Leading New Zealand’s computation evolution

Recognition of cancerous tumours, efficient rubbish truck routing, improved production of mussels – the real-world issues Professor Mengjie Zhang’s research in Evolutionary Computation touches are as varied as they are numerous.

Professor Mengjie Zhang

The Professor of Computer Science at Victoria University of Wellington (VUW) heads the interdisciplinary Evolutionary Computation Research Group, which is at the forefront of New Zealand’s fast-growing tech knowledge sector. Since 2007 Zhang has been listed as one of the top 10 genetic programming (GP) researchers in the world by the GP bibliography.

“We can safely say New Zealand is taking international leadership in evolutionary computation in a number of areas, from genetic programming to evolutionary feature selection, evolutionary deep learning, evolutionary scheduling and combinatorial optimization, and learning classifier systems,” he says.

Fresh from giving keynotes in Brazil and Shanghai, the IEEE Fellow and author of more than 500 papers notes: “It’s quite a specialist area but in many ways traditional AI can’t do what this field can do.”

Genetic programming is an evolutionary machine learning technique that employs the Darwinian principle of survival of the fittest to automatically evolve and learn computer programs for complex tasks.

Zhang’s team is using EC and GP for biometric problem solving, working alongside VUW’s School of Biology on biomedical data analysis, biomarker detection and classification, and medical image analysis. Computer vision applications work on edge detection, segmentation and object recognition. Evolutionary algorithms can then select, rank and classify images in large data sets faster and more accurately than humans.

“If we have different criteria or objectives, or potentially conflicting objectives, this is where this field excels. For example, in terms of tumour diagnosis, we want to diagnose as many real cases as we can, but we want to produce very few false alarms. If we diagnose people, they can seek treatment which is a good outcome; but if we say the person has a disease and they don’t, that might frighten them and be a cause of other problems. Automated machine learning, particularly evolutionary deep learning, can find good classifiers for these problems.

“You’ve also got different types of cancer, so it is important to classify which it is. There are 10,000 ‘biomarkers’ (features) but there are maybe 8-10 that are important ones (actual biomarkers), and our technology searches the components most relevant to solving the problem to detect these.

“The prediction model needs to be accurate in how many days or years a person might have left. To me that’s a modelling problem… but to them, that’s something very important to know. Genetic programming, symbolic regression and modelling activity can find the most efficient solutions.”

Genetic programming is also being used to ensure efficient production scheduling, rostering and dispatching. Hence the rubbish truck application: “It’s a vehicle routing problem. You have multiple vehicles and you need them to go to every location; but you don’t want them to go there multiple times. Then you have a road closed, or a truck breaks down – it’s a dynamic and uncertain situation and approaching it manually can be time-consuming and inaccurate. Automated design, using genetic programming hyperheuristic approaches, can learn rules and policies and can provide the most efficient route.”

Zhang’s team’s research also focuses on optimising the placement of sensors in wireless networks. One example is a project with New Zealand aquaculture research leader the Cawthron Institute. Using lights and sensors, EC is able to categorise the growth stages of mussels, giving accurate data to inform their feeding programmes, classification and collection. The $5m National Science Challenge SfTI Spearhead project aims to develop underwater sensing and communication tools and data analytics to let aquaculture farmers monitor their ocean farms from home base, advancing one of New Zealand’s fastest-growing export industries.

“Evolutionary Computing and Learning will have a major role in AI and machine learning in the next 10 years,” Zhang says. “It has applications in medicine, agriculture, manufacturing, space, public health, renewable energy, security, environment, economic growth – almost every area.”

It’s certainly an area he has excelled in, given his reason for studying it. After studying mechanical engineering in China, Zhang did a PhD in Computer Science (Artificial Intelligence) at RMIT in Melbourne. “In the 1980s AI was a very hot topic and now there's another peak. My supervisors made me believe it had a good future,” he laughs.

Zhang arrived in New Zealand’s capital in the year 2000, following his wife – also an academic – to a job at Victoria University. He is now a diehard Wellington fan: “Every student knows I am biased about Wellington. I think Wellington’s the best place in the world. It is small but it has everything, and Wellington Harbour is beautiful. Of all the great cities with great harbours I think Wellington is the best one.”

His department has grown over nearly 20 years, and he believes his Evolutionary Computation Research Group is now the largest in the Southern Hemisphere, with 20 staff, five post-doctorate and 30 PhD students. “We have a collaborative approach and support each other and have made great progress.”

It’s this mix of enthusiasm for his adopted hometown and his team’s strong international reputation that helped deliver a successful IEEE 2019 Congress on Evolutionary Computation conference in Wellington in June.

A Fellow of IEEE, Zhang joined VUW’s Dr Bing Xue, Associate Professor in Computer Science, to bid for the 2019 event against fierce competition.

“It’s the first time for such a big conference in this area to be held in New Zealand. IEEE was not sure about New Zealand – and then they wanted to come to Auckland, not Wellington! They finally agreed because we have a strong research group. I promised them “If you bring this conference to New Zealand, we will do very well.” We had the largest attendance for this conference in the last 10 years.

“We were really successful. The conference drew more than 540 delegates from 73 countries plus 60 volunteers, way higher than IEEE’s initial expectations of 400 people.”

Feedback was extremely positive, with particular praise for the venue – national museum Te Papa – and many delegates said this was their favourite conference, Zhang says.

Tourism New Zealand provided support from the beginning through its Conference Assistance Programme, preparing a bid document, video and presentation and, importantly, a budget to show it would be feasible to host the event in New Zealand. It also provided marketing collateral to drive delegate attendance.

“This group is one of the major groups for evolutionary computing in the world,” Zhang says. “Every year about 10 of our people attend the event. To bring it to New Zealand brought a number of benefits.

“First, when you host the greatest conference in the field in the world it greatly elevates our research reputation. Second, many experts in this field will come to New Zealand and collaborate with us. The conference finished on a Thursday; on the Friday we had 13 IEEE Fellows visit our group. That’s more than exist in New Zealand. Our students got to talk face-to-face with these world-famous experts in this field.

“That's good not only for me, but for New Zealand as a country, to collaborate with the world. That's why Tourism New Zealand supported us a lot.

“It showcases our country in a new way, too. New Zealand is very beautiful, but the knowledge level is high – and in Wellington, it’s even higher!”