Research overview
From compact quantum circuits to temporal prediction and risk allocation
My PhD investigates the expressivity, trainability, and practical deployment of Quantum Re-Uploading Units (QRUs) for streaming data and physical-system modelling.
Application layer
Applied QRU models
Particle classification and regression
I use single-qubit data re-uploading circuits for calorimetric particle identification and related regression tasks. The work studies how feature mappings, circuit depth, re-upload count, parameter sharing, optimisation choices, and NISQ constraints affect performance.
Time-series and streaming prediction
QRU models are evaluated on chaotic and environmental time series, with emphasis on low-parameter regimes, non-stationarity, spectral structure, and robust validation against classical baselines.
Climate-risk forecasting
A one-qubit QRU forecaster converts rainfall and river-level windows into short-term flood-risk indicators. These predictions feed downstream spatial risk-allocation objectives.
Theory layer
Theoretical foundations
Spectral and Fourier expressivity
I characterise how data encodings, trainable rotations, depth, and re-uploading determine the accessible frequency support of shallow quantum models. The goal is to connect circuit design to the spectral structure of the target function rather than treating architecture selection as a black-box search.
Trainability and gradients
I study gradient behaviour, parameter sharing, low-parameter architecture design, and the regimes in which additional depth improves representation or instead creates optimisation instability.
Coherent quantum pipelines
A further direction investigates how QRU outputs can be used by downstream quantum algorithms without reducing the architecture to an isolated prediction block.
Decision layer
Streaming and decision tasks
Reinforcement learning and sequential architectures
QRU embeddings and quantum recurrent components are studied in sequential decision settings, including passenger-load prediction and bus-headway regulation.
QRU–QAOA tail-risk allocation
The flood-risk project couples QRU predictions with a hydro–geo–social interaction model, a CVaR-based allocation objective, a QUBO representation, and a QAOA solver. The work explicitly evaluates where current NISQ hardware is viable and where classical baselines remain necessary.
Explainability under distribution shift
In an ongoing XAI collaboration, I analyse cross-day EMG gesture classification using attribution and attention-based diagnostics. The objective is to distinguish stable, physiologically plausible explanations from artefacts that arise under domain shift.