Multi-scale place fields provide a predictive basis for hierarchical control
This paper introduces the factorised hierarchical successor representation (fHSR), a predictive map that unifies efficient hierarchical control with the multi-scale spatial organization of hippocampal place cells, successfully explaining their characteristics in large environments and predicting a novel size-independent spatial periodicity in population firing rates.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
The brain's hippocampus is often described as the seat of memory, but for decades, neuroscientists have also viewed it as a cognitive map, a mental GPS that allows animals to navigate their world. In the controlled, small rooms of a laboratory, this map appears simple: individual brain cells, known as place cells, fire only when an animal is in one specific spot, creating a single, tidy "on" switch for a location. However, the real world is vast and complex. When rats roam hundreds of meters or bats fly tens of kilometers between their roosts and feeding grounds, these same brain cells behave very differently. Instead of a single switch, they fire in multiple, irregular bursts across the landscape, creating a chaotic-looking pattern of activity that traditional theories of navigation have struggled to explain.
This disconnect between the neat laboratory picture and the messy reality of nature has left scientists wondering if the brain's map breaks down in large spaces or if our understanding of how it works is incomplete. A prevailing theory suggested that the brain builds a predictive map, essentially calculating where an animal is likely to end up in the future based on its current position. While this idea worked well for small rooms, it failed to account for the complex, multi-spot firing patterns seen in large environments. Researchers had to ask whether this irregularity was just random noise or if it served a specific, hidden purpose in how the brain plans and moves through the world.
A new study by a team of computational neuroscientists proposes that this irregularity is not a flaw, but a feature of a highly efficient navigation system. The researchers developed a new mathematical model called the factorised hierarchical successor representation. In simple terms, this model suggests that the brain does not just plan one step at a time. Instead, it organizes movement into a hierarchy of sub-goals, such as "go to the end of the hallway" or "turn at the doorway." By breaking a long journey into these manageable chunks, the brain can plan efficiently over vast distances without getting lost in the details of every single step.
The team showed that when a computer agent uses this hierarchical approach to learn, it naturally develops a map that looks exactly like the brain activity observed in real animals. In their simulations, the model produced place cells that fired in multiple spots, with the number of spots and their sizes changing in a predictable way as the environment grew larger. This matched the data collected from rats running on tracks up to 48 meters long and Egyptian fruit bats flying over distances of up to 200 meters. The model successfully reproduced the specific statistical patterns found in these animals, including the fact that some cells fire in many places while others fire in few, and that the spacing between these firing spots follows a specific, non-random distribution.
Crucially, the researchers tested whether this hierarchical structure offered a real advantage. They found that agents using this method could learn new routes and reach new goals much faster than agents using older, simpler models, especially as the environment grew larger. The hierarchical approach allowed the brain to generalize what it learned in one situation and apply it to another, a key trait of flexible intelligence. The model also predicted that the brain's population activity would show a subtle, repeating rhythm across space, a pattern that the researchers then confirmed by analyzing the actual firing rates of rat and bat neurons. This rhythm remained consistent regardless of how long the track was, suggesting it is a fundamental property of how the brain segments space.
The study also ruled out the idea that these complex firing patterns are merely the result of random biological noise. When the researchers simulated brain activity using a model based on random functions, it failed to produce the specific repeating rhythms and the unique geometric structure found in the real animal data. The hierarchical model, by contrast, generated these features naturally as a consequence of its design. This suggests that the irregular, multi-spot firing of place cells is a direct result of the brain organizing behavior into sub-goals to handle large, complex environments.
By linking the way animals move to the way their brain cells fire, this work provides a unified explanation for how the brain scales its navigation skills from a small room to a vast landscape. It implies that the brain's map is not a static picture of the world, but a dynamic, multi-layered system optimized for planning and efficiency. The findings suggest that the brain's ability to navigate the wild, complex world relies on breaking down the journey into smaller, manageable pieces, a strategy that turns what looks like chaotic noise into a precise and powerful tool for survival.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.