Projects/Publications

A Novel Objective Function for Counterfactual Explanations Using Conic OptimizationV. Galwaduge, R. Wijethunga, A. Sadhu, J. Samarabandu 7th International Conference on Advancements in Computing (ICAC), IEEE, 2025

Combines two feature-space distance functions into a single counterfactual objective and solves it as a quasi-convex program, benchmarked against three numerical optimisation baselines on both shallow and deep classifiers.

ieee

Novel Actionable Counterfactual Explanations for Intrusion Detection Using Diffusion ModelsV. Galwaduge, J. Samarabandu Journal of Cybersecurity and Privacy 5(3), 68, 2025

A tabular diffusion framework that produces plausible, diverse counterfactuals for network intrusion detection and summarises them into global rules that filter incoming attack queries — alongside the first comparative study of counterfactual algorithms in the NIDS setting.

doi

Lexicon-based Fine-tuning of Multilingual Language Models for Low-resource Language Sentiment AnalysisV. Dhananjaya, S. Ranathunga, S. Jayasena CAAI Transactions on Intelligence Technology 9(5), 1116–1125, 2024

An intermediate-task fine-tuning method that transfers a high-resource sentiment lexicon into multilingual models, tested on three-class sentiment classification in Sinhala, Tamil and Bengali.

doi

Deep Learning Based Personalized Stock Recommender SystemN. Wijerathne, J. Samarathunge, K. Rathnayake, S. Jayasinghe, S. Ahangama, I. Perera, V. Dhananjaya, L. Sivaneasharajah International Conference on Neural Information Processing (ICONIP), 362–374, 2023

A hybrid recommender that pairs collaborative and content-based filtering to tailor stock suggestions to an investor's preferences and risk tolerance.

doi

BERTifying Sinhala — A Comprehensive Analysis of Pre-trained Language Models for Sinhala Text ClassificationV. Dhananjaya, P. Demotte, S. Ranathunga, S. Jayasena 13th Language Resources and Evaluation Conference (LREC), 2022

The first systematic comparison of multilingual and monolingual pre-trained models on Sinhala text classification, with new annotated datasets released for future work.

acl