Discretization of continuous input spaces in the hippocampal autoencoder
This work addresses a fundamental problem in neuroscience by providing a unified framework for hippocampal functions, with potential applications in AI and cognitive modeling.
The study tackled the challenge of integrating spatial cognition and episodic memory by showing that discrete memories in sparse autoencoder neurons produce spatial tuning like hippocampal place cells, resulting in neurons tiling image and frequency spaces with minimal overlap and enabling reinforcement learning agents to perform visuo-spatial tasks effectively.
The hippocampus has been associated with both spatial cognition and episodic memory formation, but integrating these functions into a unified framework remains challenging. Here, we demonstrate that forming discrete memories of visual events in sparse autoencoder neurons can produce spatial tuning similar to hippocampal place cells. We then show that the resulting very high-dimensional code enables neurons to discretize and tile the underlying image space with minimal overlap. Additionally, we extend our results to the auditory domain, showing that neurons similarly tile the frequency space in an experience-dependent manner. Lastly, we show that reinforcement learning agents can effectively perform various visuo-spatial cognitive tasks using these sparse, very high-dimensional representations.