Imagine being able to tap into machines that think in a way that’s similar to the human mind. N2Sky is a proposed system that will make computer-simulated neural networks available across multiple cloud providers.
Erich Schikuta and Erwin Mann, computer science faculty at the University of Vienna, have outlined a system (PDF) that provides neural network tools aimed at novice and experienced users, all provided through a cloud made up of different services. This practice known as “Sky Computing” federates separate cloud services to create a single cloud-like environment or “Sky” that can support large applications and make use of different cloud afﬁnities based on types of resources available.
As opposed to the rigid logic of standard computers, neural networks simulate the same process of learning that is thought to go on in the brain, and are being used to perform tasks difficult with traditional computational approaches such as computer vision, speech recognition and artificial intelligence.
Neural networks can now simulate the logic gates laid out by Alan Turing who formed the logical foundation for computers in the 1930s, but instead of reacting in the same rigid ways like in AND, OR and XAND gates, neural nets often contain a “learning rule” that will react to input patterns and modify weights and outcomes.
As such, a neural network is built, but also trained to be used as an analytical tool.
In a way, it could be thought to be learning much like how a child sees a cat for the first time, then as it sees more of them, it eventually recognizes them as cats. A neural network becomes aware of patterns and able to recognize them in useful ways.
N2Sky uses the ViNNSL (Vienna Neural Network Specification Language), an XML based language for describing, training and running neural networks on a grid infrastructure. Schikuta and Mann hope to boost the overall performance of the system by allowing neural network training to happen in parallel, and that capable resources (ie. multi-core systems and clusters) can be automatically found in the cloud.
If N2Sky can bring neural networks to researchers and developers as a an accessible and affordable service, it will help spur more knowledge and excitement around this new form of computing. Also, since labs might be taking wildly different approaches to neural networks, it may could help standardize neural networks, as well as help develop a set of common practices, and community necessary for this technology to take off.
While still in the testing stage, N2Sky is hoping to also provide a search engine for neural nets that can help connect users to nets that have been able to tackle similar problems. In this sense, users will be able to rent a specific neural network trained to solve a specific type of problem.
As the range of hardware and software as a service expands, neural networks will undoubtedly be an interesting segment to follow as more researchers use it to solve puzzles, and more entrepreneurs see opportunities to use it in their new products and services.