Artificial neural networks are software algorithms that model systems of neurons to implement machine recognition and learning. A new service, Ersatz, from Black Cloud BSG provides a cloud-based "deep" neural network service (dubbed Machine Learning as a Service, or MLaaS) that will be available for anyone to use for tasks such as data recognition, clustering, classification, predictive analysis, modeling, and novelty detection.
The thinking behind artificial neural networks goes back to at least 1943 (the McCulloch-Pitts neuron model) and while working networks were created as early as the 1960's it's really only been since the middle of the first decade of the 21st century that breakthroughs in algorithms made neural networks really effective.
Just as important has been the availability of high-performance graphics processing units (GPUs) in the last few years that provide the computational horsepower necessary to run large scale neural networks.
And this is where Ersatz comes in. First announced about one year ago the service is due to come out of beta in a couple of months. Ersatz uses Nvidia GPUs front-ended by Web servers where users can upload their data sets and run them through a variety of neural networks to perform whatever kind of analysis suits their data. Users can also submit data and get results via a simple API.
Currently Ersatz offers four neural network architectures or models: Convolutional neural networks for working with images, recurrent neural networks for time series data, deep feedforward neural networks for general purpose machine learning projects, and autoencoder neural networks for dimensionality reduction.
Dave Sullivan, the CEO of Black Cloud, pointed out that their Ersatz neural networks can function at the scale required by Big Data projects making it possible to apply machine learning and recognition to enormous data sets. He also noted that as new neural network technologies emerge Black Cloud, being experts in the field, will evaluate and incorporate them where there's value to be gained and do so far faster and at a lower cost than most organizations' development groups can achieve.
So, what sort of problems might Ersatz be able to tackle? Consider Pandora's Music Genome Project which uses only expert human effort to classify music according 450 attributes. While Pandora's system works amazingly well given the incredible amount of music available it obviously has scalability problems which could well be addressed using Ersatz. Another intriguing example has been set up as a video demo showing "a use-case for Ersatz centering on Salesforce.com opportunity scoring."
What's also interesting about Ersatz is that the service provides yet another building block for enterprise and business applications to take advantage of at a lower cost and higher implementation speed than could ever be achieved with in-house developers. Moreover, this particular building block is like nothing that's been available before.
You can apply for a beta account which provides 60 minutes of processing for free with additional time priced at $0.41 per minute.