Engineering is the bottleneck in Deep Learning Research

Engineering is the Bottleneck in Deep Learning Research gives a really important message that ’engineering’ is valuable. 90% of research is stuff that has been done before, but slightly reskinned to suit your problem. The last 10% is the innovation. You want solutions engineered to get you past this 90% as quickly and easily as possible.

The difficulty of building upon another’s work is a major factor in determining what research is being done.

All research should have engineering, by which I mean rigor. With the rate at which software falls apart, we can’t really call it software engineering. We need a way of getting baselines, evaluation, related work etc. All the crap that isn’t your innovation but needs to go alongside it.

Most researchers care more about their publications, citations, and tenure tracks than about actually driving the field forward.