ABSTRACT
The learning in the ML model can be classified into four classes, and they are Supervised learning, Unsupervised learning, Semi-supervised learning, and Reinforcement learning. In semi-supervised learning, the majority of the input data vectors in the dataset are not labelled, the remaining being minority data vectors possessing class labels. Tree topology is popular in distributed machine learning frameworks, used in ALLReduce to compute local gradients in child nodes and pass on the value to their parents. Bulk Synchronous Parallel, Stale Synchronous Parallel, Approximate Synchronous Parallel, and Barrierless Asynchronous Parallel /Total Asynchronous Parallel are the four communication models adopted in distributed machine learning ecosystem. The communication models like parameter servers do not ensure privacy in distributed systems. A gamut of machine learning algorithms and deep learning algorithms along with threats, and challenges discussed in this article will provide a glimpse into existing state-of-the-art to readers.
