learning. Experimental results on benchmark datasets demonstrate SAINT’s superiority over traditional neural network architectures and state-of-the-art tabular data models. SAINT’s row attention mechanism captures complex feature relationships, leading to enhanced predictive accuracy and robustness across diverse tasks and datasets. The model contributes to advancing neural network architectures for tabular data analysis. SAINT’s incorporation of row attention and contrastive pre-training offers a valuable tool for researchers and practitioners in harnessing the power of neural networks for tabular data tasks. The insights gained pave the way for further advancements and more accurate predictive models in real-world applications.

Tabular data, neural networks, SAINT, row attention, contrastive pre-training, predictive modeling, hierarchical structure.

"SAINT: IMPROVED NEURAL NETWORKS FOR TABULAR DATA VIA ROW ATTENTION AND CONTRASTIVE PRE-TRAINING", International Journal of Emerging Technologies and Innovative Research (www.jetir.org), ISSN:2349-5162, Vol.10, Issue 7, page no.c859-c863, July-2023, Available :http://www.jetir.org/papers/JETIR2307294.pdf

"SAINT: IMPROVED NEURAL NETWORKS FOR TABULAR DATA VIA ROW ATTENTION AND CONTRASTIVE PRE-TRAINING", International Journal of Emerging Technologies and Innovative Research (www.jetir.org | UGC and issn Approved), ISSN:2349-5162, Vol.10, Issue 7, page no. ppc859-c863, July-2023, Available at : http://www.jetir.org/papers/JETIR2307294.pdf