Category: Applied A.I.

Stepping Into the Past With Professor Chen Yuqi

In Isaac Asimov’s 1951 novel Foundation, the mathematician Hari Seldon coins the term “psychohistory” to describe his use of mathematics and statistical analysis to accurately predict future trends.

It’s an idea fit for science fiction – but it’s also been a consistent source of inspiration for University of Hong Kong Professor Chen Yuqi, a member of the Faculty of Arts and the Centre for Quantitative History, who’s using artificial intelligence (AI) tools to unlock the past.

Professor Chen, who graduated with a BA and PhD from Peking University before joining HKU in 2025, is at the forefront of history’s AI revolution.

During her PhD, she collected and built a database of 20,000 bronzes with inscriptions, which she used to reconstruct social networks in early Chinese history. While on a yearlong exchange at Harvard University in the United States, she worked with an interdisciplinary team collecting and organising a database of Chinese biographies. In turn, that sparked an interest in the historic roots of different psychologies across cultures – think collectivist versus individualistic – and historical psychology more broadly.

Those experiences – plus a formative internship at Tencent – gave Professor Chen valuable experience in coding. It also set her on a different path from many of her colleagues, as she immersed herself in the field of digital humanities, just as the advent of new AI tools opened up exciting possibilities for the study of history.

Now, she’s exploring these possibilities at HKU: everything from the psychology of ancient Chinese scholars to reconstructing the rise and fall of clans from their own histories. And on top of it all, she’s found time to release an AI-native game that builds on ideas pioneered at Stanford to immerse players in the world of the late Northern Song dynasty (960–1127 CE).

“I saw people experimenting with how AI communicates with AI,” Professor Chen says. “And as a historian, I thought: What would happen if we used different AIs to roleplay different historical figures?”

From psychohistory to historical psychology

Let’s say you want to survey a group to learn its views on a given topic – academics on individualism and collectivism, for example, or parents and children on filial piety and family values. How would you go about it?

The simplest answer is just to design a questionnaire and ask them. But what if you wanted to do the same thing for a group of people who have all been dead for hundreds of years, if not longer?

That’s where AI comes in, Professor Chen says. Using AI models finetuned on historical texts, she can trace psychological and ideological trends across millennia of history.

What would happen if we used different AIs to roleplay different historical figures?

– Professor Chen

The key – and the primary challenge – is verification. The AI models learn how ancient writers think, then she tests their training by looking at verifiable historical events such as the controversial Wang Anshi reforms that divided the Northern Song court almost until its collapse.

Almost every contemporary writer staked out an opinion on Wang’s policies, meaning it offers a perfect litmus test for the AI’s understanding of their beliefs. If the model returns correct predictions, then Professor Chen can use it to study how other attitudes may have changed over time.

For instance, Professor Chen’s models offer insight into how conservative/reformist tendencies predict scholar attitudes toward historical events, and how their relative power waxed and waned over time. They also let her see how core values, such as filial piety, are shaped by political and economic upheaval.

“In the past we couldn’t get a psychological values variable, because it’s very hard to measure from historical texts,” Professor Chen explains. “We can’t ask dead people to come to the lab and fill out questionnaires.”

Now she doesn’t have to.

Game theory

More recently, Professor Chen has been putting her time into a very different kind of project: an interactive historical simulation that lets students experience the past for themselves.

That’s the premise of “Into the Painting,” which lets players choose from hundreds of characters featured in the Song Dynasty masterpiece “Along the River During the Qingming Festival,” then play out their lives in the waning days of the Northern Song.

That’s actually a marked departure from Professor Chen’s original idea: Allowing users to play as great historical figures, from Qin Shihuang to Napoleon.

The problem, as she explains it, was that the game worked by changing history – limiting its potential as a teaching tool.

So, she went back to the drawing board in late 2025, spending six months designing, coding, and testing “Into the Painting.” The game ducks its predecessor’s issues by instead putting players in the shoes of ordinary Chinese, allowing them to change their lots in life while still remaining subject to broader historical forces.

It was a monumental undertaking. To date, Professor Chen has tagged over 200 figures in the painting, giving each a name and unique backstory in line with their depiction. Then she used a cutting-edge multi-agent AI approach to realise real-time simulation and create unique branching storylines, allowing players to navigate their characters through the Northern Song’s turbulent final years. Succeed, and you can play again as another character. Fail, and your storyline ends.

“Too much of history is about elites,” she says. “99% of individuals are absent from history. They aren’t the main characters, but they are important.”

AI-Driven

Professor Chen says it was important that “Into the Painting” be not just AI-powered – referring to the use of AI tools in the creation process – but also AI-driven.

That means the storylines are truly branching, with nearly limitless options available to players depending on their choices. In this, “Into the Painting” is a genuine step forward from previous experiments like “Stanford Town,” which saw Stanford researchers populate a town full of generative AIs to study how they would interact.

Professor Chen experimented with similar ideas, creating a sandbox to simulate debates among various Chinese literati.

99% of individuals are absent from history. They aren’t the main characters, but they are important.

– Professor Chen

But she harboured dreams of something bigger; even she was frustrated with the model’s limitations. “One thing about agent-based modelling is that you can’t always interact with it,” Professor Chen explains. “You can only observe.”

