How “deep learning” works is still a mystery, Microsoft Research head said

Peter Lee SEATTLE – Even researchers working in artificial intelligence sometimes don’t understand exactly why advancements happen. For instance, so-called deep learning, which is a kind of machine learning, often has the benefit of transfer learning, said Peter Lee, global head of Microsoft Research speaking on Thursday at the Geekwire Summit. But how and when that works isn’t entirely clear.  Here’s how that has worked at Microsoft. When Microsoft was developing Skype Translator, which translates speech into another language in essentially real-time, it started by training the system to understand English, he said. “We take that and train it on Mandarin and magically the English improves and the Chinese trains faster,” he said. “Then we throw French at it and the Chinese and English improve.” If that same type of advancement starts…


Link to Full Article: How “deep learning” works is still a mystery, Microsoft Research head said