Deep Temporal Contrastive Clustering

Recently the deep learning has shown its advantage in representation and clustering for time series data. Despite considerable progress, existing approaches mostly seek to train neural network by some instance reconstruction based or cluster distribution objective, which, however, lack ability exploit sample-wise (or augmentation-wise) contrastive information even higher-level (e.g., cluster-level) contrastiveness discriminative clustering-friendly representations. In light of this, this paper presents a temporal (DTCC) approach, which first time, our knowledge, incorporates paradigm into research. Specifically, with two parallel views generated from original their augmentations, we utilize identical auto-encoders learn corresponding representations, meantime perform incorporating k-means objective. Further, levels are simultaneously enforced capture instance-level cluster-level information, respectively. With loss auto-encoder, loss, losses jointly optimized, architecture is trained self-supervised manner result can thereby be obtained. Experiments on variety datasets demonstrate superiority DTCC approach over state-of-the-art. The code available at https://github.com/07zy/DTCC .