Beyond task-coding dimensions
Activity outside task-coding dimensions drives transient amplification and can redirect decisions in models fitted to neural recordings.
Senior Scientist · Theory Lead
Neural Dynamics · Allen Institute
We build data-constrained theories of how neural circuits generate flexible behavior.
Activity outside task-coding dimensions drives transient amplification and can redirect decisions in models fitted to neural recordings.
Trial-specific input biases reshape one circuit’s dynamics, revealing transitions between engagement and disengagement. Under review at ICLR 2027.
Maximum-entropy inference separates connectivity structure required by neural dynamics from structure the data cannot identify.
As new memories are learned, older memories transition from stable to chaotic retrieval before they are forgotten.
Learning rules inferred from visual cortex support high-capacity associative memory with stable or chaotic retrieval.