Topic: recurrent connections

A recent approach to understanding the mammalian visual system is to show correspondence between the sequential stages of processing in the ventral stream with layers in a deep convolutional neural network (DCNN), providing evidence that visual information is processed hierarchically, with successive stages containing ever higher-level information. However, correspondence is usually defined as shared variance between brain region and model layer. We propose that task-relevant variance is a stricter test: If a DCNN layer corresponds to a brain region, then substituting the model’s activity with

The hippocampal formation plays a key role in learning and memory, with CA3 thought to be the substrate of pattern completion [1]. However, its underlying computational and synaptic mechanisms are not yet understood. Here we analyze the mechanisms behind sequence replay through simulations of spiking neurons and a time-discrete model of moving membrane potential distributions. First, we simulated spiking networks of excitatory leaky integrate-and-fire cells forming a sequence of assemblies. The sequence contains both recurrent and feedforward connections (within and across assemblies, respect

A hallmark of biological and artificial neural networks is that neurons tile the range of continuous sensory inputs and intrinsic variables with overlapping responses. It is characteristic for the underlying recurrent connectivity in the cortex that neurons with similar tuning predominantly excite each other. The reason for such an architecture is not clear. Using an analytically tractable model, we show that it can naturally arise from a cooperative coding scheme. In this scheme neurons with similar responses specifically support each other by sharing their computations to obtain the desired

A head direction (HD) cell fires when an animal faces a particular azimuthal direction (Taube and Bassett 2003). Recurrent connections among HD cells create a ring-attractor network that continuously encodes HD (Ajabi et al. 2023). Indeed, in the Drosophila HD system, three rings interact to linearly integrate angular head velocity (AHV), with two of the rings receiving countervailing velocity-dependent inputs (Turner-Evans et al. 2017). In contrast, for the desert locust an alternative mechanism has been proposed (Zittrell et al. 2023; Pabst et al. 2024). There, velocity modulates synaptic

Animals continuously process information from multiple sensory modalities to assess their environment and guide their behavior. For example, a predator may rely on both visual and auditory cues to track its prey. Over time, various algorithms have been developed to describe multisensory processing, including linear and nonlinear fusion. However, these algorithms typically treat each time step independently and fail to account for temporal dependencies in these signals. To address this limitation, we introduce a novel set of multisensory tasks that systematically incorporate controlled temporal

With the recent success of Transformers in the field of automatic natural language processing [1], it has become clear that we are able to create complex, high-performance neural network architectures that can extract, combine and transform information to meet a specific objective, such as predicting the next word in a text. However, to function, Transformers need to receive a context of all N elements of the past sequence (N = 2k to 32k tokens) at each time step, causing the number of computations to grow quadratically (N^2) with the length of the sequence [1]. This "perfect" memory of a co