Her desire to change this paradigm led her to develop FISH – a Framework for the Interactive Simulation of History. Combining her previous experience at Tencent and her research, she wanted to push the boundaries of AI-based modelling to allow players to interact with the models, not just observe them.

“I don’t want ‘Into the Painting’ to just be a single, narrow game,” she says. “I want to develop it into a framework, creating general principles that anyone can use to develop interactive simulations of historical periods.”

“The thing about AI-driven games is that they are truly unlimited,” she adds. “You can have hundreds of different endings. It’s all open.”

May 22, 2026
How AI Is Rewriting the Rules of Art Conservation

“It opens up all these new horizons for art history, for connoisseurship, and for how the discipline is going to continue to form.”
— Professor Marc Walton

You’ve heard of using AI to make art, but an interdisciplinary team of researchers at the University of Hong Kong is now tackling a far more complex problem: applying AI to the field of art conservation. Their work could have outsize ramifications for the world’s art institutions, expanding access to cutting-edge art conservation tools, cutting the time needed for materials analysis and allowing even small museums to protect and preserve their collections.

A Hands-On Approach

Tucked away in a corner of the Hong Kong University Museum and Art Gallery is the only on-site, university museum research lab in Asia. There, a team of chemists, conservation scientists, and students under Professor Marc Walton of HKU’s Museum Studies programme and the Department of Chemistry’s Dr Kenneth Ng are developing, building, and experimenting with new instrumentation that could radically lower the barriers to characterising the materials comprising objects of art and archaeology.

Looking around the lab, it’s hard to imagine that, a little over a year ago, almost none of this infrastructure existed. Before Walton, who was previously Head of Conservation and Research at M+, joined HKU in 2024, he had never met Dr Ng. They were brought together by another new arrival to the university, Chemistry Professor Jay Siegel, who recognised the pair’s shared interest in both chemistry and mechanical tinkering.

For Professor Siegel, the collaboration offered a solution to two longstanding issues in university education: how to break down barriers between disciplines and give students hands-on experience with real-world applications.

“(The students) are very well trained, they know their theories, but they’ve never touched an artefact before.”
— Dr Kenneth Ng

Soon, what began as a series of informal conversations morphed into something very real, with Walton joining HKU, then teaming up with Ng and the Chemistry department to bring their vision to life. “Jay recognised that Kenneth and I were thinking along the same lines,” says Walton. “He couldn’t have been more correct. This is the type of cross-fertilisation you normally wouldn’t think about: bringing a chemist together with someone from the humanities.”

Bridging Art and Science

Viewed from the front, the UMAG lab’s “Franken-camera” doesn’t look particularly unusual. It’s only when you circle around back that the moniker’s logic starts to come into focus, revealing uncovered wires and chips that have been grafted on by the team to improve performance.

The Franken-camera is far from the only curious-looking tool in the lab. The team uses a wide variety of analysis instruments, from off-the-shelf items like the handheld XRF spectrometer – designed to mimic the look of a Star Trek phaser – and modified microscopes. Their commitment to home-brewed tech isn’t just about performance or customisability; by opting to custom build tools, they give students more opportunities to practice a variety of skills that could be useful for their futures, such as the integration of code with hardware.

“The one big hurdle that we face in teaching is that ‘fancy’ microscopes are usually very, very expensive and ‘thou shalt not touch it.’”

— Dr Kenneth Ng

One of the biggest new tools in their toolkit is artificial intelligence. While the underlying machine-learning tech has been around for decades, the proliferation of LLMs and AI applications is drastically shortening the time spent on materials analysis – a key step in the process of understanding and conserving a piece of art.

As an example, Walton points to the traditionally time-consuming task of point analysis of artefacts. AI is allowing the team to take a handful of data points – some with detailed spectroscopic information, others that cover a larger portion of the artefact and show spatial details – and merge the sets to produce an image cheaply and quickly.

Expanding Access Through AI

“AI allowed us to do that. Before, it was really difficult to be able to fuse these different things together, to be able to create something that combines the best of both worlds.”
— Professor Marc Walton

Both Walton and Ng spotlight AI’s impact in expanding access to art conservation. Traditional conservation characterisation methods are expensive and difficult to use, leaving institutions across the globe struggling to balance their desire to augment the value of their collections through study and treatment with the associated costs of bringing science into the museum. 

If the cost of conservation tools and processes could be brought down, the thinking goes, then many of these problems could be solved with some basic technical knowledge and a little ingenuity. 

“These are things that any museum around the world, any person that’s interested in duplicating our work, could conceivably be able to do it without spending a whole lot of money,” says Walton.

The Future as Blank Canvas 

The team cautions that AI isn’t a cure-all, and that many of the classical methods of conservation continue to work well. Rather than completely overturning the field’s received knowledge, they’re focused on teaching students how to develop and use new tools while maintaining a critical mindset.

“It’s very important for students to know the nuts and bolts of AI rather than just using it as a black box,” says Ng, using a common metaphor for the opacity of AI algorithms. “Especially as a scientist, you really need to put on that scientist hat and differentiate whether it’s hallucinating, or whether it’s giving you the right answers.”

Still, they remain excited about the tech’s potential for conservation, with Walton pointing to possibilities, not just for the museum world, but also for lowering the barriers to connoisseurship and changing the field of art history.

