Building and nurturing research initiatives in Information Retrieval

Link to your Favorite Post

Building the future of data discovery and research innovation

The lead copy should be more details about the work you do and it’s impact.

The Knowledge Discovery Lab (KDL) advances research in Information Retrieval (IR) by transforming classroom concepts into hands-on investigations and publishable studies. The group cultivates a supportive environment for experimentation, replication, evaluation, and writing in IR. This research lab fosters cutting-edge research in Information Retrieval, empowering graduate students to explore complex data challenges and contribute to breakthrough publications in the field.

Enhancing AI Explainability

Summarization Assistant for Academic Research: This project aims to build an intelligent summarization system that processes full-text academic papers and generates concise, user-tailored summaries. Leveraging pre-trained transformer models such as T5, BART, or BERT, the system will be fine-tuned specifically on scientific literature sourced from open-access journals and datasets like arXiv and PubMed.

The tool will enable users to generate a summary for the whole article or individual sections of the paper (i.e., Introduction, Methods, Results, or Conclusion). Beyond the usual AI summarization tools, which often act as black-boxes, this project will inform the user which specific portions of the original paper contributed most significantly to the generated summaries. By providing transparency and explainability into the summarization process, the system aims to enhance user trust and satisfaction, making it not just a summarizer but an interpretable assistant for academic research workflows.

Additional objectives include benchmarking the system’s output quality against traditional LLM-based summarization tools (e.g., Claude, Elicit), using metrics such as ROUGE and BERTScore, and developing a lightweight, accessible interface for end-users.

Pre req: Python programming, basic NLP concepts, familiarity with Hugging Face Transformers, and experience working with machine learning models

Hybrid Meta-Ensemble (HME) Framework for Pore Pressure Prediction

This project is part of the Research Runway Program supported by Khoury College of Computer Science and is an application of machine learning in the domain of geoscience. The objective is to enhance the accuracy and generalization of pore pressure prediction in complex geological settings. This framework integrates the strengths of multiple machine learning and deep learning models, including Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Deep Feedforward Neural Networks (DFNN), Random Forest (RF), and XGBoost, into a unified meta-ensemble structure. The HME framework leverages ensemble learning, deep feature extraction, and meta-modeling to capture both spatial and temporal dependencies inherent in well log data. By combining the localized pattern recognition of CNNs, the sequential trend modeling of RNNs, and the nonlinear feature interaction capabilities of ensemble tree methods and feedforward networks. We look forward to collaborating with the research and industry community to further refine and deploy this innovative solution in real-world subsurface exploration.

See All Ongoing Projects

Research Papers in Reviewed