Jeff Hawkins, the man who brought us the Palm Pilot, is back with a new streaming analytics company that’s now being used by energy-management company EnerNOC to predict the future for the institutions running our electrical grids. Hawkins’s new company, called Numenta, processes data as it streams off sensors, servers and other machines, and then quickly recognizes patterns so it’s able to predict in real time what happens next.
If you visit the web site for Numenta, which was founded in 2005 but just recently emerged from stealth mode, you’ll see lots of images or neurons, synapses and dendrites, and lots of text explaining neurological processes. Don’t be intimidated. The long story short is that Numenta’s software, called Grok, is able to recognize patterns (e.g., temporal and spatial) from streaming data and then automatically build models that allow it to predict what will happen next.
The goal isn’t necessarily to be as intelligent as the human brain, but to be as fast as the human brain when it comes to processing data that Grok understands. People love to talk about “big data,” VP of Marketing Joe Hayashi explained, but “our mission is to help people act on fast data.” In Numenta’s largely machine-to-machine world, where the data half-life might be measured in seconds, the human-driven process of big data is just too slow.
“They can only go as fast as the data scientists can build models and really understand it,” Hayashi said.
Grok, on the other hand, is continuously learning from every new data point that hits the system, and it’s always readjusting its models to account for any changes it sees in the patterns of data. Not only does this help it make predictions faster and more accurately, but it also helps Grok spot anomalies that could cause problems. Ideally, Hayashi said, the software will be part of a machine-to-machine system that makes decisions on its own, in real time, without human intervention.
For a customer such as EnerNOC, which helps energy suppliers operate more efficiently, Grok will help the company’s frequency-reserve service called DemandSMART optimally draw power from customers that are part of the program. Frequency reserve markets rely on a network of customers voluntarily (although for compensation) reducing power usage during peak times in order to ensure grid integrity. Grok could also help EnerNOC predict potential mechanical failures by identifying and flagging behavior it hasn’t seen before, or by discovering patterns that lead to failure.
Actually, Hayashi explained, EnerNOC is a really good example of where Numenta and Grok fit into the data-processing ecosystem. EnerNOC, like many Numenta users, already has a system in place for processing real-time data, but that system only lets the company see what’s happening now. Introducing Grok into the environment, will let them “know what’s going to happen,” he said.
All of Numenta’s detailed comparisons to how the brain works might be an impressive way to describe the technology, but it might also bury the importance of what the company is trying to do. As we’ll discuss at various sessions during our Structure: Data conference in March, the advent of ubiquitous sensors, webscale server farms and just an abundance of machines everywhere is generating more data, and at faster speeds, than human beings could ever hope to make sense of on their own. If we’re going to keep up, we’re going to have to learn to let software shoulder a lot of the analytical load.
Feature image courtesy of Shutterstock user pixeldreams.eu.