San Francisco, CA (PRWEB) May 15, 2013
Ersatz, a PaaS from Blackcloud BSGa San Francisco based startupis the worlds first PaaS to make “deep learning” technology widely available. It arrives at a time when major technology companies like Google, Microsoft, IBM, Apple, and Baidu are all investing heavily in “deep neural networks”, a new and rapidly developing area of artificial intelligence research. While some of these companies have already incorporated the technology into their own products, it remains difficult for companies without major R&D dollars to take advantage of the power of deep learning algorithms.
Ersatz allows its users to create ensembles of deep neural networks that learn from data in ways modelled after how the brain works. Crudely modeled, points out the companys CEO, Dave Sullivan, 27. Its modelled after our brain in the same way that airplanes are modelled after birds. Theyre really different, but theyre trying to accomplish the same thing. Were building a sort of general intelligence, not trying to reproduce the human mind in all its glory. We just need the wings, not the feathers. The ensembles that users can create with Ersatz are simply groups of neural networks that learn to uncover different aspects of the same data. It’s like having a classroom of students learn something, then asking them to take a test together. The power of their minds combined produces a better result than if they had each tried individually. The same goes with ensembles of neural networks.
Users interact with Ersatz through a web interface and an API. The web interface provides an easy point and click way of creating and testing your own neural networks. The API is used for running the neural networks to make predictions in your applications. Ersatz also automates the complex process of parameter selection, so users just point it at their data, tell it what they want it to predict, and press go. Of course, power users can exert more granular control over configuration options if they want.
Non-deep neural networks are nothing new. In fact, says Sullivan. Neural networks have already been through two hype cycles–when they were invented back in the 50s, and again in the 80s when researchers discovered back propagation. During these two periods, computers turned out to be too slow to properly implement the techniques, so the artificial intelligence community moved on to other methods that proved more practical given the computing power available at the time.
Fast forward to today, and a few things have happened to change the AI landscape. First, computers of today are orders of magnitude faster than they used to be. Also, in 2006, there was a major breakthrough in neural network research that re-ignited interest in the field–this was the discovery of deep learning. Since then, researchers have developed a number of powerful new techniques, many of which fall under the umbrella term of deep learning. Finally, the data we generate in our lives has continued to multiply in volume–many have called this phenomenon big data. The problem initially was simply how to even store and access such large quantities of data. The problem now is to figure out what to actually do with all that data. Deep learning offers a possible solution.
Ersatz is the perfect tool for those interested in making more accurate predictions on highly complex time-series datasets. There is no shortage of industries where applications might be found: banking, healthcare, manufacturing, academic research, robotics, and many others. This product is essential for just about any company where the accuracy of predictions makes a big difference to the bottom line.
As far as competitive landscape, Google is perhaps the most important player in this developing market. They already use deep learning techniques for voice recognition in their Google mobile app. They have also made several key hires in the past several months, perhaps most notably Geoffrey Hinton and Ray Kurzweil, both titans in the field. However, Google does not currently have a product that offers other businesses the ability to use this technology in their own products.
Ersatz is currently in private beta and adding users as capacity allows. Interested parties can request an invite through the Ersatz website (http://www.ersatz1.com/).