Research Statement

My central interest is emergence—in particular, the emergence of intelligence, both artificial and natural.

Even more fundamentally, I seek to understand emergence in general: how molecules assemble into living cells, how neural circuits give rise to thought, and how order arises spontaneously from local interactions. I pursue these questions through three interconnected research directions:

Research Areas

  • → 🤖

    Science of (Artificial) Intelligence

    I seek to understand current artificial intelligence through empirical and theoretical investigation, especially aiming to explain interesting and puzzling deep learning phenomena.

    I have investigated how diffusion models learn to factorize and compose concepts, revealing mechanisms of compositionality in deep generative models.

    Currently, I'm exploring evolution strategies for finetuning large language models, examining how evolutionary approaches can complement traditional gradient-based methods for model adaptation and alignment.

  • → 🧠

    Neuroscience

    To understand biological intelligence, I build models investigating how different computational properties emerge from structural patterns arising from cortical wiring rules.

    Biological connectomes exhibit rich computational behavior despite being shaped by simple, local growth rules. Each neuron forms synapses guided by spatial gradients, cell-type-specific motifs, and plasticity constraints—yet collectively, these processes give rise to structured networks capable of memory, stability, and computation.

    In contrast, artificial neural networks are typically engineered by global optimization, abstracting away the mechanisms by which structure and function co-develop.

  • → 🌱

    Evolution/Artificial Life

    Beyond intelligence specifically, I'm interested in understanding emergence of self-organization and self-replication in general—the very foundation of life itself.

    I hope to crack the code for the necessary and sufficient conditions for self-organization and self-replication to emerge, using artificial life as a platform and evolution as a tool.

    My goal is to uncover the computational and dynamical principles that make such emergence possible—and to explore whether these principles can be instantiated in artificial systems.

Current Focuses

  • →

    Evolution Strategies for LLM Finetuning

    Understanding why evolution strategies with surprisingly small population can be applied to finetune LLMs of billions of parameters.

  • →

    Emergence of Computational Capabilities from Cortical Wiring Rules

    Studying how families of generative connectivity rules give rise to distinct dynamical and computational regimes. We construct synthetic neural networks that grow rather than are designed, governed by probabilistic rules parameterized by biologically inspired gradients and inhibitory motifs.

  • →

    Structure-Dynamics-Function Mapping

    Investigating how variations in wiring statistics produce measurable changes in system behavior, and which classes of structural rules give rise to computationally favorable dynamics.

Research Impact

My work seeks to build a theoretical science of emergence—one that treats self-organization and replication as formal computational phenomena.

Just as thermodynamics unified disparate physical processes under universal laws, I believe there exist underlying informational or dynamical invariants governing the spontaneous emergence of complexity. Understanding these principles could transform how we interpret the brain, design artificial systems, and even define life itself.

My research aims to make that understanding concrete: to uncover, in precise mathematical terms, how simple rules give rise to intelligent matter.