Teaching Deep Learning, One Sore Neck at a Time

Mr DONG Chenxi, Senior Research Assistant in the Department of Mathematics and Information Technology (MIT), received the Grand Prize in the Huawei ICT Competition 2025–2026 Teaching Competition (Industry-Education Collaboration Track, Global Final, Regions Outside Chinese Mainland). He was named one of sixteen Most Valuable Instructors (MVIs) worldwide and invited to present his winning project at the Global Final Presentation, held from 2 to 5 June 2026 at Huawei’s R&D Headquarters at Huawei Xiliubeipo Village (溪流背坡村), Dongguan, Guangdong Province. He also attended the Huawei Most Valuable Instructor Forum during the event.

The Huawei ICT Competition is an annual global event for students and teachers from colleges and universities. It offers an international platform for competition and exchange, enabling participants to build their ICT knowledge, sharpen their practical skills, and experiment with the latest technologies and platforms.

In this FLASS FORWARD’s interview, the awardee explains his award-winning project, what inspired him to enter the competition, and his teaching aspirations. He also shares his experience of taking part in the Global Final Presentation.

A1: One reason comes from my own learning experience. While studying IT courses during my master’s degree, subjects such as neural networks and machine learning were taught mainly through theory. A theoretical foundation is important, of course. But students often find it difficult to connect theory with real-world applications when knowledge is taught without practical, realistic examples, and they eventually get lost in these subjects.

To make teaching more effective, the course I designed for the competition uses real tasks, real data, and a cloud-based approach to shape its workflow and content. This way, students can better understand why the technical aspects of these subjects matter and how they are applied in reality. It is also closer to what students experience in the workplace.

A2: The Huawei ICT Teaching Competition asks participants to design a short course demo, usually around one to two hours, to teach a specific ICT (Information and Communication Technology) topic using Huawei technologies or products. For my entry, I designed a course called “Teaching Deep Learning with Huawei Cloud ModelArts: Scaffolded Authentic Learning on Real Clinical Data with GenAI-Supported Reflection”. Huawei Cloud ModelArts is the platform on which the organiser required participants to run their entry code.

The main objective was to teach students how to use a neural network, a machine learning model inspired by the human brain, for medical image classification. In my design, students used Huawei Cloud to fine-tune a neural network on a real dataset of cervical spine X-ray images.

A3: Instead of relying on simplified “dummy” datasets, my course challenges learners to build a complete Deep Learning pipeline on Huawei Cloud ModelArts using raw, imbalanced medical data. It uses an open dataset of real clinical cervical spine X-ray images, and teaches students to train neural networks to screen those images and automatically detect abnormal curvatures across four categories: Lordotic (healthy), Straight, Sigmoid, and Kyphotic.

The course has several advantages over those using fake data. First, it is more authentic, since it uses noisy clinical X-ray images rather than clean academic data. Many IT and AI students suffer from abnormal cervical curvature, also known as “Tech Neck”, and neck pain after long hours in front of screens. Because the course speaks directly to their own health concerns, they are motivated to commit to the learning.

It also uses distributed cloud GPUs instead of local CPUs, which improves learners’ proficiency with the cloud. And it requires learners to treat generative AI strictly as a Socratic reflective partner, not a code generator, which improves their AI literacy.

A4: Yes, I think so. I often feel there is a gap between what students learn at university and what industry actually wants. Because of that gap, people have to learn a lot on the job that was never covered during their studies. Building a real tool like Huawei Cloud into the course design will improve students’ ability to handle practical, real-life tasks. It makes them more competitive in the job market and helps them find work more easily.

A5: To test whether my idea works, I prepared a short teaching video of around 20 minutes with matching lab code, and invited some undergraduate students to try the activity online.

The initial feedback was encouraging. Some students said that using real data and a real cloud environment made them more motivated and the learning more meaningful. They also learnt to use GenAI as a reflective assistant. For example, they asked it how to improve model performance, critically reviewed its suggestions, modified part of the lab code, reran the experiment, and reflected on the results.

A6: A key point of my design is that GenAI should not be used simply to generate answers. Instead, I framed it as a reflective partner. Students still need to understand the model, make decisions, test their changes, and judge whether the results actually improve. I think this is close to how AI will be used in future workplaces. Using GenAI responsibly to work more efficiently will, I believe, become an essential skill.