That’s not all: Asked whether AI can bridge the gap between objective and subjective analysis, Walton pauses for a moment before turning philosophical. “I always think the objective and subjective come together, because the agency of the artist is in the materials, which we can be objective about” he says. “But really, what we want to understand is the subjective part of it. So, we’re using science as a tool to assess the subjective.”

To learn more about how Professor Walton and Dr Ng are using AI to rewrite the rules of art conservation, watch the video below:

January 30, 2026
On Red Tides: How Wang Mengqiu Is Tracking Algal Blooms With AI

Forests aren’t the only fragile ecosystems at risk from climate change. In the right conditions, Sargassum, a kind of macroalgae, can trap vast amounts of carbon and store it at the bottom of the ocean. But what happens when those conditions change?

The University of Hong Kong’s Professor Wang Mengqiu doesn’t want to wait to find out. Working with satellite imagery and advanced AI algorithms, she’s tracking oceanic conditions, monitoring blooms, and trying to answer an all-important question: How much of a given Sargassum bloom will become a carbon sink – and how much will wind up a carbon source?

Across the sea

The scale of Sargassum blooms can be hard to comprehend. To demonstrate what the planet is facing, Professor Wang pulls up a map showing the Great Atlantic Sargassum Belt, which stretches from the western coast of Africa across the ocean to the Caribbean and Mexico.

A recurring natural phenomenon, the Belt was observed as early as the late 15th century. “The ocean can benefit from Sargassum,” Professor Wang says, adding that they are “very big players in the ocean carbon cycle.”

Yet the Belt’s recent growth has worried researchers, as blooms set historical highs and landfall events in the Western Atlantic smother beaches and coastal wildlife.

“Satellites are an ‘eye in the sky,’ giving us a synoptic view of what’s happening below.”

– Professor Wang Mengqiu

“In the open ocean, Sargassum acts as a carbon sink,” Professor Wang explains. “But if it reaches the coast, it releases hydrogen sulfide (a compound linked to asthma) and endangers coastal vegetation, including corals and seagrasses.”

Step one in addressing the problem is figuring out just how large the Belt has become – and where and when landfall events are likely to occur. But how do you measure something that spans an entire ocean?

Professor Wang shows the scale of the Great Atlantic Sargassum Belt

AI in the sky

Typically, tracking Sargassum blooms might involve observations from a boat.

Both of these methods are expensive and time-consuming. So researchers like Professor Wang are turning to satellite and remote sensing technology, leveraging tools like AI and big data to quickly and efficiently identify Sargassum patches and map their locations.

Using satellites designed to monitor colour changes in the ocean, scientists can get a much better idea of how big the Belt – actually a thousands-of-miles-long stretch of discrete blooms – really is.

Although the underlying technology has been around since the 1980s, the field has benefitted in recent years from new hyperspectral sensors, higher-resolution images, and the shift from machine learning to deep learning.

“We don’t want to just react, we need to be able to make predictions.”

– Professor Wang Mengqiu

The launch of NASA’s Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) Satellite in 2024 was particularly exciting, Professor Wang says. “The satellite is a game changer because it’s hyperspectral,” she explains. “It will give us a much finer, more detailed knowledge of global changes.”

Previously, that level of detail would have been hard to analyse. But new deep learning techniques are making even hyperspectral imaging analysis possible, Professor Wang says. Her own work is one reason why: She just published a paper showing how to automatically de-noise images to better monitor algal blooms.

These advances will ultimately make it easier for scientists to predict the size and impact of future blooms, allowing for better management and mitigation.

Professor Wang explains her work using AI to track algal blooms

The search for solutions

Professor Wang’s next project is to develop algal bloom distribution maps that can be used to guide policymaking.

The Atlantic isn’t the only ocean facing growing problems related to Sargassum. In the Yellow Sea and East China Sea, for instance, blooms have caused havoc in recent years.

Professor Wang hopes to work together with the Hong Kong government to understand public needs and design real-world applications for researchers’ growing trove of remote sensing data, including alarm systems for Sargassum landfall.

“The goal is to convert Sargassum into something useful: energy, building materials, or other applications.”

– Professor Wang Mengqiu

While these ideas are still a long way off, she believes they could one day help mitigate or even eliminate the damage done by Sargassum to coastal ecosystems like Hong Kong’s – and even help vulnerable populations like the city’s fishers.

“I want to convert satellite products into something broadly useful,” she says. “As scientists, we have to serve the community.”

Professor Wang poses on the HKU campus

January 9, 2026
Jin Wu Wants the World to See the Forest for the Trees

When most people look at a forest, they see an example of nature at its best, a planetary lung that is one of our most reliable defences against climate change. But University of Hong Kong Professor Jin Wu sees something else: a delicately balanced ecosystem that, if not properly managed, could play havoc with earth’s future.

That’s because not all forests are created equal. The Amazon, for example, was traditionally dominated by evergreen trees. More recently, however, a mix of environmental changes, drought, and human encroachment have led many of these to die off and be replaced by deciduous variants.

