A Novel Hierarchical Clustering Technique to Analyze Style and Content Factorization During Image Recognition

Advancement in various technologies such as 5G and IoT has led to massive data generation. Data proliferation has brought along enormous privacy threats. Often privacy threats are overlooked by users. In this chapter, we exhibit the hidden privacy threat of user identification from handwriting. We propose a novel unsupervised clustering technique by which the attacker learns to group users according to the writing style. Distinctively, we use the handwritten digits dataset to describe the potential threat. Experimental results on digit recognition are discussed to elaborate on the possibility of exposing hidden information without affecting the actual performance. The motivation for the proposed work is to emphasize on the effects of overlooking privacy. Through our experimental results, we showcase the possible threats underlying in digit recognition task that is a popular machine learning task.