Abstract
The paper provides a quantitative and qualitative description of deep learning research using bibliometric indicators covering global research publications published during 14-year period 2004-17. Global deep learning research registered 106.76% high growth per annum, and averaged 7.99 citations per paper. Top 10 countries world- over dominate the research field with their 99.74% global publications share and more than 100% global citations share. China ranks the top with the highest (29.25%) global publications share, followed by USA (26.46%), U.K. (6.40%), etc. during the period. Canada tops in relative citation index (5.30). International collaboration has been a major driver of research in the subject with 14.96% to 53.76% of national-level share of top 10 countries output appeared as international collaborative publications. Computer Science is one of the most popular areas of research in deep learning research (76.85% share). The study identifies top 50 most productive organizations and 50 most productive authors and top 20 most productive journals reporting deep learning research and 118 highly cited papers with 100+ citations per paper.
Highlights
Deep learning is an advanced way of achieving artificial intelligence using neural network algorithms
Top 10 countries world- over dominate the research field with their 99.74% global publications share and more than 100% global citations share
Global publications output on deep learning research in 14 years cumulated to 10027 publications during 2004-17
Summary
Deep learning is an advanced way of achieving artificial intelligence using neural network algorithms. Deep learning ( known as deep structured learning or hierarchical learning) is a specialized form of machine learning research. It is designed on the way human brain processes information and learns. Neural networks ( known as artificial neural networks) are trained to first learn how to perform artificial intelligence tasks by exposing them to a labeled data set and to defined neural network architecture. Deep networks can have as many as 150 layers Learning can be both supervised and unsupervised and it is applied to train and fine-tune neural networks using class target labeled data set of inputs and expected outputs. Neural networks come in several different architecture types such as deep neural networks, deep belief networks, and recurrent neural networks [1,2,3]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
