Research
Papers
- Metropolis-Adjusted Diffusion Models
K. H. Lam*, T. Farghly*, C. Williams*, J. Yang, Y. W. Teh, A. Doucet
NeurIPS, 2026Mitigating discretisation bias of diffusion models, by adding Metropolis-Hastings adjustments to Langevin corrector steps. - Rao-Blackwellised Reparameterisation Gradients
K. H. Lam, T. D. Bui, G. Deligiannidis, Y. W. Teh
NeurIPS, 2025Rao-Blackwellising reparam gradients and casting local reparam gradients as Rao-Blackwellised reparam gradients for BNNs. - LegendreTron: Uprising Proper Multiclass Loss Learning
K. H. Lam, C. Walder, S. Penev, R. Nock
ICML, 2023Generalising generalised linear models by learning link functions with invertible neural networks.[Code]
Presentations
- Metropolis-Adjusted Diffusion ModelsMolSS Reading Group, Virtual, 2026
- Rao-Blackwellised Reparameterisation GradientsOxford Young Statisticians Seminar, Oxford, United Kingdom, 2025
- LegendreTron: Uprising Proper Multiclass Loss LearningTeh research group, Oxford, United Kingdom, 2024
- LegendreTron: Uprising Proper Multiclass Loss LearningDeligiannidis research group, Oxford, United Kingdom, 2024
Posters
- Metropolis-Adjusted Diffusion ModelsNeurIPS, Sydney, Australia, 2026
- Metropolis-Adjusted Diffusion ModelsISBA World Meeting, Nagoya, Japan, 2026
- Rao-Blackwellised Reparameterisation GradientsISBA World Meeting, Nagoya, Japan, 2026
- Rao-Blackwellised Reparameterisation GradientsLondon Meeting on Computational Statistics, London, United Kingdom, 2026
- Rao-Blackwellised Reparameterisation GradientsNeurIPS, San Diego, United States, 2025
- LegendreTron: Uprising Proper Multiclass Loss LearningICML, Honolulu, United States, 2023