Video & world models
Understand how causal generation drifts across long rollouts, then intervene where memory and representation errors compound.
Open to research scientist and applied scientist opportunities
Zinuo (Henry) You
Generative systems
that hold their course.
I research video and world models, low-step generative inference, and continuous-time learning—turning ambitious models into systems that remain efficient, controllable, and measurable.
MODEL
TIME
NFE
ρ(t)
Learn the path.
Preserve the system.
ICML 2026Spotlight
AAAI 2026First author
ICASSP 2024First author
3 UKRI projectsResearch lead
Research agenda
Three connected ways to make generation more dependable.
Efficient inference
Learn where a fixed solver budget should be spent instead of accepting a uniform schedule as the default.
Continuous-time learning
Model irregular observations and latent trajectories without erasing the structure carried by time itself.
Selected work
Research told as systems, not abstracts.
Each case study starts with the failure mode, makes the modelling choice legible, and separates public evidence from ongoing work.
Causal Video & Interactive World Models
Diagnosing and reducing the representation drift that compounds across long autoregressive rollouts.
GenODE / GICO
Learning context-conditioned integration clocks for frozen generative ODE and flow-matching backbones.
Latent Laplace Diffusion
Parallel generative forecasting over irregular timestamps using stable latent trajectories.
Selected publications
Published research, with implementation close at hand.
2026
2026
2026
Manuscript · under-review
Refractive Clocks: Diffusion-to-Flow Schedule Transfer for Conditional Time Series Flow Matching
A temporal-reparameterization view of transferring diffusion schedules into frozen flow-matching systems.
ZYApproved portrait
to be added
to be added
About
Research direction,
engineering discipline.
I am a Computer Science PhD candidate at the University of Bristol, leading modelling and implementation across UKRI-funded projects. My work moves between mathematical formulation, large-scale experiments, and the release-quality code needed to make results useful.