LinkedIn Donates Feature Store to Linux Foundation


LF AI & Data is the Linux Foundation's umbrella organisation for big data and AI initiatives, and today LinkedIn announced that Feathr, its open source feature store, has joined the foundation.

Feathr was originally created by LinkedIn to aid in the management and service of features used in the company's machine learning applications. Feathr automates and standardises how the data type,

which is used in both the training and inference stages of machine learning is interacted with, so that users don't have to spend time working with features by hand as part of their own data pipeline

Feathr was inspired by a desire to improve machine learning software's reliability, precision, and efficiency. Once the data features utilised by ML programmes have been defined in a shared

feature namespace, they can be accessed "by name" from within ML workflows. Because of this, the same features may be employed in different ML programmes, raising both efficiency and precision. 

Also, the centralization of feature storage and serving in feature stores improves the performance of ML serving at the inference stage, which is not the case with all feature stores. 

The adoption rate of Feathr has increased since its 2017 debut within the company. It seems that LinkedIn is using the feature store to keep tabs on the thousands of features it employs.

According to a blog post published today by LinkedIn data infrastructure engineer Hangfei Lin, "It has decreased the engineering time necessary for introducing and experimenting with new features 

from weeks to days." The authors write that it's also up to 50% faster than the proprietary feature processing pipelines it replaced. LinkedIn hopes that by donating Feathr to The Linux Foundation's 

LF AI & Data group, the open source project will be better governed, resulting in an increase in both users and contributions. In April, LinkedIn made available to the public the Feathr codebase

Labriola win over the Bengals

Click Here To See