Research profile
Adaptive AI that works in the real world
My research connects advances in machine learning with problems that matter in healthcare, neural engineering, organisations, and the environment. A recurring question guides the work: how can intelligent systems remain reliable when people, data, and operating conditions change?
Methodological foundations
Building intelligence for changing conditions
My methodological work provides a common foundation across the application areas below. It focuses on models that adapt, explain their reasoning, and operate efficiently beyond controlled laboratory settings.
Domain adaptation
Learning methods that remain useful when training and deployment data come from different distributions, subjects, devices, or settings.
Dataset-shift detection
Online and retrospective approaches for recognising covariate shift early and supporting reliable model adaptation.
Efficient vision transformers
Lightweight, explainable vision models trained for specialist domains, with particular emphasis on medical imaging.
Agentic & multimodal AI
Human-centred systems that combine language, vision, organisational knowledge, and accountable decision support.
Applied research clusters
From methods to measurable impact
Each cluster brings together a research problem, selected evidence, and the funded programmes through which the work is translated into practice.
01Adaptive & trustworthy AIDomain adaptation, dataset-shift detection and efficient visual intelligence
Domain adaptation for non-stationary systems
My research develops learning methods that remain effective when the statistical properties of data change between training and deployment. I have designed adaptive mechanisms that update a model’s understanding as new patterns emerge, improving resilience across subjects, devices, sessions, and real-world operating conditions.
This work has produced active and passive adaptation approaches, ensemble-learning frameworks, and methods for handling covariate shift in online systems (Raza et al., 2014; Raza et al., 2015; Raza et al., 2016; Raza & Rathee, 2018; Raza et al., 2019).
Change and covariate-shift detection
I contributed methods for identifying changes in a data distribution in both online and retrospective settings. These tests help determine when a deployed model can no longer rely on its original assumptions and when adaptation or retraining is required (Raza et al., 2013; Raza et al., 2013; Raza et al., 2015).
Efficient and explainable vision transformers
With my doctoral researchers, I investigate lightweight Vision Transformers that can be trained without dependence on generic ImageNet weights. The work focuses on data-efficient learning, interpretability, robustness, and specialist architectures for medical images and other resource-constrained domains (Ali et al., 2024; Ali et al., 2025; Ali et al., 2026).
Trustworthy, agentic and multimodal AI
My current work extends these foundations to human-centred systems combining language, vision, organisational knowledge, and decision support. The emphasis is on explainability, fairness, dependable evaluation, and responsible adoption rather than capability alone.
02AI for health & clinical decisionsEarlier diagnosis, intelligent triage and explainable medical vision
Skin-cancer detection
In partnership with Check4Cancer, I led research to develop an AI engine for melanoma detection using dermoscopic and digital images collected over two decades. The work advances explainable computer vision and efficient Vision Transformers, with particular attention to robustness and reducing false-negative decisions in a mobile diagnostic pathway (Islam et al., 2024; Islam et al., 2024; Islam et al., 2026).
AI-assisted NHS patient triage
With Provide Community, our team designed, developed, and deployed a Smart Referral System for musculoskeletal services. The architecture supports real-time triage, directs patients to appropriate clinical pathways, standardised referral routes across 12 clinical services, and reduced decisions that previously took weeks or months to a matter of hours. The partnership received Best Associate and Best Academic recognition.
Forensic dentistry and medical imaging
I collaborate internationally on AI for dental radiography, periodontology, and forensic dentistry. Contributions include age and sex estimation from molar and panoramic radiographs, mental-foramen detection and segmentation, dental-disease classification, and the development of structured imaging datasets (Raza et al., 2024; Raza et al., 2024; Malik et al., 2026).
Predictive health analytics
My wider health research includes intelligent clinical pathways, decision support using electronic health records, and population-health analysis. A longitudinal study linked accelerometer and health-record data to identify gestational and postnatal factors associated with physical activity in 12-month-old infants (Raza et al., 2017).
03Applied & responsible AIEnvironment, enterprise, engineering and public value
Environment, agriculture and conservation
I apply machine learning, computer vision, graph methods, and connected sensing to forest-fire prediction, biodiversity monitoring, crop analysis, water and ecosystem wellbeing, and evidence-led policy. My contributions include IoT-based early-warning approaches and event-detection methods for environmental monitoring (Singh et al., 2022; Singh et al., 2024; Pandey et al., 2023).
Enterprise, engineering and operational AI
Through Knowledge Transfer Partnerships and collaborative R&D, I translate research into dependable products and organisational processes. This includes intelligent construction planning with real-time information delivery, engineering and manufacturing analytics, logistics optimisation, revenue intelligence, generative AI, and assessment of organisational readiness for responsible adoption.
The work combines technical development with knowledge exchange so that models are embedded into practical workflows, evaluated with domain experts, and capable of delivering sustained organisational value.
04Neural engineering & brain–computer interfacesAdaptive learning for EEG, MEG, motor imagery and rehabilitation
Adaptive EEG-based BCI
I helped develop a covariate-shift detection test and adaptive learning algorithms for non-stationary EEG data (Raza et al., 2015; Raza et al., 2016). I subsequently proposed a covariate-shift-estimation adaptive ensemble for motor-imagery BCI (Raza et al., 2019) and evaluated automatic frequency-band-selection methods for learning discriminative Common Spatial Pattern features (Raza et al., 2015). The resulting online system has continued to support research in the BCI laboratory at Ulster University.
Open MEG data and cognitive decoding
Our team released a public 306-channel MEG dataset recorded from 17 participants across hand, feet, subtraction, and word-generation imagery tasks, enabling wider development and comparison of MEG pattern-recognition methods (Rathee et al., 2021). International collaboration also produced new evidence on how numeracy and literacy are represented in the human brain (Nara et al., 2023).
Personalised stroke rehabilitation
With collaborators at IIT Kanpur, I contributed to a BCI-operated neurorehabilitation system for recovering finger movement after stroke. We developed cortico-muscular-coupling and EEG–EMG feature methods, integrated covariate-shift adaptation, and evaluated the system with healthy participants and stroke survivors (Chowdhury et al., 2019; Chowdhury et al., 2017; Chowdhury et al., 2018).
Research philosophy
Rigorous methods. Responsible translation. Enduring partnerships.
The strongest research moves fluently between theory and practice. My work combines methodological depth, interdisciplinary collaboration, and sustained engagement with the people who will use—or be affected by—the resulting technology.