By: M’KIS Marketing & Communications Department

Nick Sen Chua, Grade 11, has been passionate about computer science since he learned how to code in Grade 9. “Computer science was a field that I immediately fell in love with from the moment I walked into my first class. It combines everything I love about math, some aspects of the sciences, and overall was a fresh experience that, although incredibly alien to me at that time, really clicked when I learned how to code.”

When he was in Grade 10, Nick Sen pursued an independent study course in computer science as an elective, where he worked on researching and creating real-world products, such as a video game, for his Middle Years Programme (MYP) Personal Project. 

Not one to stop his rapid improvement in coding, Nick Sen pursued an internship this past summer. After contacting multiple companies, individuals, and universities, Nick Sen secured a two-month internship with Associate Professor Tomas Maul from the School of Computer Science, University of Nottingham Malaysia.

For two months, Nick Sen conducted research with Professor Maul, who specializes in neural computation and has research interests in machine learning, automation, and artificial intelligence. Their research culminated in the article “Antimatter Networks for Combating the Dying ReLU Problem,” published in the academic journal Machine Learning and Intelligent Systems. 

As a published academic, Nick Sen was invited to present the paper’s findings at the Machine Learning and Intelligent Systems conference held at the Universiti Tunku Abdul Rahman, Kampar campus. The international conference hosted professionals and academics from countries such as the United States, China, Japan, and India.

While he was the youngest participant at the conference, Nick Sen confidently presented the paper’s findings.

Nick Sen described neural networks, a method in Artificial Intelligence that teaches computers to process data in a way that attempts to replicate human brain processes. In simplified terms, this machine learning process uses computer nodes to process data, analogous to how neurons power the human brain. 

However, replicating neural networks is an exceedingly complicated task. One way in which computer scientists replicate neural networks is through the use of activation functions, which simulate decision-making in the brain. 

An example of an activation function is Rectified Learning Unit (ReLU) nodes, which are often easier to train and often achieve better performance. However, when ReLU nodes are used, it takes the maximum of 0 and the value, meaning that a value will remain perpetually 0 throughout training a neural network, not contributing to its training, which leads to the deactivation of a node.

To solve this problem, Nick Sen and Professor Maul investigated using activation swapping mechanisms, which try to activate the existing deactivated nodes and inverse, negative ReLU nodes in creating neural networks. 

Their paper concluded that combining regular ReLU nodes with inverse, negative ReLU nodes and activation swapping mechanisms was the most effective way to resolve the dying ReLU problem in neural networks. Inspired by the nature of the solution and the complex relationship between matter and antimatter in the universe, they also coined a new informal term for neural networks combining these variables, antimatter networks. 

Nick Sen said these findings could help “improve ReLU neural networks to train itself to make better accurate predictions on new data as well as perform consistently in test conditions and simulations.”

On Nick Sen’s work and research, Professor Maul added: “Nick implemented the first successful proof-of-concept of a novel idea in deep learning entitled antimatter networks, and to the best of my knowledge, is the youngest person in our school to publish his research in an international conference. As such, I believe Nick will inspire future students, undergraduate and even postgraduate, to be more ambitious and optimistic with their work."

To access the journal article, click on the following link.