Deep learning just may be the buzziest, most misunderstood term in the tech world. And October was a banner month for the enigmatic subject.
The term “deep learning” is, essentially, a synonym for a type of neural network approach. And though voguish at the moment, the goal of the field has long been to create a system that can independently learn. Now, research into machine learning is rapidly accelerating, and those intellectual seeds are bearing widely applicable fruits. Artificial intelligence and deep learning startups are regularly popping up, with AI juggernaut Yoshua Bengio launching a Silicon Valley-style incubator dedicated to the field just last week in Montreal.
But the ways in which deep learning systems and technology are being leveraged are beginning to diverge considerably
Last month, DeepMind, the Google-owned, London-based artificial intelligence and machine learning firm, unveiled a new system, known as a differential neural computer, or DNC, which can independently learn from its own experiences. The program blends neural processing with data storage, allowing systems to individually solve tasks such as expanding a family tree, playing vintage video games, or figuring out an ideal subway route. The system’s ability to solve a task differs from other applications because it deduces solutions by learning from millions of examples — not just pre-programmed data. The asterisk here is that the solutions become increasingly accurate as the program merges memory with complex neural processing.
“We hope that DNCs provide both a new tool for computer science and a new metaphor for cognitive science and neuroscience: here is a learning machine that, without prior programming, can organise information into connected facts and use those facts to solve problems,” DeepMind researchers Alexander Graves and Greg Wayne wrote in a blog post explaining the technology.
DeepMind’s announcement came on the heels of the firm’s other great accomplishment of 2016: having a machine handily beat a human champion in the board game Go.
But advances in the field aren’t coming just from big players like Google. Igal Raichelgauz, for one, has taken a slightly different approach to deep learning with his AI imaging firm, Cortica, based in New York and Tel Aviv.
Founded in 2007, the bedrock of Raichelgauz’s company, which he co-founded with researchers Karina Odinaev and Yehoshua Zeevi, is tethered to research he conducted on cortical tissue taken from a rat’s brain. (Yes, like the rodent.)
Raichelgauz is confident his software system, which is based on the anatomical functions of the brain, is more readily applicable to real-world scenarios than those of other deep learning ventures, including DeepMind. Those applications include self-driving cars, medical imaging technology, and personal image recognition, the latter of which has soaked up much of the company’s consumer resources. Proposed uses for Cortica’s technology include integrating advertising into images, using mobile devices to identify objects in real time, and organizing photos in a “human way,” according to a user’s specific interests, Raichelgauz said.
“With DeepMind, the initial focus was on gaming problems, where the rules of the game are well defined — first with computer games and then Go,” he said. “It’s very easy to define a game just in a few rules, but when it comes to images…it becomes difficult to define what are those rules.”
He added, “I think the biggest difference between Cortica and the deep learning or machine learning approach is the ability to learn in an unsupervised way. That’s the main advantage.”
And Raichelgauz isn’t alone in his confidence in image recognition systems based in neurobiology. Companies like San Francisco-based Vicarious have developed related technologies, and, at least according to Web sources, raised more money. Vicarious has reported raising about $70 million in VC funding, while Cortica’s number is around $40 million, according to CrunchBase. (Raichelgauz said his company has raked in far more than $40 million, but he declined to specify exactly how much.)
Cortica’s significant investors have included Horizons Ventures, Lanta Capital Holdings, Mail.Ru Group, and Ynon Kreiz, according to the company’s CrunchBase profile. Meanwhile, Facebook’s Mark Zuckerberg, Amazon’s Jeff Bezos, and Tesla’s Elon Musk have each invested in Vicarious, according to the company’s website.
Also in New York, Union Square Ventures-backed Clarifai, founded in 2013, is rolling out similar video and image recognition technology. The firm announced it closed a $30 million Series B round in late October, led by Menlo Ventures, bringing its total amount raised to about $40 million.
Raichelgauz said Cortica could raise another round of funding in 2017, and is planning to double down on consumer product investments. The firm already has outposts in Beijing and New York and is eyeing the expansion of another location in the Silicon Valley area next year.