PhD researcher at Queen Mary University of London (School of Biological and Behavioural Sciences), supervised by Prof. Axel Rossberg. I am a machine learning researcher working on generative models for sparse, high-dimensional spatiotemporal data.
My work is on conditional generative modelling: recovering full system states from partial, noisy observations, with ensemble-based uncertainty. The methods are domain-agnostic. I currently test them on species distribution data, and have previously worked on multimodal classification and automated assessment.
Methods I work with: denoising diffusion models, graph neural networks, temporal transformers, U-Net architectures, ensemble calibration, individual-based and ODE simulation.
ORCID: https://orcid.org/0009-0000-7847-6242
| Degree | Institution | Period | Result |
|---|---|---|---|
| PhD Spatio-temporal process-aware models for biodiversity prediction | Queen Mary University of London | 2025 – Present | — |
| MSc Artificial Intelligence | Anglia Ruskin University | 2023 – 2024 | Distinction |
| BSc Machine Learning & Artificial Intelligence | Goldsmiths, University of London | 2020 – 2023 | First Class |
Demonstrator — Queen Mary University of London · 2026 – Present
Teaching Python, R, machine learning and data science, and assisting as a Demonstrator on the MSc module "AI and Data Science in Biology" (School of Biological and Behavioural Sciences).
PhD Student Researcher — Queen Mary University of London · Jan 2025 – Present
Spatio-temporal process-aware models for biodiversity prediction. Simulating biodiversity dynamics with ODE, IBM and PSD models; training a diffusion model (MetaDiffusion) on simulated communities; predicting species distributions with quantified uncertainty from sparse data; producing replacement metrics for IUCN Red List assessments.
PhD DSG Participant — Maritime Oil Slick Detection using AI — The Alan Turing Institute · Sept 2025
Two-week Data Study Group with Cefas: detecting oil leaks from shipwrecks in satellite SAR images. Built a reproducible remote-sensing pipeline for oil-slick delineation from Sentinel-2/1 imagery in a multidisciplinary team. Implemented grouped k-fold cross-validation by site to avoid leakage; produced before/after benchmarks (mIoU/F1).
Research Assistant — Anglia Ruskin University · Mar 2024 – Aug 2024
Analysed camera-trap imagery for wildlife monitoring; developed a machine learning model for badger identification and designed a software interface for biometric identification, working with external partners.
Research Student — Anglia Ruskin University · Dec 2023 – Mar 2024
Developed a deep learning approach for lung tumour detection; analysed and interpreted data, contributing to research reports and project planning.
Research Assistant Intern — Natural Language Processing — University of London · May 2023 – Aug 2023
Research in NLP using BiLSTM and BERT models; managed computational workflows and recorded outcomes to support text-processing semantics.
IT Systems Support — University of Cambridge · Nov 2018 – Dec 2020
Systems and user administration; hardware and software installation, configuration and troubleshooting; audio-visual support in seminar and meeting rooms; database support.
Internship Experience, Technology — Bright Network · Jun 2020 – Jul 2020
Technology internship; contributed to face-recognition features for a banking app and gained experience in agile software development and project management.
Languages & Frameworks
DevOps & Tools
|
Conditional Diffusion for Sparse Spatial Data A 79.6M-parameter conditional diffusion model that reconstructs multi-species range maps from sparse occurrence records, with ensemble-based uncertainty. Highlights:
Poster B-S.B.31, ECCB 2026 · Paper · DOI
|
IBM / ODE / PSD Framework Multi-patch individual-based model with infinite-pool assembly, ODE and PSD baselines, dispersal kernels, and invasion tracking. Highlights:
Key finding: PSD replicates IBM/ODE qualitative dynamics across body-mass regimes. PSD framework: unifies four metacommunity modelling paradigms as limiting cases of one process, running roughly two orders of magnitude faster than the ODE approach. Preprint: 10.64898/2026.04.21.719942
|
|
Vision-Language Fusion CNN (image) + BERT (text) with late fusion classification. Results:
|
🎙️ NART GradingSpeech Analysis for Clinical Assessment Automated grading of National Adult Reading Test pronunciations. Features:
|
|
Wildlife Biometrics & Re-identification Self-supervised features + metric learning for camera-trap individual identification. Highlights:
|
Attention-UNet & Adversarial Defenses Ultrasound segmentation with robustness testing under noise and adversarial perturbations. Results:
|
Some repositories are private during active research. |
1. MetaDiffusion: A process-aware conditional diffusion model for multi-species spatial range reconstruction from sparse observations
Sara Shahin, Axel G. Rossberg Poster B-S.B.31 · ECCB (European Conference on Computational Biology) · 2026
Sara Shahin, Jacob D. O'Sullivan, Axel G. Rossberg Submitted to Ecology Letters · bioRxiv preprint · 2026
DOI: https://doi.org/10.64898/2026.04.21.719942
3. Evaluation of Ensemble Learning for Mitigating Adversarial Attacks in Industrial Object Detection
Shashank Pulijala, Sara Shahin, Tapadhir Das, Raj Mani Shukla
Conference Paper · IEEE · 2024
DOI: 10.36227/techrxiv.171561173.31420188/v1
Preprint DOI: https://doi.org/10.36227/techrxiv.171561173.31420188/v1
| Award | Details | |
|---|---|---|
| 🏛️ | Alan Turing Institute — Data Study Group | Selected PhD researcher · Sept 2025 |
| 🌍 | Open Source Summit Europe 2025 | Fully funded scholarship · Amsterdam |
| 🧬 | Bertelsmann Scholarship | Genomic Data Science Track |
| 👩💻 | Women in Computer Science & Engineering | Sponsored Delegate · Milan |
- IBM AI Engineering Specialization
- IBM Data Science Specialization
- Supervised Machine Learning: Regression and Classification
- Machine Learning
- Foundations of Data Science: K-Means Clustering in Python
- Persian (native or bilingual)
- English (full professional)
- Arabic (limited working)
💡 Note: Most of my active development happens in private repositories for ongoing PhD research. Public contributions don't reflect the full scope of my work.
self.phd_focus = "Generative AI for spatio-temporal biodiversity prediction" self.methods = [ "MetaDiffusion: conditional diffusion model trained on simulated communities", "Individual-based models (IBM), ODE and PSD simulation", "GPU-accelerated simulation", "Uncertainty quantification from sparse observations" ] self.applications = [ "Species distribution prediction", "IUCN Red List metrics (area of occupancy, extent of occurrence)", "Conservation science" ]
