Research firm Cognilytica recently released a briefing note focusing on iMerit’s data labeling work and the role of labeling in the development of machine learning algorithms.
Here’s an excerpt from the note:
In order for many machine learning algorithms to be trained, especially supervised learning algorithms, they need to be fed relevant data that has been appropriately “labeled” with the required output that needs to be learned. Algorithms must be trained on hundreds to thousands of images that have been precisely marked in the data labeling process. For example, image recognition systems that use deep learning neural network approaches need large volumes of clean, normalized image data where the image has been properly labeled as the desired output to train the system over multiple training iterations to build a model that can generalize properly to recognize future images. Such labeling needs to happen for any supervised learning application.
Data preparation activities, including data labeling, takes up a significant amount of the time for most AI and machine learning projects.
Customers have the data, but they don’t have the resources to label large data sets, nor do they have a mechanism to insure accuracy and quality. Raw labor is easy to come by, but the assurance of quality is not easy to guarantee. In addition, labeling projects involve multiple steps as well as requiring human subjective decision making.
The primary competitive advantage that iMerit has is its delivery excellence and quality of its solution architects. They also have greater flexibility on technology as they aren’t rigid on the required platform of interaction and work with customer’s existing technologies and platforms. Their recent partnership with AWS also shows the high quality of the product they deliver.
