Code Search (2021-2024)
Research project focusing on retrieving programming language artifacts related to some natural language queries from a pool of possible programming language artifacts and ranking them by relevance, using dual encoder models. Model is built in Python using TensorFlow and Keras modules. This model showed an average improvement of 10.03% over state-of-the-art methods in terms of MRR scores. The research was conducted under the guidance of Dr. Zhe Yu. This work has been published in Empirical Software Engineering (EMSE) Journal and presented at FSE 2025 in the journal-first track.
https://github.com/hil-se/CodeSearch
Comparative Learning (2023-2026)
Research project focusing on modeling learning comparative judgments for Agile story point estimation through machine learning and human subject experiments. Machine learning experiments involved building a model to learn from pairwise story point data and rank them. These experiments involved using GPT2, SBERT, FastText language models and traditional machine learning methods. The framework was built using TensorFlow modules. The proposed model showed an average increase of 21.84% in Spearman’s rank correlation coefficient scores over state-of-the-art models. The research has been conducted under the guidance of Dr. Zhe Yu.
https://github.com/hil-se/EfficientSPEComparativeLearning
Explainable image classification (2024)
Image processing and Explainable AI-based research project focusing on explaining a pre-trained VGG model's classification decisions on face image data. This work involved fine-tuning a pre-trained VGG model on SCUT face image data for classification, and using the model's gradients on the images to explain why the model made those decisions.
Comparative learning for face image attractiveness (2024)
Research project focused on modeling comparative learning on face image data. This work involved using the comparative judgment framework with a pre-trained VGG model as the encoder to predict a ranked preference order for the images.
Comparative learning for image captioning (2024 - 2026)
Research project focused on modeling comparative learning on image and associated caption data. This work involved using the comparative judgment framework on this multi-modal data to predict whether a paired image and text caption are likely to be connected.
Outdated comment detection for repository commits (2024 - 2025)
Research project focused on detecting whether the comment associated with repository commits are up-to-date or outdated after new commits. This work involved the use of various deep learning structures, including dual encoders.
Modeling Art Evaluations from Comparative Judgments (2024 - 2026)
Research project focused on modeling comparative learning on image data. This work involved using the comparative judgment framework on image based data to evaluate direct and comparative judgments on image data.
Bangla Abstractive Text Summarization using Encoder-Decoder Model (2019-2020)
A research project on constructing a dataset for the task of abstractive text summarization in Bangla, and constructing a deep learning based model capable of using said dataset. The model was written in Python using Tensorflow modules. The research was conducted as the final year research project at University of Dhaka under the supervison of Dr. Muhammad Asif Hossain Khan.
https://github.com/monoshizmkhan/Bangla-Abstractive-Text-Summarization