While it’s too early to say how this will impact the climate, the end result could be a vicious cycle, Professor Wu says. As temperatures rise, trees need more water to sustain themselves. If the water runs low, more evergreens die and are replaced by deciduous trees, which do not transmit water from the soil to the air as efficiently. That means less rainfall and even more evergreen loss, releasing their stored carbon into the atmosphere as part of a process scientists call “Amazon dieback.”

“People think of forests as a way to fight climate change, but they can also facilitate climate change.”

– Professor Jin Wu

All is not lost, however. Professor Wu and his team are among the many scientists working to understand and track changes underway in the Amazon, in his case, through satellite data and remote sensing. “Remote sensing lets us scale up the knowledge we have of individual leaves and plants to a global level,” Professor Wu says with a characteristic smile. “Then we can turn that global knowledge into science-based decision-making.”

Professor Wu poses with his students in his lab

From tree to forest

The field of remote sensing is developing quickly, thanks in large part to new tools like machine learning and AI algorithms, Professor Wu says. One of the most exciting developments his team is working on is using satellite data to track the chemical fingerprint of individual plants, offering an unprecedented window into forest composition.

The idea comes from a 2009 study by ecologist Dr Greg Asner, who figured out how to identify individual plant species based on the chemical compositions of nitrogen and phosphorus. Because these plant chemical traits interact with light in different ways, satellite imagery could theoretically allow scientists to identify the functional composition of an entire forest.

“Plants all have unique chemical fingerprints, and remote sensing plus AI can help us identify them.”

– Professor Jin Wu

The key hurdle is data. Commonly used hyperspectral imaging techniques are prohibitively expensive and time-consuming. A few years ago, generating a hyperspectral image for a 300-by-300-meter plot cost around HK$20,000.

In search of an alternative, Professor Wu’s team turned to a multispectral approach. By looking at 10 criteria across dense image time-series within a year – as opposed to the over 100 needed for hyperspectral imaging – they believe they can develop a complete picture of plant coverage and the underlying high-dimensional chemical compositions around the world at a fraction of the cost.

If it works, it could allow scientists to track how plant species are changing across both space and time. And that knowledge, Professor Wu says, might prove vital in the fight against climate change.

Professor Jin Wu shows off some of the equipment he uses to monitor changes in forests.

Shrubs and CPUs

Professor Wu and his team aren’t just looking at the Amazon. They’ve also trained a machine learning model to identify shrubs in the Inner Mongolian grassland. Long-term, they plan to map vegetation across the steppe, keeping an eye on this important ecosystem.

Crucially, because their data is compatible with long-running databases, they can also look backward in time, allowing them to study the changes that have already taken place over the past 30 years.

“The field is evolving very quickly, and a lot of things we’ve never even dreamed of will soon be possible.”

– Professor Jin Wu

AI has helped make this possible, Professor Wu says, but the computational power needed represents a strain on the resources of smaller labs. Ideally, AI resources could be centralised so they benefit multiple teams at once – what he calls a “centralise and service” approach.

Science with a smile

Despite the challenges, Professor Wu remains optimistic about the future of remote sensing.

His sunny attitude toward the sometimes slow speed of research progress is reflected not just in his near constant smile, but also in the list of advice for new PhD students on his office whiteboard, which includes both old saws like “be patient” and reminders to “be generous to yourself” and to not think of yourself as purely a helper.

Professor Wu almost always has a smile on his face, even when explaining complex science.

There’s also an item on there about the importance of the spirit of “discovery,” rather than just safely trying to refine existing methods. It’s advice he appears to have taken to heart, as even the mere mention of remote sensing’s prospects in the coming decades causes him to wax excitedly about new mechanisms and techniques that could revolutionise our understanding of the planet.

“This is an important field for the 21st century, and we’re well placed to connect science and policy,” he says. “I just hope the University of Hong Kong can leverage this curve and pioneer novel research.”

January 6, 2026
Can Satellites Point the Way to More Liveable, Sustainable Cities?
In just 40 years, China went from a predominantly rural society to a highly urbanised one. What happened? And what lessons can other urbanising regions around the globe draw from that experience?

University of Hong Kong Professor of landscape architecture Bin Chen believes we can find the answers in satellite data. An expert in remote sensing and Director of HKU’s Future Urbanity and Sustainable Environment (FUSE) lab, he’s using satellite imagery to track everything from the historical growth of cities to emerging problem areas like urban heat, air pollution, and green/blue space loss.

It’s a cutting-edge field – and one with implications far beyond China. Done correctly, it could provide one of the first-ever windows into how cities grow and evolve in the real world. This emerging “urban intelligence” will in turn have major ramifications for rapidly urbanising countries across the Global South, allowing them to avoid mistakes made by other cities while empowering them to create more equitable built environments for all residents.

We asked Professor Chen to share his thoughts on the nature of his work, the importance of remote sensing and AI to urban planning, and how the world can build more equitable cities.

Professor Bin Chen points to some of his research on satellites and urban intelligence

The “intangibles” of sustainability

Professor Chen characterises his work as looking at the “tangibles and intangibles” of the built environment.

The tangibles are relatively straightforward. But the intangibles – air, heat, sunlight, light pollution, shade, and noise – play just as important a role in our lives and health, while being much harder to quantify.

That’s where satellites can play a role, Professor Chen says, as new data and methods allow researchers to look at cities on a granular level, identifying problematic heat islands, light pollution, and air quality.

