Abstract
Human beings learn different things from their experiences in life, beginning from when they were born to this planet and they keep improving as the day goes by. Nevertheless, machine learning approaches are becoming more useful in all facets of life, becoming better every day just like human beings. Thus, on the contrary to machine learning models for learning a sole function from a well-organized dataset in a short frame of time, humans learn various functions cumulatively. sHowever, to prevent this learning method from suffering a setback when marking errors amass, there is a need to ensure the system intermingle with a human for few minutes per day to aid it remains on track. Owing to this, the objective of this study was aimed at reviewing never-ending learning language techniques and framework or architecture, which is an intelligent machine that runs without ending and on each day extracting, reading info from the web pages to populate a growing structured knowledge base and learn to carry out tasks better as days goes by. Our never-ending language learner (NELL) architecture and techniques propose 4 subsystems that ran for about 66 days, implementing about 241,000 applicant’s beliefs with a predictable exactness of 73%. Our finding is of great benefit in the sense that it exhibits the advantages of employing a different set of mining knowledge techniques that are responsive to learning and KB-knowledge base which enables the storage of the applicant’s pieces of evidence and beliefs.
Keywords: Never-ending learning language, Machine learning, knowledge base, Human learning.
Introduction
The success of machine learning (ML) as a division of artificial intelligence (AI) and its wide acceptance and application for tasks ranging from speech detection to spam filtering, to face detection, to credit-card swindle exposure (Mitchell et al., 2015). Even with these great feats, computer learning is remarkably narrow once matched to human learning. When a never-ending learning language is mentioned, it refers to another prototype to ML that is more narrowly replicates the competence, diversity, and cumulative nature of human learning.
To demonstrate this, it is pertinent to note that in each of the aforementioned applications of machine learning, the computer learns only a sole function to execute a distinct assignment in isolation, generally from human marked working out instances of feedbacks and outcomes of that function. In the case of spam filtering, the working out patterns involved of particular emails that are non-spam or spam marks for each. This method of learning is frequently known as supervised or controlled learning, owing to the non-concrete learning difficulty, which is to estimate some anonymous function, f: X → Y given a training set of feedback/outcome sets {〈xi, yi〉} of that function. Apart from the supervised function, there are others classes of ML models that are in existence. These include unsupervised learning, semi-supervised learning, reinforcement clustering, and topic modeling, just to mention a few, on the other hand, these machine learning paradigms are typically acquired just a distinct function or data pattern from a sole database.