Google may have recently lost artificial-intelligence luminary Andrew Ng to Chinese search company Baidu, but that hasn’t stopped the Mountain View, Calif.-based company from ratcheting up the wisdom of its computers.
Baidu and Microsoft have made progress on deep learning, which involves training systems called artificial neural networks on information derived from audio, images, and other inputs, then presenting the systems with new information and receiving inferences about it in response.
But today Google announced it had won first place in the “classification and detection portion with additional training data” category of the ImageNet large-scale visual recognition challenge — an annual academic competition for the best computer vision system for analyzing images based on a given data set.
According to Google software engineer Christian Szegedy, in the last year, Google has doubled the quality of its ability to apply the right labels to images, as well as correctly spot many items in the images. And deep learning, as the cognoscenti know, can be applied to computer vision, as well as text processing and voice recognition.
Szegedy notes that the team behind the system went by the name GoogLeNet, in honor of researcher Yann LeCun, who popularized convolutional neural networks that represent the foundation of the Google system. (LeCun recently joined Facebook.)
It’s not shocking that Google has managed to better its systems, given that most people in the deep-learning community work for Google. And Google continues to its reign in terms of talent acquisition by way of its many deep learning acquisitions, including DeepMind, DNNresearch, and most recently Jetpac.
The question worth pondering now is what Google will do with its all its deep-learning smarts. The tech giant could point its smarter-than-ever system at popular services like Google image search, Google video chat, Google Drive, and YouTube videos with an eye toward improvement to edge out competitors. Then again, Google could expose its artificial brain for developers to use, like other services running on Google’s computing infrastructure.
But at least it’s clear Google hasn’t given up on deep learning.