Design Perspectives of Multi‐task Deep‐Learning Models and Applications
Yeshwant Singh, Anupam Biswas, Angshuman Bora, Debashish Malakar, Subham Chakraborty, Suman Bera · 2022
Multi-task learning (MTL) has performed exceptionally well in a variety of applications in recent years. While single model training has vowed remarkable results over the years, it overlooks vital information that may help us predict an evaluation metric more accurately. MTL generalizes deep-learning models even more than single-task learning in learning-related activities. MTL aims to learn generalized feature mapping models by exchanging features across tasks and using inductive transfer learning. Furthermore, we are interested in understanding the task linkages between different activities to gain more benefits from MTL. The goal of this chapter is to visualize existing MTL models, compare their performances, discuss the methods used to evaluate the performance of these MTL models, and discuss the problems encountered during the design and implementation of these models in various domains, as well as the benefits and milestones attained by them.