Machine Learning: From The Realm of Academia To Industry

One of the most striking observation I made in the past couple of years in the Machine Learning domain is the gradual shift in the demographics of ML Engineers from academia to that of the industry.

A couple of years ago, ML and AI applications were built by a group consisting predominantly of academicians and researchers in the ML domain. This could be seen from the mass hiring of entire ML departments, students and teachers included, of colleges like MIT by companies like Google, Apple etc.

The main reason for this could be attributed to the niche factor of the applications of ML during the formative years of ML in the software industry. Most of the ML models that these researchers were hired to build were highly niche, project specific and turnkey products that were not extendible. This had a ripple effect on the ML industry where there was no incentive for building any frameworks with extendible functionality for building ML models. Thus, in order to build any ML model, every component had to be built from ground up using first principles. This discouraged adoption from the engineers in the industry who are always looking for easy solutions for fast prototyping and not spend months building a certain model. Thus, ML was relegated to the experimental section of the projects in the mainstream software industry, where not much importance was given to the necessity of such applications.

This was the situation for a long time before Google made the decision to build TensorFlow and make it open source. As ML applications gradually took a turn from being niche products to being ubiquitous, there was more and more demand for a reusable, extendible framework where focus was on rapid prototyping rather than on building the components from scratch.

If the realm of ML applications is observed from a birds' eye, most of the components are inherently reusable. Be it the neurons in a Neural Network, or the Decision Nodes in a Decision Tree, every component implements a predefined function that remains constant irrespective of the use case for which the model is being built. This led to the explosive adaptation of ML frameworks such as TensorFlow, Theano or Caffe. Now, the focus shifted from "how to build the module" to "what module to add to the model". Building a ML model was reduced to playing with pre-built modules, similar to playing around with Lego blocks.

With the advent of even higher level of abstraction over these frameworks like Keras, building an ML model became even simpler. There is increased penetration of these technologies in the software industry with engineers with no formal degree in ML or statistics trying their hands on ML applications.

With the release of TensorFlow v1.0 and the in built support for Keras, I believe that there would be a noticeable shift in the demographic of people building ML applications. It will no longer be dominated by researchers/academicians. Rather, it will be widely accepted by the industry as well as the developers, along the lines of web/mobile development. Thus leading to the shift from "Machine Learning Scientist" to "Machine Learning Developer".