“It’s not just about climate. It’s also about liveability.”

– Professor Bin Chen

Importantly, these are also problems of equity and ESG, with poorer residents often having to go without access to parks, shade, and clean air.

“Take shade, for example,” Professor Chen says. “In a city of high-rises, where you have people living in subdivided units with no access to sunlight year-round, shade is a social issue.”

Even if every building in a given district meets regulations, it can be hard to tell what the overall outcome will be – a problem satellite imagery and AI modelling can help solve.

While Professor Chen’s team can identify the issue, he hopes work on the solutions will be an interdisciplinary affair. “We need to bring experts together,” he says. “We need knowledge from social economics, environmental studies, data science, and urban planning.”

Professor Bin Chen poses with a sign for the Department of Landscape Architecture and HKU

Urban intelligence

The long-term goal is to develop what Professor Chen calls “urban intelligence” – a deeper understanding how people, cars, buildings, and the environment interact – then apply that knowledge to developing regions around the world.

Perhaps surprisingly, given their importance in modern life, the actual mechanisms by which cities grow are not always well understood. It wasn’t until 2008, when satellite data became more widely accessible, and 2015, when large-scale cloud computing and machine learning caught up, that researchers could closely examine the growth of modern cities.

“AI makes everything more efficient.”

– Professor Bin Chen

Professor Chen points to Shanghai’s Pudong New Area as a classic example, noting that while policy documents offer a window into its growth, remote sensing technology allows researchers a seamless, transparent view of how the district grew into a global financial capital.

Leading the way

But taking advantage of these advances will require more – and more open – data.

Still, researchers can make significant progress via virtual collaborations. A few years ago, Professor Chen worked with HKU Professor Peng Gong and partners from 23 universities and institutes across China on a database of Chinese urban land use.

By mixing satellite data with on-the-ground verification, they were able to create the country’s first nationwide parcel-level essential urban land use categories map, allowing researchers to easily compare cities around China, from Beijing to Shenzhen and Wuhan.

“Remote sensing is becoming more and more powerful. We used to have to focus on individual cities, but now we can look at entire countries, even the whole globe.”

– Professor Bin Chen

That empowered numerous follow-up studies, and the team hopes to expand their map to the rest of the world in the coming years.

“We’re entering a new stage, going beyond remote sensing with multimodal data,” says Professor Chen. “But pushing the field forward will require more data sharing.”
Professor Bin Chen, an expert on satellites, remote sensing, and urban intelligence, speaks to a student
Professor Bin Chen, an expert on satellites, remote sensing, and urban intelligence, poses in front of a sign for his Future Urbanity and Sustainable Environment Lab
Professor Bin Chen, an expert on satellites and urban intelligence, poses with some of his awards
December 19, 2025
Shunlin Liang Wants to Feed the World, One Orbit at a Time
Could satellites one day help us solve pressing socioeconomic problems like food insecurity? What if they could directly tell farmers when to fertilise their crops? Or how much water they should use to irrigate?

It sounds like science fiction, but according to University of Hong Kong Chair Professor of Remote Sensing Shunlin Liang, it’s not as far-fetched as you might think.

Professor Liang, who is also the Director of HKU’s Jockey Club STEM Laboratory of Quantitative Remote Sensing, has spent much of his thirty-year career trying to maximise the social benefits of remote sensing and satellite imaging technology. Now he and his team have developed AgriFM, a new artificial intelligence foundation model that could revolutionise how farmers in food-insecure regions manage their crops.

We asked him to share his thoughts on his new model, how remote sensing can improve lives around the world, and why it’s so important for researchers to have access to open-source tools.

Laying the foundation

Roughly half of the people dealing with food insecurity worldwide live in Asia. When Professor Liang saw photos of their plight, it motivated him to take action, and the result of that work is AgriFM, an open-source foundation model that offers superior crop-mapping performance over conventional deep learning methods.

While smallholder farmers generally know every inch of their land, managers on larger farms struggle to track crop growth and needs across a wide area. Remote sensing and AI technologies can help, Professor Liang says, allowing them to quickly identify when and where additional water, pesticides, or fertilisers are needed. That, in turn, will save farmers money while boosting yields.

“We need to maximise the benefits from minimal inputs.”

— Professor Shunlin Liang

That’s the short-term goal. Long term, Professor Liang says, the technology could prove even more transformative, predicting crop yields months in advance, helping governments and NGOs identify problem spots before they lead to hunger.

Open source, open access

In some ways, AgriFM is a natural continuation of Professor Liang’s decades-long interest in developing open-source tools and making satellite data more accessible.

In 2013, Professor Liang developed the open-source Global LAnd Surface Satellite (GLASS) suite, allowing researchers anywhere in the world to work with satellite data from all the leading public databases.

While organisations like NASA and ESA release satellite data to the public, they do not make it cross-compatible. That’s where GLASS comes in, making it possible to achieve much wider coverage by drawing from and unifying data from across various space agencies in one-easy-to-use package.

“In the future, foundation models mean we’ll need fewer people working on methodological problems, freeing them to focus on real-world applications.”

— Professor Shunlin Liang

More data means more potential new models like AgriFM, which Professor Liang believes could dramatically change the remote sensing field in the next five to 10 years. Already, he’s working on upgrades to GLASS like High GLASS, which offers greater resolution.

