Entries with tag neural network.

A neural network that Google developed has taught itself to recognize cats. Google X Lab researchers created a deep-learning artificial brain of 16,000 computer processors with machine-learning capabilities, then provided it with thumbnail images randomly extracted from 10 million YouTube videos. The researchers claim they did not prompt the neural network on what to look for in training and say the system conceptualized the cat on its own. According to the New York Times, it assembled a “digital image of a cat by employing a hierarchy of memory locations to successively cull out general features after being exposed to millions of images.” The scientists said it appeared they had developed a cybernetic cousin to what takes place in the brain’s visual cortex. The work also debunked the widely held concept that images must be labeled to train a face detector, an important issue in the field of photo detection and recognition. The researchers are scheduled to present their work later this week in Scotland at the International Conference on Machine Learning. (The Telegraph)(Wired UK)(The New York Times)

New self-learning software system from Siemens uses neural-network technology to help grid operators more accurately forecast the electrical output from renewable energy sources. The researchers say their forecast over a 72-hour period was more than 90 percent accurate. Grid operators can use this type of information to calculate network power demand, which can help them more accurately place advance orders with suppliers for additional electricity. The Siemens algorithm calculates projected transfer losses directly from electricity-consumption forecasts. Switzerland’s Swissgrid national electricity-transmission company estimates using the software will save it about 200,000 euros per year. (PhysOrg)(Siemens)