MU researcher probing deep learning’s possibilities
Talk to Tony Han, associate professor of computer science at the MU College of Engineering, about deep learning, and it doesn’t take long to figure out how deeply he believes in the method’s possibilities. Big Data, smart cities, biomedical innovations, security — all are areas he said he believes will be heavily impacted by deep learning.
Han currently is on a research sabbatical, working on cutting edge research in the field of deep learning. Take Deep Speech 2 as an example. Han led a team of Silicon Valley researchers that developed Deep Speech 2, a speech recognition system used by Baidu, China’s most popular search engine. The Massachusetts Institute of Technology Technology Review named the system one of the 10 breakthrough technologies of 2016 because of it’s uncanny ability to recognize words despite stutters, irregular speech patterns, accents and more.
“Humans usually can achieve 78 percent accuracy at the sentence level. If you do a committee with the best people worldwide doing this transcription, the sentence level accuracy is about 90 percent,” Han said. “But using our speech engine, it’s 85 percent, which means it’s better than the single best person but just slightly inferior to the global committee.”
Han also is working on using deep learning to help Baidu bring autonomous-driving vehicles closer to market. Deep learning technology helps self-driving vehicles recognize objects and pedestrians to avoid collisions. Han said the goal is to eventually make the technology used in such vehicles affordable enough for more consumers.
“We’re working on autonomous driving, and now we’re actually working in the city environment,” Han said. “The vehicle has the capability of doing pedestrian detection, obstacle detection and can drive in a controlled, simple environment.”
The possibilities for deep learning are vast, Han said, and touch on many areas in which University of Missouri researchers already are doing high-level work.
“I really believe deep learning is the direction and will have an impact on all aspects of computer science,” he said. “People are doing speech recognition, object recognition, natural language processing. … I also think Big Data, smart cities — deep learning can play a big role in these projects.”