1. Introduction
This project will develop a mathematical and algorithmic framework for adaptive sensing of environmental point processes.
Environmental monitoring often requires detecting and localising sparse events or sources: pollution spikes, particle emissions, acoustic detections, leaks, faults, animal calls, or other spatially distributed phenomena. These events are rarely observed directly. Instead, they are measured through noisy, indirect, and resource-constrained sensing systems such as fixed sensors, mobile platforms, UAVs, cameras, acoustic arrays, or sampling networks.
This project will develop a mathematical and algorithmic framework for adaptive sensing of environmental point processes. The hidden state will be modelled as a random marked point measure, representing event locations together with marks such as intensity, class, time, uncertainty, or physical magnitude. Observations will be modelled as outputs of controllable sensing operators: for example, the choice of sensor location, sampling time, viewing direction, acoustic aperture, mobile route, or measurement mode.
The central research question is:How should an AI system choose its next observation when the hidden object is a random spatial event process and sensing is costly?
The project will combine point-process modelling, Bayesian inference, active learning, information-driven control, and sparse recovery. A key aim is to reduce sensing, communication, and computation costs while preserving the ability to detect, localise, and quantify important environmental events. The student will develop methods for uncertainty-aware sensing policies based on posterior covariance, Fisher information, expected information gain, ambiguity reduction, or energy-aware utility functions.
Possible case studies include urban particle-emission mapping, acoustic biodiversity monitoring, low-power environmental sensor networks, or mobile robotic sensing. The project is mathematically grounded but application-facing: it will produce models and algorithms for deciding where, when, and how to sense under environmental and energy constraints.
Relevant background includes point-process theory, spatial statistics, Bayesian experimental design, active learning, sparse sensing, sensor management, and statistical signal processing.