Research

Publications, areas of focus and work in progress.

Focus

Research areas

Artificial Intelligence

Exploring Artificial Intelligence methods for developing adaptive systems and data-driven solutions to complex real-world challenges.

Machine Learning

Exploring ML techniques to develop intelligent systems, focusing on prediction, pattern recognition, and data-driven solutions.

Deep Learning

Reproducible training, drift detection and the engineering practice that keeps a deployed model honest over time.

Natural Language Processing

Retrieval, summarisation and extraction over domain documents, with a particular interest in evaluating what these systems get wrong.

Computer Vision

Detection and segmentation in settings where labelled data is scarce and the cost of a false negative is high.

Intelligent Diagnosis

Developing intelligent systems to analyze data, identify patterns, and support accurate, timely, and reliable diagnosis.

Publications

6 in total

2026

  1. Calibrated Uncertainty Estimates for Clinical Triage Models

    Humza Imran. Placeholder venue — replace with the real journal. Journal article

    Abstract

    Placeholder abstract. Triage models are usually evaluated on accuracy alone, which says nothing about whether a confident prediction deserves that confidence. This entry exists so the Research page has something to lay out; replace it with your own.

  2. Retrieval-Augmented Generation over Long Technical Documents

    Humza Imran. Placeholder venue — replace with the real conference. Conference paper

    Abstract

    Placeholder abstract. Chunking strategy tends to matter more than the choice of retriever once documents pass a certain length. Replace this entry with your own publication.

2025

  1. Label-Efficient Defect Detection in Industrial Imaging

    Humza Imran. Placeholder venue — replace with the real journal. Journal article

    Abstract

    Placeholder abstract. Annotation budget, not model capacity, is usually the binding constraint in industrial inspection. Replace this entry with your own.

  2. Practical Drift Detection in Production Machine Learning Pipelines

    Humza Imran. Placeholder venue — replace with the real workshop. Conference paper

    Abstract

    Placeholder abstract. Most drift alarms in production are caused by upstream schema changes rather than by genuine distribution shift. Replace this entry.

2024

  1. Evaluating Summarisation Without Reference Summaries

    Humza Imran. Placeholder preprint server — replace. Preprint

    Abstract

    Placeholder abstract. Reference-free evaluation is attractive precisely where it is least reliable. Replace this entry with your own preprint.

2023

  1. Placeholder Doctoral Thesis Title — Replace This Entry

    Humza Imran. Placeholder institution — replace. Thesis

    Abstract

    Placeholder abstract for the thesis entry so the Research page has one of each publication type to display. Replace or delete it.