“Foundation models will change everything,” he says.

A computing challenge

That future is still some ways off, however, in large part due to data storage constraints.

Thanks to support from the Jockey Club and HKU, his team currently has 10 petabytes of data storage, but Professor Liang has set an ambitious goal of growing that to 50 petabytes in the next few years.

“Launching satellites and collecting data is expensive. I don’t want that (data) to be limited to a small community.”

— Professor Shunlin Liang

The idea is to expand the GLASS suite’s impact. To that end, he’s hoping to partner with data vendors and donors to commercialise some products while making others free for use by “millions” of people around the world.

“I want to make the data open to everyone,” Professor Liang says. “As a university professor, this is what I should do.”
Professor Liang is a global leader in remote sensing. Now he wants to use satellites to solve food insecurity.
Professor Liang is a global leader in remote sensing. Now he wants to use satellites to solve food insecurity.
Professor Liang is a global leader in remote sensing. Now he wants to use satellites to solve food insecurity.
Professor Liang is a global leader in remote sensing. Now he wants to use satellites to solve food insecurity.
December 16, 2025
To Chu Zhiqin, Diamonds Are Forever – and for Everything

“One of the biggest limitations in fields like AI is power consumption, and the key bottleneck is that their semiconductors are based on silicon.”
— Professor Chu Zhiqin

Most people know diamonds as age-old symbols of eternal love, but to the University of Hong Kong’s Professor Chu Zhiqin, they’re the key to our technological future.

It’s no exaggeration to say that that we’re living in the age of silicon. The element has become the building block of modernity, powering the semiconductors in everything from precision medical equipment to our phones. But silicon is far from an ideal conductor, and scientists have long dreamed of fashioning superior alternatives out of more suitable materials like diamonds.

The challenge lies in the very attributes that make diamonds so desirable: their durability and toughness. How can you grow diamonds that aren’t just better conductors than silicon, but also have the flexibility needed for the next generation of computing devices, including both power-intensive uses like AI and difficult-to-manufacture wearables?

That’s where Professor Chu comes in. His research – for which he was just named a finalist in the prestigious Falling Walls Science Breakthrough awards – is revolutionising the field of diamond-based semiconductors, heralding the future arrival of flexible yet durable chips with more than 10 times the thermal conductivity of their silicon counterparts.

The Future of Semiconductors

In theory, diamonds are almost the perfect vehicles for a future “fourth generation” of semiconductors. They are relatively easy to grow – all you need is hydrogen, methane, and electricity – strong, and can handle far higher wattages and more heat than silicon.

The trick is in manufacturing. While the basic process is well established, researchers have long struggled to produce diamond structures suitable for next-generation devices like wearables: ones that are flexible without sacrificing durability.

Cracking the code took Professor Chu and his team over a year of hard work. But they finally developed a trade secret process that allows them to reliably produce high-quality, micrometre-thin diamond semiconductors that can be rolled and flexed like sheets of paper.

“That’s the beauty of being a scientist: You know more about your field than anyone else in the world. If you dig long enough and hard enough, no one will be better at it than you.”
— Professor Chu Zhiqin

The Next Generation of Computing

The barriers to replication are steep: Although the team’s work has already spawned imitators, the competing diamonds tend to crack quickly, Professor Chu says, making them unsuitable for industrial uses.

The next challenge is commercialisation. Professor Chu is trying to interest manufacturers – always difficult with any new material – and is focusing his energies on high-upside, future-centric sectors like heat spreaders for GPUs and electric vehicles. If they succeed, they could radically increase the efficiency of some of the world’s most important products.

Among the developments Professor Chu is keeping his eye on are new AI-based measuring tools. Although accumulating training data remains difficult, AI offers more efficient, dynamic measurements. “Everything can be analysed by AI,” Professor Chu says. “It tells us what the best measurement parameters are.”

“Basically, the way we’re doing measurements now is the Stone Age, but with AI, we can move forward.”
— Professor Chu Zhiqin

A Scientific Spirit

But Professor Chu isn’t solely fixated on cutting-edge technologies. As the interview winds down, he muses on another significant challenge facing the field: cultivating the next generation of scientists.

Many aspiring scientists find the process discouraging and gradually drift away from the field. “They think learning math is hard, and learning physics is boring,” he says, adding that they can’t see how they’re supposed to master everything.

That’s an unfortunate misconception, Professor Chu believes. “Scientists actually are highly specialised,” he says. “It’s not that scientists can solve everything, but we can solve a few, very specific things.”

This preoccupation is reflected in his lab’s work. When asked what his favourite device is, he doesn’t mention the powerful Zeiss microscope, but a miniature quantum mechanics learning kit developed by one of his graduate students, Madhav. The box allows teachers to show young students quantum mechanics in action, rather than theory, letting them experience for themselves what it’s like to be a scientist.

It may seem small – or an unusual project for a lab of this stature – but it’s just one more way Professor Chu and his team are shaping the future of science.

October 31, 2025
Teaching Machines to Think Quantumly: Qi Zhao on the Frontier of AI-Driven Computing

“Quantum computers don’t just calculate. They learn from the rules of nature itself.” — Prof. Qi Zhao

Artificial intelligence is everywhere — in our phones, our cities, and the tools we use to think.

