Ensuring Data Integrity Through Robustness and Explainability in AI Models

Swathi Chundru

Abstract


Maintaining data integrity is crucial for machine learning programs to be effective and trustworthy in the era of artificial intelligence (AI). Data accuracy and reliability are more important than ever since AI systems are becoming more and more integrated into decision-making processes across a wide range of industries, including autonomous vehicles, healthcare, and finance. Any breach in this area can result in serious mistakes and risks. Data integrity refers to the correctness, consistency, and security of the data used to develop and assess AI models. This study explores resilience and explainability, two essential components of data integrity. The term "robustness" describes an AI model's resistance to adversarial attacks and data manipulation, which guarantees the model's dependability even under challenging circumstances. To make AI systems more resilient to many types of disruptions and attacks, strategies like adversarial training, data augmentation, and robust optimisation are investigated. By using these techniques, the risks related to data corruption are reduced and the models' ability to produce accurate and trustworthy results is maintained. Conversely, explainability aims to help users understand AI models' decision-making processes. It is imperative that consumers understand the process and rationale behind decision-making to promote trust and accountability. There are described approaches to clarify model predictions and enable meaningful interactions with AI systems, such as Shapley additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME). Our results demonstrate the necessity of combining robustness and explainability in order to create transparent and dependable AI systems. These components work together to enable us to develop AI solutions that protect data integrity, build user confidence, and guarantee sound decision-making in crucial applications.

Full Text:

PDF

References


S. J. Kim, E. K. Lee, and J. K. Lee, “Adversarial Training for Neural Networks: A Comprehensive Review,” IEEE Transactions on Neural Networks and Learning Systems, vol. 31, no. 3, pp. 912–926, Mar. 2020.

K. B. Tjandra and K. A. Barai, “The Impact of Data Augmentation on Deep Learning Models for Image Classification,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 42, no. 5, pp. 1102–1114, May 2020.

C. Szegedy, W. Zaremba, I. Sutskever, et al., “Intriguing Properties of Neural Networks,” in Proceedings of the 2014 International Conference on Learning Representations, 2014.

S. Ribeiro, C. Singh, and C. Guestrin, “‘Why Should I Trust You?’ Explaining the Predictions of Any Classifier,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016, pp. 1135–1144.

M. Ribeiro, S. Singh, and C. Guestrin, “Model-Agnostic Interpretability of Machine Learning Models,” in Proceedings of the 2018 Conference on Fairness, Accountability, and Transparency, 2018, pp. 409–418.

B. L. P. Ribeiro, R. K. Gupta, and C. J. Harris, “Visualization Techniques for Machine Learning Models: A Survey,” IEEE Transactions on Visualization and Computer Graphics, vol. 25, no. 5, pp. 1978–1990, May 2019.

Ronakkumar Bathani (2020) Cost Effective Framework For Schema Evolution In Data Pipelines: Ensuring Data Consistency. (2020). Journal Of Basic Science And Engineering, 17(1), .Retrieved From Https://Yigkx.Org.Cn/Index.Php/Jbse/Article/View/300


Refbacks

  • There are currently no refbacks.