The number of papers are indications of interest and not impact. Indeed more people working on the same problem can generate more ideas. But more ideas do not necessarily generate more impactful ideas when ideas are constrained by groupthink.
https://twitter.com/stochastician/status/1443666778454888451
What is driving the interest in deep learning is of course its phenomenal success. This leads to more funding and more advanced tools. There are diminishing returns in every field as the low-hanging fruit is picked.
Like in any field the early adopters are always rewarded disproportionately more than the latecomers. Unfortunately, it is human bias to recognize more the pioneers.
To gain the greatest upside requires greater risk. Risk gets magnified when the cost of doing business increases. In Deep Learning, the cost of research is growing exponentially. Hence the greater risk leads to less revolutionary and impactful discoveries.
Furthermore, as a field matures, the majority of innovation is more like engineering where we are optimizing different combinations of known solutions to tackle a new problem. I would argue that most Deep Learning papers are of an engineering nature.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
