AI-DEEP LEARNING BASED NOISY REDUCTION IN IMAGE LABELS WITH QUALITY EMBEDDING WITH CONVOLUTIONAL NEURAL NETWORKS (CNNs)

In the era of data-driven artificial intelligence, large-scale image datasets have become a cornerstone of success in deep learning applications.However, collecting clean and accurately labelled datasets at scale is both expensive and labor -intensive.As a result, real-world datasets often contain noisy labels-incorrect, ambiguous, or misleading annotations-that significantly impair the performance of deep neural networks.Addressing this challenge, this work introduces a novel framework for learning robust visual models from such noisy datasets by explicitly modelling the quality of image labels.