Google’s letting the cash flow. Fresh off its $3.2 billion acquisition of “conscious home” company Nest, which makes the Nest Learning Thermostat and Protect smoke and carbon monoxide detector, it’s spending some comparative pocket change -- $400 million – on artificial intelligence startup DeepMind Technologies.
The news was first reported at re/code here, where one source describes DeepMind as “the last large independent company with a strong focus on artificial intelligence.” The London startup, funded by Founders Fund, was founded by Demis Hassabis, Shane Legg and Mustafa Suleyman, with the stated goal of combining machine learning techniques and neuroscience to build powerful general purpose learning algorithms.
Its web page notes that its first commercial applications are in simulations, e-commerce and games, and this posting for a part-time paid computer science internship from this past summer casts it as “a world-class machine learning research company that specializes in developing cutting edge algorithms to power massively disruptive new consumer products.”
According to this article at RobotEnomics, the machine learning DeepMind specializes in is reinforcement learning “to solve high-dimensional uncertain sequential decision-making problems,” providing Google with the means “to improve on predicted modeling and provide a solution to reduce the amount of human intervention and enhance decision making.” For more on the topic of reinforcement learning, head here for a DeepMind Technologies-authored paper on “Playing Atari with Deep Reinforcement Learning.”
It presents what is described by the authors as the first deep learning model to successfully learn to control policies directly from high-dimensional sensory input using reinforcement learning.” It cites that it applied its approach to various Atari 2600 games with the goal of creating a single neural network agent able to successfully learn to play as many of the games as possible – learning as a human would, from nothing but the video input, the reward and terminal signals, and the set of possible actions. It performed all previous reinforcement learning algorithms on six of the seven games attempted, and surpassed a human expert player on three of them, the paper notes.
Google, according to this article in the MIT Technology Review, “has become a magnet for deep learning and related AI talent,” citing as one example the vendor’s purchase of Prof. Geoffrey Hinton’s University of Toronto deep neural network startup DNNresearch in 2012, and plans to apply its tech to image recognition, search and natural language understanding. It also notes that deep learning has improved voice search on Android devices.
Its continuing acquisition of deep learning smarts and talent in the area should come as no surprise. This article discusses that Google is in a deep learning arms race with Facebook, Microsoft and other tech giants to get its hands on the relatively few researchers who are experts in the field of teaching computers to see, hear and learn as mere mortals do.