Watch scientists from Google, MIT, Stanford and Spotify explain how these self-teaching algorithms power our world. Here, you can see their recent presentations at the RE.WORK Deep Learning Summit
You’ve heard of deep learning – the algorithms behind speech recognition, computer vision and machine translation. The field is very hot right now, even though anyone who’s tried the latest transcription software or image recognition app realizes it still has a long way to go.
If you want to know how deep learning works and where it’s going, you need look no further than the RE.WORK Deep Learning Summit, which took place in San Francisco. The gathering featured top algorithm developers and researchers from the likes of Google, MIT, Flickr, Spotify, Emotient, Facebook and Stanford to explain when and how they are using deep algorithms. These weren’t so much the rock-star CEOs and headliners you’ll find at TechCrunch Disrupt or Web Summit. Rather, they were the actual scientists in the field explaining their work.
You can check out videos from the conference at their YouTube channel.
Among the highlights were Google Brain scientist Quoc Le discussing how deep learning can be used to understand texts without much prior knowledge. In particular, he explained how algorithms can learn the vector representations of words, and use these vector representations to translate unknown words between languages. These vector representations preserve the semantics of sentences and documents and therefore can be used for machine translation, text classification, information retrieval and sentiment analysis.
Roberto Pieraccini, Director of Advanced Conversational Technologies at Jibo, discussed the latest advances in speech recognition while Dr. Joshua Susskind, Co-Founder and Senior Data Scientist at Emotient, discussed recent developments in facial recognition and expression analysis.
One of the highlights of the conference was a fireside chat with Andre Ng, the former head of the Google Brain project and now chief scientist at Baidu. Ng was bearish in describing recent advances in machine learning and said that these algorithms are more of a “cartoon version of how the brain works” than an actual simulation of the brain.
When asked how close artificial intelligence researchers are to a “single algorithm of the brain,” he said “we have no idea how the brain really works.” He also admitted that “there is no killer app for computer vision yet.”
With these caveats, he says, deep learning algorithms can do a lot.
Baidu, for instance, uses the algorithms for image services. For instance, if you see someone’s sweater, you can snap a photo and Baidu will tell you where to buy it. They also use deep learning to power their advertising, as well as for speech recognition. Engineers within Baidu are using the deep learning algorithms to predict when hard disks will fail.
Ng said Baidu’s “holy grail” for deep learning is to get better and better at learning to map XY pairs — or what is known as “supervised learning.” Eventually, for instance, you will be able to input an audio clip into an algorithm and it will output a transcript.
“Baidu has made enormous progress in speech recognition,” Ng said. “Imagine if all of you could talk to your cell phones, even in noisy environments, even in cars and say, ‘Hey cell phone, please text my wife—let her know I’ll be five minutes late.’”
“It will totally change the way we interact with technology.”
Ng said he has no patience for predictions that evil robots will take over the world. He says that these predications are actually a distraction from what he sees as the real potential threat of technology and artificial intelligence: the impact of technology on labor.
“The United States took about 200 years to move from 98 percent farming to today, less than 2 percent. What that meant the descendants of farmers had to do something else. This time the transformation will be much faster. Let’s say we succeed in building self-driving cars. That’s 3.5 million truck drivers that might have to find different employment and our education system has historically found it difficult to retrain people who are already alive today to do a different job.”
Ng, who in the past was an executive of Coursera, says that he is concerned that even Coursera might not be up to the task of retraining so many people.
“This is a serious issue and government and business leaders need to have a serious discussion about it.”
Watch Andrew Ng and other videos from the conference here.