What we research
The human brain represents our most powerful computational system, processing information with remarkable efficiency while consuming minimal energy. We seek to understand the computational principles that make it a sophisticated thinking machine — and apply those insights to advance both our understanding of cognition and the technology we build. Despite significant advances in mapping neural circuitry, we are only beginning to understand the rich computational principles that give biological neural systems their unparalleled abilities, efficiency, and adaptability.
Understanding neural computation will transform both neuroscience and technology.
We believe that new discoveries about how neural circuits process information will significantly advance neuroscientific understanding — for both basic science and clinical applications — while inspiring technological innovations. Uncovering these principles can guide the development of more capable and efficient artificial systems, including neural networks and neuromorphic hardware with reduced energy consumption.
Three directions.
Our research develops mathematical and computational theories that build upon state-of-the-art discoveries in experimental neuroscience and neurobiology. We focus on three key areas: the centrality of active dendrites in computation and learning; the significance of temporal coordination and spike synchrony in neural coding; and the interaction of network-level coordination with completely local plasticity rules. To connect theory with data, we also develop specialized machine-learning tools for dynamical systems, particularly for systems that undergo sudden regime shifts or bifurcations.
Recent work.
Recent work has yielded several discoveries and methodological advances. We've shown how dendritic plateau potentials let neurons process spike sequences across multiple timescales, so a single neuron can detect complex temporal patterns with timing invariance. We've demonstrated how spike synchrony serves as a robust mechanism for encoding and processing information, particularly in recurrent networks shaped by experience-dependent plasticity. And on the methodological front, we've developed novel trajectory-based methods to train machine-learning models embedded in dynamical systems, overcoming traditional limitations in modeling bifurcating dynamics.
We put these results to work through collaborations on neuromorphic hardware design for efficient AI accelerators — benefiting energy-constrained applications and lowering the AI CO₂ footprint — and by developing new AI algorithms for neuromorphic sensors such as event-based cameras. Our machine-learning tools also support research in other domains, helping environmental scientists study the ecological dynamics of phytoplankton in the Baltic Sea.