Sizhuang He

About Me

I work on the intersection of machine learning and biology:

LLMs and Agentic AI: Autonomous Systems for Biological Discovery
Generative Modeling: Flow Matching, Diffusion, Discrete Diffusion
Operator Learning: Modeling Continuous Spatiotemporal Dynamics, Integral Equations
Computational Biology: Single-cell Transcriptomics Data Analysis

[Sep 2025] Our paper Non-Markovian Discrete Diffusion with Causal Language Models is accepted to NeurIPS 2025 (San Diego, CA).
[Jul 2025] Our paper COAST: Intelligent Time-Adaptive Neural Operators is accepted to AI4MATH@ICML 2025 (Vancouver, Canada).
[Jan 2025] I received the Fan Family Fellowship of Yale University.
[Jan 2025] Our paper Intelligence at the Edge of Chaos is accepted to ICLR 2025 (Singapore).

Soft-Rank Diffusion

We propose Soft-Rank Diffusion, a discrete diffusion framework for learning distributions over permutations on $S_n$ that replaces shuffle-based random walks with a structured soft-rank forward process. By lifting permutations to continuous relaxed ranks, we obtain smoother and more tractable trajectories, while our contextualized generalized Plackett–Luce (cGPL) denoisers provide a prefix-conditional generalization of PL-style reverse models for inherently sequential decision structure. Across sorting and combinatorial optimization benchmarks, Soft-Rank Diffusion consistently outperforms prior permutation diffusion baselines, with especially strong gains on long sequences. — In Review

Cell2Sentence

C2S-Scale scales this framework to 27 billion parameters trained on a billion-token multimodal corpus—achieving state-of-the-art predictive and generative performance for complex, multicellular analyses. Visit the project page. Read more about this work in our blog post and another blog post. — In Review

CaDDI

We introduce a novel approach to discrete diffusion models that conditions on the entire generative trajectory, thereby lifting the Markov constraint and allowing the model to revisit and improve past states. CaDDi treats standard causal language models as a special case and permits the direct reuse of pretrained LLM weights with no architectural changes. — NeurIPS 2025

CaLMFlow

We present Volterra Flow Matching, a novel generative modeling framework that reformulates ODE-based flow matching frameworks with Volterra Integral Equations, hence avoiding a core challenge in ODE-based methods, known as stiffness. We show the connection between Volterra Integral Equations and causal transformers, the backbone of modern Large Language Models and hence demonstrates that causal language models can be naturally extended to generative modeling over continuous data domains through the lens of Volterra Flow Matching. — arXiv

Intelligence at the Edge of Chaos

By training LLMs on elementary cellular automata rules of varying complexity, we pinpoint a ‘sweet spot’ of data complexity that maximizes downstream predictive and reasoning abilities. Our findings suggest that exposing models to appropriately complex patterns is key to unlocking emergent intelligence. — ICLR 2025