Document worth reading: “Small Data Challenges in Big Data Era: A Survey of Recent Progress on Unsupervised and Semi-Supervised Methods”

Small data challenges have emerged in many learning points, given that success of deep neural networks often relies upon on the provision of an infinite amount of labeled data that is pricey to collect. To deal with it, many efforts have been made on teaching superior fashions with small data in an unsupervised and semi-supervised pattern. In this paper, we’re going to evaluation the newest progresses on these two predominant lessons of methods. A huge spectrum of small data fashions will in all probability be categorized in a large picture, the place we’re going to current how they interplay with each other to encourage explorations of new ideas. We will evaluation the requirements of learning the transformation equivariant, disentangled, self-supervised and semi-supervised representations, which underpin the foundations of newest developments. Many instantiations of unsupervised and semi-supervised generative fashions have been developed on the premise of these requirements, considerably rising the territory of current autoencoders, generative adversarial nets (GANs) and totally different deep networks by exploring the distribution of unlabeled data for further extremely efficient representations. While we focus on the unsupervised and semi-supervised methods, we are going to even current a broader evaluation of totally different rising topics, from unsupervised and semi-supervised space adaptation to the fundamental roles of transformation equivariance and invariance in teaching a big spectrum of deep networks. It is just not potential for us to jot down an distinctive encyclopedia to include all related works. Instead, we aim at exploring the first ideas, guidelines and methods in this area to reveal the place we’re heading on the journey in path of addressing the small data challenges in this large data interval. Small Data Challenges in Big Data Era: A Survey of Recent Progress on Unsupervised and Semi-Supervised Methods