Uncovering the Role of Research in Machine Learning Industry

Research plays a crucial role in the development of machine learning technology. With the advancement of machine learning, the research community has adopted a “bigger, better” approach, requiring a massive amount of data and computing power, which costs tens of millions of dollars. Unfortunately, most companies and academic institutions cannot afford to pursue pure research due to these high costs.

Fortunately, there are only a handful of machine learning research labs in the world, and most of them are funded by corporations such as Alphabet, Microsoft, Facebook, and Tencent. These labs can be found by browsing the affiliations of published papers at major academic conferences, including NeurIPS, ICLR, ICML, CVPR, and ACL. In 2019 and 2020, Alphabet accounted for over 10% of all papers at NeurIPS.

Research vs. Applied Research

Some companies might have roles that involve applied research, which is somewhere between research and production, but much closer to research than production. Applied research involves finding solutions to practical problems but doesn’t involve implementing those solutions in actual production environments.

Applied researchers are researchers who come up with novel hypotheses and validate them, but they need to have subject matter expertise since their hypotheses and theses deal with practical problems. For example, in machine learning, a research project would be to develop an unsupervised transfer learning method for computer vision and experiment on a standard academic dataset. An applied research project, on the other hand, would be to develop techniques to make that new method work on a real-world problem in a specific industry, such as healthcare.

Research Scientist vs. Research Engineer

There is much confusion about the roles of a research scientist and a research engineer. A research engineer is a rare role often seen at major research labs in the industry. If the role of a research scientist is to come up with original ideas, the role of a research engineer is to use their engineering skills to set up and run experiments for these ideas. While the research scientist role typically requires a Ph.D. and/or first author papers at top-tier conferences, the research engineer role doesn’t, though publishing papers always helps.

In some teams, there is no difference between a research scientist and a research engineer. Both come up with ideas and implement them, with the researcher acting as an advisor guiding research engineers in their own research. It’s not uncommon to see research scientists and research engineers be equal contributors to papers. The different job titles are mainly a product of bureaucracy, with research scientists supposed to have bigger academic clout and often being better paid than research engineers.

Startups may be more generous with job titles to attract talent. For instance, a candidate chose a startup over a FAAAM company because the startup gave him the title of a research scientist, while the big company gave him the title of a research engineer.

Conclusion

The machine learning industry is evolving rapidly, and research is a vital component of this evolution. However, pursuing pure research is expensive, and most companies and academic institutions can’t afford it. Applied research is a more practical approach, which aims to solve real-world problems. Additionally, research scientists and research engineers have distinct roles, although they might be equal contributors to papers. Regardless of the job title, individuals in the machine learning industry need to have subject matter expertise and engineering skills to be successful.