His work has been accepted as a full paper at the 28th ACM SIGEVO Genetic and Evolutionary Computation Conference (GECCO 2026) — the largest peer-reviewed conference in evolutionary computation.
Titled “Teaching the Teacher: The Role of Teacher–Student Smoothness Alignment in Genetic Programming-based Symbolic Distillation,” the research was conducted during his Summer 2025 attachment at the NUS Department of Electrical and Computer Engineering, in collaboration with Mr. Fong Kei Sen, ECE PhD student.
The study addresses a key challenge in symbolic distillation, understanding why neural networks often fail to translate into accurate, interpretable symbolic models. The team identified a fundamental mismatch in complexity between “teacher” (neural network) and “student” (symbolic) models as the root cause. By addressing this mismatch, their large-scale study demonstrates a robust approach to generating symbolic models that significantly outperform conventional methods, paving the way for more reliable and interpretable machine learning systems.
This achievement underscores the impact of faculty mentorship at NUS, with Associate Professor Mehul Motani guiding Soumyadeep through advanced research in machine learning and computational intelligence. It also highlights the value of the IRIS@NUS programme, which provides outstanding undergraduate and master’s students from around the world with opportunities to engage in cutting-edge research and gain exposure to graduate-level studies.
