Saffron Gets $7M to Build Brain-Like Learning Machine
Machine learning is all the rage today in the analytics space, but it’s not the right big data tool for all circumstances. A company called Saffron Technology today announced it received a $7 million investment to continue building its big data analytic software that mimics how the human brain learns.
Saffron Technology was founded 14 years ago by two former IBMers, Manny Aparicio and Jim Fleming, with the idea of creating a natural learning platform that worked like the human brain. This was a tall task back then, and it’s still a tall task, because nobody knows exactly how the human brain works.
“What sits on your shoulders is the most complex object in the known universe,” theoretical physicist Michio Kaku recently said on The Daily Show with Jon Stewart. “We would have to build a computer the size of a city block, cooled by a river, and energized by a nuclear power plant to [recreate] the computational power of your brain, which does it for 20 watts.”
Saffron‘s software doesn’t require a computer the size of a block, nor a nuclear power plant. In fact, software, which it calls the Natural Intelligence Platform, runs on standard “pizza box” commodity servers, CEO Gayle Sheppard recently explained to Datanami.
“Our main focus has been figuring out how to bring to big data the same capabilities that the human mind has, and that means real-time learning and pattern recognition and the automatic association of ideas that we naturally do as humans,” she says. “I don’t want to leave you with the impression that we have modeled the human brain. We use an aspect of how we learn and how we reason, and we mimic that in our technology.”
A Modern Memex Machine
Aparicio and Fleming looked to the research of Vannevar Bush–the 20th century American inventor who was the founder of defense giant Raytheon and the head of the U.S. Office of Scientific Research and Development (OSRD) during World War II–on which to model Saffron’s technology.
One of Bush’s inventions was something called memex machine, what’s described as a “proto-hypertex” system that would closely mimic the associative processes of the human mind. The invention (in some ways a precursor to hypertext processing on the Web) would have combined electromechanical controls and microfilm cameras and readers, and was meant to imbue humans with the power of permanent recollection.
Saffron’s offering, called the Natural Intelligence Platform, creates something like Bush’s memex machine, but instead of micro-mechanical devices and microfilm, it uses advanced algorithms and streams of structured and unstructured data.
The Natural Intelligence Platform is composed of several layers, including data ingestion tools, the core associative graph database, an API layer, and a visualization layer. The data ingestion layer extracts meaningful information from data, which could be anything from news stories and GNIP‘s Twitter fire hose to phone calls and emails. Depending on the industry, natural language processing (NLP) algorithms or statistical routines written in R might be called on to help prepare the data for processing. Hadoop can be used to store data before and after processing.
The heart of the company’s intellectual property is an associate graph database, called the MemoryBase, that processes data on three types of entities–people, places, and things. This database is geared toward helping users find meaningful patterns in the data, including who’s connected to who, why, when, and where, across time and space. This approach is modeled on the associative nature of human intelligence, notably the ability spot patterns in large amounts of seemingly unrelated data.
“What Saffron does is create an associative index over the data. We’re not storing the data. We’re reading and learning about what’s in the data and creating a big hyper graph of associations at the entity level,” Sheppard says. “From those patterns, we can begin immediately using the knowledge that’s being created to do similarity analysis, to look at emergent or convergent trends, to do temporal trending of associations that are moving up and down together.”
Real World Use Cases
Saffron’s technology has been implemented about two dozen times in the defense, energy, healthcare, and manufacturing fields. Depending on the industry, the data types and initial processing algorithms will change. But once the data is loaded into the MemoryBase, it’s all about measuring the distance between points on a graph.
|A depiction of Vannevar Bush’s memex machine|
“A lot of our users are line of business people, and they’re trying to make sense of the world. Sometimes it’s geopolitical risk or competitive risk or social risk,” Sheppard says. “In some particular case, we’re helping them anticipate what’s going to happen next or where something might be needed next, so they can act on that and prepare. And ultimately, as new data arrives, we’re always learning.”
One law enforcement organization has successfully used Saffron to help put a dent in drug trafficking in South Florida. The organization (which Saffron asked not be named) loads data from more than 80 different sources into the database, and then relies on the database to make connections among people, places, and things.
“We learn about people’s interest, their association, the cartels they belong to, the submarines they own, the planes they fly, the high speed boats they use, the locations and drops they choose,” Sheppard says. The database can make connections among many entities in a matter of seconds, as opposed to the days or weeks it would require to manually make those connections. This approach also shortens the training time required to bring new intelligence officers up to speed, because the knowledge is kept in the machine.
The software is also used to help track incidents that occur at nuclear powered electricity generation stations around the country. There is a plethora of information to be gleaned from the multitude of event reports, which are written in a highly technical language. Once loaded into Saffron, the technology can detect patterns that may be buried in the data. “We learn about the events that have occurred to help predict that events will occur unless we take action,” Sheppard says. “It’s a very proactive approach to managing these plants.”
Saffron’s empirical approach is also handy to counter-terrorism groups, who use the technology to help spot potential threats to national security. “As we learn about people, we’re getting the context of what they’re talking about and we’re able to score whether or not the context of the communications is in fact a possible high threat or simply just chatter,” Sheppard says. “We can establish very quickly what’s a real threat and what’s just noise, without rules or models.”
Moving to Silicon Valley
The company will use the $7 million in Series B funding announced today in part to fund a move of its headquarters from Cary, North Carolina to Silicon Valley. The company plans to keep its Cary location as an R&D center, while its new HQ in Northern California will help it emerge from “stealth mode” and elevate its profile in the rapidly moving big data analytics space.
The company also plans to use the funding to fuel its sales and marketing initiatives, and to bolster R&D. Among the projects in the works is looking for a way to enable Saffron to perform unsupervised learning, which will help the product detect unknown unknowns. Currently, you need to have some idea of what you’re looking for, although you don’t necessarily need a theoretical model, as the machine learning approach requires.
It’s all about helping people, Sheppard says. “We’re not replacing humans. We’re just making those humans smarter in the decisions they make, allowing them to make decisions more accurately and faster than they were able to before,” she says.