In the world of computer science, A.I. (artificial intelligence) is one of the hottest topics as it is closely connected to our daily life today, while machine learning – building a model to learn from data to make predictions and decisions without explicitly programmed – is an indispensable part in A.I. technology.
Joined HKBU as an Assistant Professor of Department of Computer Science in late 2019, young machine learning researcher Dr. Bo Han has been bringing his expertise in trustworthy machine learning to the faculty and conducting a series of impactful research projects under the support of RGC Early CAREER Scheme, NSFC Young Scientists Fund, and University and Departmental Start-up Fund.
Holding a visiting position at RIKEN Center for Advanced Intelligence Project (RIKEN AIP), a top-tier machine learning centre in Asia, Dr. Han says it was Professor Andrew Ng – one of the most influential experts in the field of machine learning – opened his eyes to machine learning.
“It was around 2012, I took an open source course delivered by Professor Andrew Ng from Stanford University, and learned about machine learning from the course. I was very excited after the course, and I wanted to be a researcher and build my career on machine learning,” he recalls his time being a master’s student at Ocean University of China. “At that time, A.I. was not that hot yet. Now, everybody talks about A.I.”
Delved into machine learning in crowdsourcing research when he was a PhD student at University of Technology Sydney, Dr. Han says, like many young researchers, he struggled through some gloomy time when he first started in the research field. “After two, three years [of research], I had some minor results. But I was rejected by all the top-tier ML/AI conferences continuously. It was a very hardship time for me. It’s like I couldn’t see a light in the dark. In my third year as a PhD student, I only had one tier-two conference paper published. I was really suffering,” he looks back on the early days of his researcher career.
A shed of light came through when he was offered an internship with RIKEN AIP in 2018. “My mentor in RIKEN AIP, Masashi Sugiyama, who is very famous in machine learning, said to me, ‘your previous background in crowdsourcing is very similar to label noise problem’,” he says. He then shifted his focus to weakly supervised learning and developed some algorithms to handle label noise issues during his internship, working with Masashi Sugiyama, Gang Niu and Mingyuan Zhou. He also published two fundamental papers, Co-teaching: Robust Training of Deep Neural Networks with Extremely Noisy Labels and Masking: A New Perspective of Noisy Supervision, which got over 550 and 110 Google citations in three years respectively.
In 2019, he got the opportunity to be a postdoctoral researcher at RIKEN AIP, working with Masashi Sugiyama, he thought it was time to move on to something else and started his exploration of adversarial learning. “I hope to do something else to push my knowledge boundary,” says Dr. Han, who received the RIKEN BAIHO Award (2019), RGC Early Career Scheme (2020) and NSFC Young Scientists Fund (2020). “In fact, my research transition comes very naturally in these several years. I just use my domain knowledge to catch up with the trend and I’m still evolving.”
As a promising researcher, Dr. Han aspires to create greater impact to the industry and society with his machine learning research efforts and results. One of his ways to achieve that is sharing his weakly supervised learning algorithms online. “I open source a lot of codes online and my papers are getting citations, which means that I’ve already made an impact to the academic society,” he says. “I hope to continuously open source my research codes, so all the young and junior students can learn that research can be reproducible. That’s very important. The reason why computer science and machine learning can be developed so quickly is because most of the researchers open source their codes and contribute to the computer science society.”
Dr. Han also collaborates with companies such as Alibaba and Tencent to work on his federated learning and graph learning projects. For example, Device-Cloud Collaborative Learning for Recommendation, one of his works collaborating with Alibaba Research, was published in top-tier data science conference KDD2021 recently. “Federated learning is growingly important nowadays. Its high level concept is that you don’t need to distribute the data to a centralized server, instead the data is kept on your mobile phone and that would preserve user’s privacy,” he says.
In the near future, Dr. Han would like to focus more on causal representation learning well as trustworthy machine learning. “As a young researcher, I can’t just stay in my comfort zone. I can’t stay in the label-noise and adversarial area for my whole life. I need to delve into more new areas, such as causal representation learning” he says.
Now juggling his research and teaching with his family life, the dedicated researcher says he hopes he could have 48 hours a day. Meanwhile, he is grateful for all the helps and guidance from the faculty members and supporting staff in the Department of Computer Science at HKBU. “I truly appreciate the research atmosphere here, and HKBU Department of Computer Science is like a warm family” says Dr. Han. “I hope more young researchers can join us and work with us together. All the senior professors here are very nice to help your research, teaching and guiding you to build up your career.”
