Research

We study how intelligence emerges through learning under constraint. Real systems learn from limited data, noisy observations, uncertainty, changing environments, and local mechanisms of adaptation. These conditions shape how structure is extracted from experience, how representations stabilize, and how knowledge becomes reliable enough to guide action.

Our work develops theories and models of robust adaptive learning across machine learning, neuroscience, and cognition. We focus on the variables that govern this process—signal, noise, uncertainty, dimensionality, curvature, effective complexity, learning time, and control—and use them to understand how systems form internal models of the world. This includes research on uncertainty within deep models, local learning rules in neural systems, adaptation in embodied agents, and the dynamics of learning in high-dimensional regimes.

A central aim of the lab is to identify principles that travel across domains. Biological learning sharpens questions about mechanism, machine learning provides a formal language for testing them, and cognitive theory helps situate them within perception, thought, and behavior. Together, these lines of work form a unified research program centered on adaptive structure formation: how systems detect regularities, represent ambiguity, update under pressure, and remain stable while the world changes.

We extend the same perspective to scientific discovery itself. Discovery is a learning process shaped by hypothesis generation, evidence evaluation, uncertainty management, and iterative model revision. Our work on agentic science grows from this view. We build systems for inquiry by drawing on the same principles that govern minds and learning machines.

NightCity Labs is building a theory-driven NeuroAI program focused on the principles of learning, adaptation, and discovery.