Making problem solving more efficient
The recently promoted Hao Wang from the Leiden Institute of Advanced Computer Science teaches computer programmes how to solve real-world problems. His goal is to make the process of problem-solving more efficient. On 1 November, he obtained his doctorate with the predicate cum laude. ‘I hope to solve difficult problems we haven't been able to solve so far.'
Hao Wang on Machine Learning
This video can not be shown because you did not accept cookies.You can leave our website to view this video.
Wang works in the domain of Machine Learning, or more specifically Optimisation. ‘This research area is mainly about how to teach a computer programme to solve real-world problems for us’, he explains. With the optimization technique, a computer programme selects the best solution from the available alternatives, with respect to some criteria. Wang particularly investigates what he calls ‘the efficiency issue’: how to make the algorithms which calculate the best solution, work more efficient in terms of time and costs? His new advancements are very applicable for solving many real-world problems. ‘For instance, the vehicle routing problem, where we want to minimise the costs in the delivery service or calculate the quickest route for an ambulance. Here, we search for the best route based on real-time traffic information. I want to solve such problems in a more efficient way.’
Among others, Wang worked on industrial processes. ‘For TATA Steel, we aimed at optimising their one-kilometre-long production line’, Wang explains. ‘Sometimes, defects arise in the produced steel, the company logically wants to minimise this number of defects.’ Tens of thousands of parameters, such as temperature and rolling speed, control the whole production process. Wang and his colleagues assembled all the data that the company collects in each production step. Then, they used this data to build a predictive model. ‘We can see at which settings of the controlling parameters the most defects occur. We abstract these observations into a model. With this model, given certain parameter settings, we can predict whether or not a defect will occur.’
After the model is validated, it can be used for optimising the production process. Wang: ‘We ask the model: what is the optimal setting that gives us the minimal number of defects? In this way, we apply our knowledge to the industry to help with their production.’
Solving new problems
Wang came up with several novel methods and algorithms for Optimisation and Machine Learning, filling in gaps in the research domain. He combined techniques and methods from different subdomains, for example the so-called Evolutionary Algorithms with the Bayesian Optimisation methods, which resulted in well-performing new algorithms. Wang will continue his research at Leiden University as a postdoc. He wants to focus on a theoretical analysis of some of the methods he proposed in his thesis. ‘I want to see if the theoretical findings reconcile with the empirical performance of those methods, so as to affirm they work correctly. That way, my advancements will be highly reliable if applied to a different scenario.’ Furthermore, Wang wants to apply the new methods and algorithms to more real world problems, such as the routing problem. ‘If we make optimisation more efficient and faster, we can solve problems we haven't been able to solve so far.'