CUHK Research: Changing the world

“Few understood computer vision” When Jiaya Jia entered Hong Kong’s job market in 2004 with a PhD focused on computer vision, almost nobody outside the university knew what computer vision could do. “I doubted I could get a decent academic job because it was such a small research field,” he recalls. “Few understood its ability to make real life better.” CUHK was the first with an offer, which (then) Dr Jia speedily accepted. CUHK’s instincts were rewarded. Today, he is a professor with CUHK’s Department of Computer Science and Engineering and acknowledged as a world-leading scientist in computer vision. This is in addition to founding SmartMore i n 2019, a fas t- g rowi ng company fo r manufacturing optimisation and automation (MOA ) , s e r v i ng adva nced i ndu s t r i a l enterprises in economies where human hands are increasingly in short supply. Next generation “smart machines”, Professor Jia notes, have already replaced “stupid traditional machines” in factories across Germany, Japan, the US and mainland China. Operating in the dark 24/7 without a human onsite, their sensors are equivalent to eyes, ears and noses. They have an AI “brain” intelligent enough to monitor production lines, spot defects and make decisions. “AI was not a decent word” Riding off advances in deep learning, computer vision is today part of AI in that it simulates human perception so that a computer can, say, distinguish between images of a cat and a dog. Professor Jia’s discoveries are already in practical use. For example, to create a poster or beautify a photo, a user needs only key in an instruction and the computer does the rest. Some 20 years earlier, “AI was not a decent word,” according to the professor. “It could not achieve any challenging intelligence goal without the power of deep learning. AI looked fake – more science fiction than reality.” AI’s turning point The breakthrough, says Professor Jia, came in 2012 with research into neural networks. Up till then, algorithms connecting hundreds or thousands of computer “neurons” – like synapses in a human brain – mostly produced gibberish. But when Canadian-based scientists began applying algorithms to millions of computer neurons, things suddenly changed. Outputs became ordered and semantically meaningful. Computers could at last tell the difference between a cat and a dog. “After that, we were called AI researchers,” Professor Jia laughs, “because computer vision was not fiction anymore. It was really beginning to work. We could sense it would become powerful, useful, and applicable to a lot of areas. We were confident it could solve real problems in industry.” So began another decade of hard work for Professor Jia and PhD candidates under his supervision. Racing time It is a race against time. “It’s not just that computer engineering is changing fast. The intelligence level of AI is getting higher and higher,” Professor Jia observes. “We need to The intelligence level ofAI is getting higher and higher.We need to keep our eyes very closely on what’s hot in this area if we’re going to catch the train. 12