These are composed into a single vector representing position by a similarity-preserving, conjunctive vector-binding operation.

Advances in generative models have recently revolutionised machine learning. Meanwhile, in neuroscience, generative models have long been thought fundamental to animal intelligence. Understanding the biological mechanisms that support these processes promises to shed light on the relationship between biological and artificial intelligence. In animals, the hippocampal formation is thought to learn and use a generative model to support its role in spatial and non-spatial memory. Here we introduce a biologically plausible model of the hippocampal formation tantamount to a Helmholtz machine that we apply to a temporal stream of inputs. A novel component of our model is that fast theta-band oscillations (5-10 Hz) gate the direction of information flow throughout the network, training it akin to a high-frequency wake-sleep algorithm. Our model accurately infers the latent state of high-dimensional sensory environments and generates realistic sensory predictions. Furthermore, it can learn to path integrate by developing a ring attractor connectivity structure matching previous theoretical proposals and flexibly transfer this structure between environments.

These are composed into a single vector representing position by a similarity-preserving, conjunctive vector-binding operation.

Advances in generative models have recently revolutionised machine learning. Meanwhile, in neuroscience, generative models have long been thought fundamental to animal intelligence. Understanding the biological mechanisms that support these processes promises to shed light on the relationship between biological and artificial intelligence. In animals, the hippocampal formation is thought to learn and use a generative model to support its role in spatial and non-spatial memory. Here we introduce a biologically plausible model of the hippocampal formation tantamount to a Helmholtz machine that we apply to a temporal stream of inputs.

Advances in generative models have recently revolutionised machine learning. Meanwhile, in neuroscience, generative models have long been thought fundamental to animal intelligence. Understanding the biological mechanisms that support these processes promises to shed light on the relationship between biological and artificial intelligence. In animals, the hippocampal formation is thought to learn and use a generative model to support its role in spatial and non-spatial memory. Here we introduce a biologically plausible model of the hippocampal formation tantamount to a Helmholtz machine that we apply to a temporal stream of inputs.

Spatial cognition in hippocampal formation is posited to play a crucial role in the development of self-localization techniques for robots. In this paper, we propose a self-localization approach, DEQ-MCL, based on the discrete event queue hypothesis associated with phase precession within the hippocampal formation. Our method effectively estimates the posterior distribution of states, encompassing both past, present, and future states that are organized as a queue. This approach enables the smoothing of the posterior distribution of past states using current observations and the weighting of the joint distribution by considering the feasibility of future states. Our findings indicate that the proposed method holds promise for augmenting self-localization performance in indoor environments.

The hippocampal formation has been shown to be involved in spatial mapping (O'Keefe and Dostrovsky, 1971; Taube et al., 1990; Hafting et al., 2005). Analogies to geographic maps or computer graphics suggest that spatial mapping should involve associating a Euclidean space of unique location representations, for example Cartesian coordinates, with a set of locations in the real world. In neuroscience, such a Euclidean space is often theorized to be provided by entorhinal grid cells (Hafting et al., 2005). A single grid cell module can represent Euclidean space ambiguously, while an array of grid cells can theoretically create unique location representations (Fiete et al., 2008). In models that simulate spatial mapping of the brain at a low level, the population activity of multiple grid cell modules is often used in this way (Lewis et al., 2019; Whittington et al., 2020), or they use a single grid cell module to complement traditional Cartesian coordinates (Milford and Wyeth, 2008). These models imply that the brain maps space in a way that is fundamentally similar to metrical geographic maps or computer graphics.

A plethora of studies in rodents have described spatially tuned neurons, including place cells in the hippocampus and grid cells in the medial entorhinal cortex (MEC), suggesting a crucial role of the hippocampal formation in spatial navigation (1). Human studies have, in turn, shown that the hippocampal formation is involved in declarative memory (memories of facts and events) (2). What, then, is the function of the hippocampus? Is it involved in memory or in spatial navigation, or does it have a more general function that encompasses both? Several studies have shown that place cells remap, changing the location at which they respond, following geometrical changes in the environment, and that they can be modulated by nonspatial factors, according to the animals' specific tasks (3).