HOrIc& RO7HrEs#uNl NB% GDjxO2Y0
> PhD at , advised by Caglar Gulcehre and Martin Jaggi
> Now: Student Researcher at Google DeepMind, London
I work on diffusion language models as a fast, controllable, and principled alternative to standard LLMs.
Research Focus
Bridging continuous and discrete diffusion
Diffusion geometry
Samplers and few-step generation
Efficient architectures for DLMs
Selected Work
Preprint 2026
Simplex Diffusion Models
Simplex Diffusion Models operate on the probability simplex and address the information collapse of discrete diffusion models.
TMLR 2026
Language Modeling with Hyperspherical Flows
饾晩-FLM corrupts normalized token embeddings with random rotations on the hypersphere, instead of using Gaussian noise.
ICLR 2026 路 oral
Partition Generative Modeling: Masked Modeling Without Masks
Partition Generative Models (PGMs) eliminate [MASK] tokens.
Latest News
- Sep. 2026
Language Modeling with Hyperspherical Flows was accepted at TMLR.
- Sep. 2026
Fixed-Point Masked Generative Modeling was accepted at NeurIPS 2026.
- Jul. 2026
I started a research internship at Google DeepMind in London!
