Supervised Shallow Multi-task Learning: Analysis of Methods

The last decade has witnessed a continuous boom in the application of machine learning techniques in pattern recognition, with much more focus on single-task learning models. However, the increasing amount of multimedia data in the real world also suggests that these single-task learning models have become unsuitable for complex problems. Hence, multi-task learning (MTL), which leverages the common path shared between related tasks to improve a specific model’s performance, has grown popular in the last years. And several studies have been conducted to find a robust MTL method either in the supervised learning or unsupervised learning paradigm using a shallow or deep approach. This paper provides an analysis of supervised shallow-based multi-task learning methods. To begin, we present a rationale for MTL with a basic example that is easy to understand. Next, we formulate a supervised MTL problem to describe the various methods utilized to learn task relationships. We also present an overview of deep learning methods for supervised MTL to compare shallow to non-shallow approaches. Then, we highlight the challenges and future research opportunities of supervised MTL.