A7: On the morning of 3 June, I visited the Huawei Shannon Exhibition Hall in Shenzhen. There I saw Huawei’s latest technology up close: AI data centres, the Ascend series of AI computing GPUs, and real-world AI agent applications. To be honest, I was amazed by how much AI has already been put into practice. I think universities in both the Chinese Mainland and Hong Kong will see similar AI infrastructure upgrades in future. Right now, the corporate world is clearly ahead of academia in applying AI.

In the afternoon, I visited Huawei R&D Headquarters at Huawei Xiliubeipo Village (溪流背坡村), beside Songshan Lake in Dongguan. Around 30,000 R&D staff work in this huge village, often called the Huawei Ox Horn Campus. It is so large that staff use a red tram system to travel between locations. The campus recreates the architecture of 12 European cities, including Heidelberg, Verona, Bologna, and Bruges, with castles, palaces, and other European-style buildings housing Huawei's research centres. It is a sign of the tech giant’s enormous commitment to technological development.

A8: Exchanging views with Most Valuable Instructors from around the world was a valuable experience. They are visionary in leading change in their own fields in the face of the AI revolution. Professors from the Chinese Mainland shared their work on tech and AI localisation, such as writing textbooks for domestic (Chinese-developed) programming languages and building courses optimised for Chinese-made AI chips. I also saw university projects exploring AI-teacher-student-administrator interaction systems.

But what really inspired me came from Mr Isah Abdulazeez, the MVI from Nasarawa State University in Nigeria.[TSM[1] To give his students practical, hands-on training rather than asking them to memorise theory, he went all out to earn eight to ten industry engineer certifications despite his limited resources. His passion for education was striking. It made me realise that, as AI educators, we should make sure our teaching meets industry needs, and always equip ourselves with the knowledge the industry needs.

A9: My panel emphasised that, to develop students’ ability to apply theoretical knowledge to real-life problems, project-based assignments should be introduced as early as possible to undergraduate studies and used throughout. We also recommended that universities collaborate with both local and international industry partners to give students access to the latest technologies used in the industry, such as high-performance computing hardware. This will keep students in step with current industry practice.

Everyone on my panel agreed that AI tools shouldn’t replace students’ need to understand core concepts, and that educators must therefore reinforce foundational theory rather than weaken it. We also agreed that AI should be a collaborator and adviser, not a decision-maker, and that teachers need to define clearly how AI is used in the classroom.

One of the panellists, Dr Shafika Isaacs of UNESCO, pointed out that critical thinking and checking the accuracy of AI’s suggestions matter more than ever. She stressed that in the AI era, collaboration matters more than the final result. This means we need better ways to assess how students use AI, not just marks for what they produce.

A10: At the presentation, I saw one MVI-affiliated university using an AI agent to manage the entire teaching process, covering data collection, automated student records, and smart dashboards.

Students on a programming course there upload the code from their assignments to the AI-powered platform. The agent then runs the code, checks whether the result meets the requirements, and reviews its own work. It can even break performance assessment into separate components: code robustness, innovation, algorithms and data structures, coding standards, and team contribution. Finally, it compiles all this data into a personalised dashboard for each student. It handles these complex tasks on its own; teachers only need to oversee the process from start to finish.

This shows that AI has already become part of almost every stage of teaching and learning, and it’s delivering impressive results. I truly believe AI will play an even more central role in education. Educators really need to build up their AI skills to keep pace with this fast-moving technology. Only by doing so can we collaborate effectively with AI in future classrooms.

A11: The professors at EdUHK demonstrate a genuine commitment to students’ overall well-being and career development. I am especially grateful to Dr Alpha Ling Man-ho, the acting head of the Department of Mathematics and Information Technology. After completing a Master of Arts in Mathematics and Pedagogy at this University (2019-20), I wanted to pursue a postgraduate programme in AI and data science in Hong Kong. Dr Ling thoughtfully recommended several universities and programmes, and generously wrote letters of recommendation for me. I went on to enrol in a Master of Science in Data Science at the City University of Hong Kong (2020-21) and began my career in AI and data science research. I came back to EdUHK in 2021 and since then have been working in Professor Li Wai-keung’s research team on projects involving educational data analytics.

The energy of EdUHK’s academic environment, and the passion the professors on the team have for educational research and student learning, made a real impression on me, and gave me a strong interest in joining the Huawei teaching design competition. In addition to Dr Ling and Professor Li, I am very grateful to Professor Philip Yu Leung-ho from MIT department for his detailed advice on pedagogical design, including how to better explain the role of GenAI and how to develop my competition course into a lower-code version for non-IT students in future.

Note: Mr Dong’s entire short course material (video, ppt) has been uploaded to Huawei Talent webpage. Click here to review it.