DeepCluster is a novel unsupervised learning method for training convolutional neural networks (CNNs) on large datasets. It jointly learns the parameters of the CNN and the cluster assignments of the resulting features using a standard clustering algorithm, k-means. The method iteratively clusters the features and uses the cluster assignments as pseudo-labels to update the network weights. DeepCluster is applied to the unsupervised training of CNNs on datasets like ImageNet and YFCC100M, achieving state-of-the-art performance on various standard benchmarks. The approach is robust to changes in architecture and training set, demonstrating its effectiveness in diverse scenarios. The paper also discusses the current evaluation protocol in unsupervised feature learning and evaluates DeepCluster on instance-level image retrieval tasks, providing insights into the quality of features produced by unsupervised methods.DeepCluster is a novel unsupervised learning method for training convolutional neural networks (CNNs) on large datasets. It jointly learns the parameters of the CNN and the cluster assignments of the resulting features using a standard clustering algorithm, k-means. The method iteratively clusters the features and uses the cluster assignments as pseudo-labels to update the network weights. DeepCluster is applied to the unsupervised training of CNNs on datasets like ImageNet and YFCC100M, achieving state-of-the-art performance on various standard benchmarks. The approach is robust to changes in architecture and training set, demonstrating its effectiveness in diverse scenarios. The paper also discusses the current evaluation protocol in unsupervised feature learning and evaluates DeepCluster on instance-level image retrieval tasks, providing insights into the quality of features produced by unsupervised methods.