But for Professor Qi Zhao at The University of Hong Kong’s School of Computing and Data Science (CDS), the next leap in AI may come from a place far smaller than any silicon chip.

His research explores how quantum physics and machine learning can work together to create a new kind of intelligence — one that learns the way the universe learns.

From Theory to Computation

Zhao trained as a quantum information theorist, studying how data behaves when stored in particles rather than bits.
At HKU CDS, he leads a group that builds hybrid computing models combining classical algorithms with quantum processors.

His goal is simple to state but hard to achieve: use quantum systems to make AI faster, smarter, and more energy-efficient.

“Classical computers follow fixed paths,” he explains. “Quantum computers can explore many paths at once. That difference changes how learning works.”

Reimagining Computation

Traditional AI trains neural networks through repetition — adjusting parameters until patterns emerge.
Quantum computers take a different approach.

They rely on variational quantum algorithms, where a small quantum circuit learns by tuning itself with help from a classical controller.

Think of it as teamwork: the quantum part handles exploration; the classical part handles evaluation. Together, they solve problems that would take ordinary machines far longer to compute. Zhao’s team studies how this cooperation could transform optimization tasks, from image recognition to material design.

Quantum Machine Learning in Action

Inside his lab, AI helps control fragile quantum hardware.

Algorithms adjust pulse shapes, timing, and temperature to keep qubits stable. The system learns which conditions produce reliable results and adapts automatically when the environment changes. “It’s feedback learning in the truest sense,” Zhao says. “The machine is teaching itself how to stay coherent.”

These experiments do more than improve performance. They show how AI and quantum physics can enhance each other.
AI stabilizes quantum devices; quantum mechanics gives AI new mathematical tools for creativity and pattern discovery.

Learning from Quantum Data

Zhao believes that the next revolution will come when AI no longer just analyzes quantum data — it learns inside quantum data.

His group explores models where quantum systems perform the learning directly, finding relationships hidden from classical logic.
Such systems might recognize molecular structures or financial correlations beyond human intuition.

“This is where AI stops imitating intelligence,” he explains. “It begins to share it.”

Mentorship and Collaboration at CDS

As a mentor, Zhao encourages students to cross boundaries between physics and computer science.
He collaborates closely with Prof. Giulio Chiribella, Prof. Yuxiang Yang, and Prof. Ravi Ramanathan, creating a bridge between theory, experiment, and data science.

In class, he simplifies complex formulas into visual intuition. His students learn not only to code algorithms but also to think about why an algorithm works. “The most exciting discoveries,” he says, “often happen when we try to explain them simply.”

Looking Ahead: The Shape of Quantum Intelligence

Zhao imagines a future where AI systems powered by quantum hardware design drugs, manage energy grids, or simulate ecosystems in real time.

These machines will not replace human reasoning; they will extend it.

“Intelligence isn’t just logic,” he says. “It’s the ability to learn from limited information. That’s what quantum mechanics has been doing for billions of years.”

In his view, teaching machines to think quantumly is not just about computation — it’s about understanding learning itself.
And at HKU CDS, that journey has already begun.

August 22, 2025
When Randomness Becomes Intelligence: Ravi Ramanathan on Quantum Security and the Limits of AI

“In both physics and life, uncertainty isn’t a flaw — it’s what keeps everything interesting.” — Prof. Ravi Ramanathan

In a world driven by algorithms, certainty feels powerful. Yet for Professor Ravi Ramanathan of The University of Hong Kong’s School of Computing and Data Science (CDS), the opposite may be true.
He studies how randomness and trust shape the future of both quantum security and artificial intelligence — two fields that depend on data, but also on doubt.

From Theoretical Curiosity to Digital Trust

Ramanathan’s path began in theoretical physics. He was fascinated by the strange mix of order and unpredictability inside quantum systems. Over time, that curiosity evolved into a question that now defines his research: Can uncertainty itself protect information?

At HKU CDS, his group designs quantum cryptographic protocols that don’t rely on trusting the devices used to send or receive data. Instead, they use the laws of physics — not human assurances — to guarantee security.
“It’s like replacing a lock built by people with one built by nature,” he says.

Building Unhackable Systems

Traditional encryption depends on mathematical puzzles that powerful computers might one day solve. Quantum cryptography flips that logic.
It uses quantum particles, whose behavior changes when observed, to detect any eavesdropper immediately.

Ramanathan’s work focuses on device-independent security — a method where users don’t even need to know how their devices are built. As long as the results obey specific quantum correlations, the communication is secure.

This idea has major implications for AI as well. “AI systems make decisions based on patterns in data,” he explains. “If those data streams are ever compromised, the intelligence that depends on them becomes fragile. Quantum security keeps the foundation solid.”

The AI Connection: Intelligence Without Certainty

While most AI operates deterministically — producing the same output for the same input — the quantum world thrives on probabilities.
Ramanathan believes that future forms of quantum AI may combine these two views of intelligence: the structured logic of algorithms and the creative randomness of quantum mechanics.

“Learning,” he says, “might require a balance between prediction and surprise.”
By introducing controlled randomness, quantum systems could explore possibilities that classical AI would never consider. The result might be machines that don’t just calculate outcomes — they imagine them.

