Finite-Time Corrective Transport for Unified Optimization and Distillation
Optimize a finite-step generative process through a corrective field over a frozen base transport.
PhD researcher · University of Bristol
I work on generative modelling, sequence modelling, and reinforcement learning. My research explores how to generate coherent sequences over long horizons, improve generative models through reinforcement learning and distillation, and make inference more efficient.
I am pursuing a PhD in Engineering Mathematics and Computer Science. My earlier work studies graph representation learning and Bayesian optimization.
Optimize a finite-step generative process through a corrective field over a frozen base transport.
Generate a whole forecast window at arbitrary timestamps, without stepping through physical time.
Study how representation drift compounds through causal rollout, and how models preserve motion, state, and control.
Choose GNN width and depth by treating representation learning as an information channel.
Combine multiple relation graphs with a retained feature branch for stock-trend representation learning.
Adjust variance pressure as an optimization budget is spent on an expensive, noisy black-box objective.
Keep feature transformation separate from relational mixing over a dynamic stock graph.
Extended version · LNCS 2025 · ICAART 2024 revised selected papers: Multi-Stock Movements Classification: A Decoupled Graph Diffusion Neural Network Approach
History-conditioned flow matching for generating level-2 limit order books.
Build traceable answers that connect retrieval, numerical reasoning, executable computation, and evidence.