Sequence of Actions Recognition Using Continual Learning

Abstract

Continual Learning is the basic fundamental idea behind Artificial Intelligence and Machine Learning. With C.L, the machine will autonomously learn and translate the data into meaningful information like humans. Additionally, it can be defined as a model that automatically keeps on fine-tuning itself by acquiring knowledge over time throughout its lifespan. Moreover, this paper tries to solve a long-remained problem of human action prediction using continual learning. The human action prediction is an indispensable component for featuring a seamless human-robot interaction in the future. And along with that, such models and implementations have tremendous scope in almost every field in the future ranging from security to elderly or child care or law and order maintenance and automation and whatnot. Numerous implementations use simple machine learning models, which wear out over the period. Using continual learning, a model can be designed which improves itself without requiring any human intervention. And the model will be able to produce substantially improved results because of continual learning. The model will not only be able to get better with the objects coming in its way in the future but also will be able to improve the model and get additionally polished results.