Future paths: ongoing learning and AI
Globally, a lot of research is being undertaken in the field of Evolutionary Computation (EC), a branch of artificial intelligence (AI), where algorithms are inspired by biological evolution. EC techniques have proved extremely useful in identifying patterns and pulling out relevant information from vast amounts of labelled or categorised data.
As EC methods have global search ability, they are flexible and therefore applicable to a wide range of problems. EC approaches have become prominent due to their ability to provide high-quality solutions to data mining problems within a reasonably quick timeframe (from a computation perspective).
However, what happens when data is not labelled? For example, if a medical centre has collated information from its 5000+ patients in an unstructured manner, how can AI techniques help identify patterns in this vast pool? Similarly, how can seismologists gather meaningful insights from unlabelled recordings of earthquakes, which include a plethora of variables such as magnitude, depth, and location?
Increasing size and complexity of data in recent years has led to the creation of new AI techniques, which often end up producing results that are complicated for humans to understand. This is where unsupervised learning, a fundamental category of machine learning that works on data with no pre-existing labels, helps identify intrinsic patterns in the data. Where unsupervised learning is very sensitive to high-dimensionality, Andrew Lensen’s research questions the use of EC for feature manipulation being restricted to supervised learning,
Andrew’s research provides a comprehensive investigation into the use of evolutionary feature manipulation for unsupervised learning tasks. A variety of tasks were investigated, including the well-established one of clustering, as well as more recent unsupervised learning problems, such as benchmark-dataset creation and manifold learning. The study has shown the ability of evolutionary feature manipulation to improve the performance of algorithms and the interpretability of solutions in unsupervised learning tasks.
Speaking about what inspired him to pursue this field, Andrew says “it started with a particularly inspiring and encouraging lecturer. I took the Introduction to Artificial Intelligence course with only a vague understanding and slight interest of what AI was about. I remember listening to Meng (Mengjie Zhang) introducing evolutionary computation with vigorous passion, and found the idea of Genetic Programming to be my favourite part of the whole course.
Meng convinced me to do a summer project with him and Harith (al-Sahaf), after which I continued my Honours project with him, Harith and Bing (Xue), and then went to start my Master’s degree. Nine months in, Meng and Bing talked me into converting my degree into a PhD.”
Going on to talk about his experiences along the PhD journey, Andrew says “Doing a PhD is, quite literally, an experience like none other. The truth is that it is a bit of rollercoaster. One of the biggest challenges I faced was imposter syndrome. Every time I had a paper accepted, my response was ‘How have they not realised that I’m not nearly as smart enough to be doing any of this?’ Academia is fundamentally based on questioning everything: if you don’t have good justification or evidence for something, your work loses a sense of legitimacy. This has an effect on how you think, and how you see yourself, and is something that I think every academic and student struggles with,” he says.
“I’ve learned a great deal about myself, and have improved in a number of skills that are hard to quantify—writing, time management and presentations, to name a few. My supervisors have played an immense role in my life over the past 5 years, both as academic role models and as supportive friends. In a broader context, I think that when doing a PhD, you can’t be successful if you don’t collaborate with others.
“I’ve also had the opportunity to travel to Germany and Japan during my PhD. Both of these have been really great experiences, academically and otherwise. For now, I’m looking forward to graduating in December and I enjoy being a staff member – I’m in the first year of my two-year Postdoctoral Research Fellow position and this allows me to continue doing interesting research with my colleagues,” he says.