In fact, the company has been at the front row seat as far as dealing with fake news as well as filtering offensive content using artificial intelligence and machine learning.
What’s more, Facebook has not only been concentrating on resolving problems but also creating products that are linked to recommendation engines, image and text analysis.
Thanks to the leadership of deep learning guru Yann LeCun, FAIR has performed exemplary in both computer vision and coreML, which created cutting-edge algorithms that can comprehend text.
These algorithms are incorporated into the machine learning (ML) platform to promote and grow ML from the training to deployment.
1. Building Perception
LeCun pioneered the idea behind building on the convolutional neural network. Building Perception’s neural networks are trained to comprehend a wide array of data including voices, videos and photos better.
2. FB Learner Flow
Labelled as AI’s backbone. This platform has the potential to reuse algorithms in various products, scale to operate thousands of concurrent custom experiments and administer them easily.
What’s more, FB Learner Flow also offers functionality including automatic generation of user interface experiences. The platform is utilized by over 25 per cent of Facebook’s engineering team.
Since the launch of FB Learner Flow, over a million models have undergone training while their forecast service has grown to make over 6 million forecasts per second.
Among the key highlights of this program is its ability to get rid of the time spent on feature engineering as well as allowing all engineers to run numerous experiments.
In fact, FB Learner Flow can conduct simulations of a whopping 300,000 machine learning models each month.
3 Deep Text
The DeepText feature builds on DeepText and utilizes deep learning methods to deal with language problems that cannot be solved using conventional natural language processing (NLP) approaches.
Additionally, this Facebook’s text understanding e leverages a mathematical idea that can comprehend the semantic relationship that exists between words and it comes in handy when creating language-agnostic models.
4. Facebook’s Facial Recognition application, DeepFace
This application enabled it to identify individuals in images with an astounding accuracy rate of 97%.
According to FB research, DeepFace is trained on the biggest facial dataset and can recognize a labelled dataset made up of four million facial images that belong to over 4,000 identities.
What’s more impressive about the application was the fact that it reached an accuracy rate of 97.35 %, particularly on the Labeled Faces in the Wild (LFW) dataset. It minimizes the current state of art’s error by over 27%, which makes its performance close to that of humans.
5. Open Sourced AI Hardware Design
According to a Facebook post, the social media giant open-sourced its artificial intelligence (AI) hardware design, which was termed as the best globally back in 2015.
Big Sur, the Open Rack-compatible hardware that is designed for artificial intelligence (AI) computing, was created in conjunction with partners and can operate on networks that are double its size.
What’s more, Facebook also open-sourced its Deep Learning module Torch, which is utilized in creating neural networks.