Deep clustering is a recent deep learning technique which combinesdeep learning with traditional unsupervised clustering. At the heartof deep clustering is a loss function which penalizes samples forbeing an outlier from their ground truth cluster centers in the latentspace. The probabilistic variant of deep clustering reformulates theloss using KL divergence. Often, the main constraint of deep clusteringis the necessity of a closed form loss function to make backpropagationtractable. Inspired by deep clustering and adversarial net, we reformulatedeep clustering as an adversarial net over traditional closed formKL divergence. Training deep clustering becomes a task of minimizingthe encoder and maximizing the discriminator. At optimality, thismethod theoretically approaches the JS divergence between the distributionassumption of the encoder and the discriminator. We demonstrated theperformance of our proposed method on several well cited datasetssuch as MNIST, REUTERS and CIFAR10, achieving on-par or better performancewith some of the state-of-the-art deep clustering methods.