DATA ANNOTATIONS     AI/ML

Democratizing ML in the Enterprise

July 19, 2020

ML is everywhere now and it is an accepted truism that enterprises that harness it better will lead the future. Now, traditionally the CTO organization has been leading the tech innovation(s) in an enterprise, but ML is very different. It entails a lot of experimentation and domain knowledge before any successful outcome can be orchestrated. The CTO organizations and the data scientists may not be fully geared to meet this challenge.

In an ideal world, the business users who stand to gain the most from this should be driving this. There are two major hurdles for this. Firstly the tech seems to be out of bounds for most business users. Increasingly though platforms like Azure, Sagemaker and GCP are creating “zero code” ML platforms – which will only get better over the years. Now, added to this is the “800 pound gorilla” in the room; most of the successes in the ML space are in “supervised learning” which means humans first need to label the “occurances of interest” in the data before the machine can start learning the patterns.

This annotation process is usually extremely laborious and time consuming – hence most enterprises outsource this out at the lowest cost. The problem here of course is two fold:

  • Most annotation tasks require a high level of domain expertise.

  • If the annotations are incorrect, the entire ML exercise of building sophisticated models is largely wasted – between a better model and more accurately annotated data, the data always wins, especially in this deep learning era.

The key to having business users interested in annotations and driving the same is a better software platform, that not only makes annotations simpler but also uses techniques like active learning so that just the minimal number of samples required need to be manually annotated.

In no way, I am suggesting that annotations are the only important thing in getting ML to work in enterprises, but just that more attention needs to be paid on developing software platforms that make annotations easier for business folks to work on.

The key to making ML more democratic within Enterprises and having business users interested is ensuring that they get more involved within the annotation process.