Weak Supervision, Future of Data Labeling

Overview of data labelling for AI, new paradigms, and size of the growing data labelling market.

Daulet Nurmanbetov
The Startup

--

Photo by Simon Matzinger on Unsplash

Data Labeling Landscape

With the emergence of AI, many firms coming to realize the real bottleneck to bootstrapping Machine Learning is a lack of labelled datasets.

In response to the need, many companies emerged that offer labelling services, labelling platforms or some other labelling solutions such as providing specific domain-experts to label data for AI algorithms. Specifically, in the last couple of years, many older services companies started offering data labelling solutions and many new upstarts were formed to tackle lack of labelled data, catering specifically to the AI domain. Here how the data labelling landscape looks today —

Data Labeling Landscape on Jan-2020 by Daulet Nurmanbetov

With data labelling being in such demand new paradigms have been gaining traction — Active Learning and Weak Supervision, witch stand to disrupt general, non-specialized crowd-sourced labelling services.

Alternatives to pure crowd-labelling

--

--