Data Regeneration from Poisoned Datasets

The administration of data in soft copies is growing increasingly popular as organizations switch from keeping documents in hard copies to more contemporary information handling systems that enable faster sharing and exchanging of data. Unfortunately, as the data has gotten easier to manage, modify, and store, it has also become easier to lose or corrupt them. It is well known that many approaches, including Autoencoders, K means, and generative models, can identify noise and regenerate data in response. Using the unique modified Adamic Adar technique, which was previously used for link prediction in the context of graphs, we tackled the aforementioned issue in this study. On the same medium-sized dataset, our suggested solution with an accuracy of 78% outperformed more established techniques like K means, which had an efficiency of only 50%. We further regenerated them using several regression and deep learning models. The research also discusses the extension of GANs, which are well renowned for their effectiveness in image generation, to lost data regeneration.