Deep Neural Network Learns Language from Wikipedia

Share Article

Ersatz (made by Blackcloud BSG) is neural network software that has learned to model language using text from Wikipedia. This has exciting future uses in search, email analytics, and translation software.

Big news for big data: the makers of Ersatz (, a platform for building "deep neural networks" in the cloud, have fed their algorithm over 4 million Wikipedia articles, and this word cloud is what it learned:

The 3D word cloud combines over 25,000 words into a highly ordered network of associations. To accomplish this task, Ersatz used the patterns of the English language to learn what words go where and how they are most often presented. Much like a child learning to read, Ersatz was able to recognize semantic relationships between words that are commonly used in conjunction with each other. Ask Ersatz what it knows about “car,” for example, and it will show you other comparison words, like “automobile,” “motor,” “driver” and “auto.” In a nearby cluster, Ersatz has learned to group auto manufacturers: "dodge", "ford", "jeep", and "cadillac".

Feed Ersatz the language of your business, and there’s no telling what it could learn.

Blackcloud BSG, the brain-behind-the-brain that is Ersatz, is looking to extend this process to a business' e-mail data in order to cluster different sales opportunities or identify particularly effective employees, among other applications. The thinking is that much like the English language, a business is a complex web of ideas and associations which needs to be properly ordered before big decisions are made. Traditionally, businesses with big data have relied upon the work of analysts to conjure this order and its value. Blackcloud BSG is changing all of that by positioning Ersatz as an invaluable tool for companies that want to get started with deep learning without hiring their own fleet of deep learning experts.

Ersatz provides an API for building custom applications and can be fed virtually any kind of data, so long as the data is computerized. Possible use cases span almost every industry. Unlike your brain, Ersatz does not get tired and doesn’t complain. It simply shapes static data into patterned models, which can lead human decision makers to new lines of insight and thinking.

Major companies like Google, Facebook, Microsoft, and Baidu are currently competing to hire fleets of deep learning researchers so they can improve their products with this type of technology. Since the larger companies will no doubt keep "the best stuff" for themselves, this leaves many companies out in the cold and potentially falling behind. Ersatz is one solution--rather than build your own, why not buy a ready-made implementation from a vendor? Ersatz is like a Google Brain or Facebook AI that people can actually buy.

What will Ersatz learn next? Its users will get to decide.

Share article on social media or email:

View article via:

Pdf Print

Contact Author

Lucy Karpova
Follow us on
Visit website