The Ethics of Uncertainty

As AI becomes more autonomous, questions of control and trust follow closely behind. Ramanathan’s research in quantum randomness adds a unique ethical layer: unpredictability can protect privacy.
“In cryptography, unpredictability is freedom,” he says. “If every decision were predictable, there would be no security — and no choice.”

He often compares the challenge of securing algorithms to the challenge of keeping human decision-making free from bias. Both require space for uncertainty. Both depend on humility before complexity.

Collaboration and Teaching at CDS

At CDS, Ramanathan collaborates with Prof. Giulio Chiribella, Prof. Yuxiang Yang, and Prof. Qi Zhao to connect foundational theory with practical technology.
He teaches courses that blend mathematics, physics, and computer science, encouraging students to question what it really means to “know” something in a computational world.

His mentoring style mirrors his research: open-ended, curious, and slightly unpredictable. “Students learn best when they discover answers for themselves,” he notes.

Looking Forward: Trusting Uncertainty

Ramanathan sees a future where quantum communication and AI-driven reasoning merge into systems that are both intelligent and secure.
These technologies may protect digital infrastructures, power next-generation networks, and even redefine how machines reason about risk.

“Randomness is often seen as noise,” he says. “But in nature — and in intelligence — it’s also creativity. Embracing it might be our best safeguard in an age of perfect prediction.”

August 15, 2025
How Quantum Sensors Learn: Prof. Yuxiang Yang on Intelligence Beyond Silicon

“To make better sensors, we need to teach them how to learn from the world around them.” — Prof. Yuxiang Yang

When most people think about intelligence, they picture circuits and code — the realm of artificial intelligence. But for Professor Yuxiang Yang of the School of Computing and Data Science (CDS) at The University of Hong Kong, true intelligence may start not in software, but in how we sense reality itself.

Yang’s research bridges the boundary between quantum physics and machine learning, using AI-driven methods to make quantum sensors more precise, adaptive, and even self-correcting. His work explores what happens when perception — human, mechanical, or quantum — becomes an intelligent process.

A Physicist Who Thinks Like an Engineer

Before arriving in Hong Kong, Yang trained as an optical physicist fascinated by the limits of measurement. “Every experiment,” he says, “is a conversation with nature. The better you listen, the more you understand.”

Now at HKU CDS, he leads projects that bring AI algorithms into quantum laboratories, transforming how sensors detect the tiniest fluctuations in magnetic fields, light, or vibration. Where most people see noisy data, Yang sees a new kind of language — one that can be decoded through learning.

The Science of Quantum Perception

Quantum sensors exploit delicate quantum states to measure the world with extreme accuracy — detecting signals a billion times smaller than what conventional devices can capture. The challenge, however, is noise: every vibration, temperature shift, or stray photon can overwhelm the signal.

Yang’s team tackles this by training AI models to separate meaningful information from random interference. Instead of manually tuning every parameter, algorithms learn to predict, filter, and adapt in real time. In effect, the sensor begins to “think” about its own environment.

When Sensors Get Smarter

The concept is revolutionary: a sensor that improves through experience. By combining reinforcement learning with quantum-optical hardware, Yang’s group designs systems that adjust their measurement strategies automatically.

In one project, machine-learning routines identify patterns in experimental noise and recommend control settings that boost sensitivity. In another, neural-network architectures inspired by human vision help interpret complex interference patterns in light-based sensors.

The results are striking — faster calibration, higher precision, and lower error rates. What used to take days of manual adjustment can now happen in minutes. “It’s like giving intuition to an instrument,” Yang says.

The AI–Quantum Feedback Loop

For Yang, the relationship between AI and quantum science runs both ways. Quantum experiments provide unique datasets that could inspire new machine-learning architectures; conversely, AI provides tools that push quantum systems to their limits.

“Quantum systems don’t process information the way classical computers do,” he explains. “They can explore many possibilities at once. Understanding how they ‘learn’ from data may reshape how we define intelligence itself.”

This feedback loop — between artificial learning and quantum understanding — is precisely where Yang believes future breakthroughs will occur.

Mentorship and Collaboration at CDS

At HKU CDS, Yang works closely with colleagues such as Prof. Giulio Chiribella, Prof. Ravi Ramanathan, and Prof. Qi Zhao in a research environment that encourages cross-disciplinary thinking.

His students are trained not only to run experiments, but to write code, model data, and reason from first principles. Many take part in HKU’s Common Core course “The Quantum Revolution,” where Yang introduces AI-driven approaches to measurement as a window into the nature of knowledge itself.

“We’re building a new generation of researchers who speak both languages — the language of physics and the language of data,” he says.

Looking Ahead: Perception as Intelligence

As AI systems become ubiquitous, Yang believes the next frontier is not only smarter computation but smarter perception — machines that can sense the world with the nuance of quantum systems.

His vision is ambitious yet grounded: quantum sensors that monitor ecosystems, detect disease, or guide autonomous vehicles with unprecedented precision — all while learning continuously from the environment.

“In the end,” he reflects, “AI and quantum physics are asking the same question: how do we extract meaning from uncertainty? The answer may begin with how we choose to observe.”

August 15, 2025