The Growth of AI and Machine Learning in Computer Science Publications

One of these days, while working, I randomly found a dataset containing information about computer science (CS) papers on major journals and conferences. I then decided to have a look at how the CS publications’ topics shifted over time and how machine learning (ML), deep learning and artificial intelligence, the current CS buzzwords, have been used lately.

I first plotted how many papers were published in each year from the 1930s until 2018. Figure 1 shows a huge increase in the last two decades or so.

Figure 1: Number of publications over time.

Every computer scientist, student and actually anyone in the world has heard about AI, ML and all these buzzwords. It is literally everywhere. I chose 6 terms that are related to AI and ML: artificial intelligence, deep learning (DL), machine learning, k-means, clustering and SVM. The last three were chosen in order to analyze what the eruption of deep learning has done to more classical machine learning methods, like k-means and SVM. Finally, I counted how many times these terms were included in the title of a paper. Figure 2 shows the absolute number of papers with each one of the terms over time and Figure 3 shows the ratio between the number of papers with the terms and the total number of papers in a designated year.

Figure 2: Absolute number of publications with the selected terms in their title over time.

Figure 2: Ratio of publications with the selected terms in their title over time.

Having a look at Figure 2, the first thing to notice is the exponential increase in papers related to AI and ML in the last 2~3 decades. Analyzing Figure 3, it is possible to see that both AI and ML had peaks in the late 1950s and early 1960s, with AI maintaining this trend until the early 1980s and ML reborning in the middle of the 1980s. Deep learning became really famous in this last decade, which resulted in the decrease of papers about classical ML algorithms, such as clustering and SVM, which were really important in the inception of ML research.

Finally, I also checked how many papers were published in 6 of the top CS conferences (CVPR, NIPS, KDD, ACL, EMNLP and ICML), where many ML and AI works are published nowadays. Figure 4 shows the number of yearly papers published by conference. Most conferences started to grow in the 2000s with really big jumps in the last decade or so. Some remarkable events such as the establishment of DL and the ImageNet Challenge resulted in a big increase in publications in NIPS and CVPR, respectively.

Figure 4: Number of publications in the selected conferences over time.

Machine learning, especially deep learning, and artificial intelligence are certainly part of our present and future. These buzzwords are here to stay, both in academia and our lives.