In the realms of computer vision, it is evident that deep neural networks\nperform better in a supervised setting with a large amount of labeled data. The\nrepresentations learned with supervision are not only of high quality but also\nhelps the model in enhancing its accuracy. However, the collection and\nannotation of a large dataset are costly and time-consuming. To avoid the same,\nthere has been a lot of research going on in the field of unsupervised visual\nrepresentation learning especially in a self-supervised setting. Amongst the\nrecent advancements in self-supervised methods for visual recognition, in\nSimCLR Chen et al. shows that good quality representations can indeed be\nlearned without explicit supervision. In SimCLR, the authors maximize the\nsimilarity of augmentations of the same image and minimize the similarity of\naugmentations of different images. A linear classifier trained with the\nrepresentations learned using this approach yields 76.5% top-1 accuracy on the\nImageNet ILSVRC-2012 dataset. In this work, we propose that, with the\nnormalized temperature-scaled cross-entropy (NT-Xent) loss function (as used in\nSimCLR), it is beneficial to not have images of the same category in the same\nbatch. In an unsupervised setting, the information of images pertaining to the\nsame category is missing. We use the latent space representation of a denoising\nautoencoder trained on the unlabeled dataset and cluster them with k-means to\nobtain pseudo labels. With this apriori information we batch images, where no\ntwo images from the same category are to be found. We report comparable\nperformance enhancements on the CIFAR10 dataset and a subset of the ImageNet\ndataset. We refer to our method as G-SimCLR.\n