Sen Lu and colleagues at the University of Michigan demonstrated that state space models can produce response profiles resembling time cells and oscillatory behaviors. The study suggests these patterns emerge from a structured multi-timescale complex rotational basis before any training occurs. This pre-existing temporal structure allows the models to adapt to specific tasks like event counting. The findings offer a computationally tractable framework for understanding how low-level neural phenomena give rise to abstract cognitive functions.
Read the full story at nature.com.
Photo: geralt / Pixabay
