Ersatz Labs, the company behind the deep learning platform Ersatz™, has announced the out-of-beta launch of its flagship product. Ersatz is the first deep learning platform and is available either as a cloud service or as a deep learning “appliance”, a combination of neural network hardware and software geared toward large enterprises. Companies can now use Ersatz to build and deploy neural networks so they can benefit from recent breakthroughs in the field without hiring a fleet of neural network experts.
Deep learning involves teaching a neural network (a type of supervised machine learning algorithm) to learn from data in a way that is architecturally similar to the brain. Teaching these neural networks—and getting them to work correctly—is difficult, and the people who know how to do it are rare. Tech giants are already investing hundreds of millions of dollars acquiring companies and teams of deep learning researchers as they compete to turn the mountains of data they generate into a durable competitive advantage.
A typical use case starts with a company deciding what kind of problem they want to solve. For example, they might want to predict a number in the future (eg. sales or inventory numbers), a score for how likely it is that something will happen (eg. clicking on an ad or “liking” a song), or the identification of an object in a picture (eg. a face or a pedestrian). After choosing a business specific objective, a company collects historical data related to the problem at hand, uploads this data, and instructs Ersatz to learn from it. After a “training” period during which the neural network “learns” from your data, Ersatz will be able to make predictions about future results based on the examples you showed it.
Behind the scenes, Ersatz uses specialized Graphics Processing Units (GPUs) to crunch numbers up to 40x faster than their CPU-based counterparts. Because video games and learning machines use much of the same types of math to function, GPUs have become practically essential for effective deep learning.
As a cloud-based solution, Ersatz charges companies a per-minute usage fee whenever Ersatz is being “trained”. For companies preferring unlimited usage and greater control, Ersatz offers its deep learning appliance. This solution includes hardware (specialized “GPU” number crunchers), the same Ersatz software as the cloud version, and the Ersatz source code so it can be audited and integrated more deeply into customers’ products.
While neural networks have been in existence since the 60s, a variety of factors have coalesced to put deep learning in a position where it has the opportunity to shape the field of machine learning for many years to come. First, the advent of the internet means there’s more data available than ever before and companies are increasingly using it to grow their businesses. Meanwhile, algorithmic improvements to neural networks (“deep learning”) have drastically improved their performance on various types of problems so they are now state of the art in several domains. Finally, the application of the graphics processing units to neural network training delivered a 40x speedup over alternatives, making neural networks at least 40x as attractive as before.
Machine learning has reached a major inflection point thanks to deep learning. Making this technology available with a platform like Ersatz will open up a whole new class of intelligent software,” said Dave Sullivan, co-founder and CEO, Ersatz Labs. “We’re a long way from the type of machine intelligence we see in movies, but we’re a short way from some truly amazing applications that will only increase the importance and impact of data in our lives. I see neural computing—and deep learning specifically—as a major competitive advantage for companies in the near term. Long term, the technology will become ubiquitous.”
With the launch, Ersatz Labs emerges from a year-long beta during which over 2200 customers signed up to use the platform. Users come from a range of industries: algorithmic traders trying to understand financial markets, medical researchers looking to analyze medical scans, and energy companies looking to analyze massive quantities of seismic data, among others.
Sign up for the free insideBIGDATA newsletter.