Topic: memory capacity

Neurons with recurrent connectivity can store memory patterns as attractor states in their dynamics, forming a plausible basis for associative memory in the brain. Classical theoretical results concerning fully connected recurrent neural networks (RNNs) with binary neurons and Hebbian learning rules find that they can store at most O(N) memories, where N is the number of neurons. However, under the physiological constraint that neurons are sparsely connected, this capacity is dramatically reduced to O(K), where K is the average number of presynaptic neurons (estimated to be O(10^3 ∼ 10^4) in m

A fundamental goal of neuroscience is the functional understanding of neural circuit architecture. Previous studies revealed the importance of network depth in visual and auditory circuits, but it remains elusive why predominantly non-convolutional circuits, such as the cerebellum or olfactory systems, have multiple feedforwardly connected layers. Recent deep learning works revealed that a two-hidden-layer network has a quadratically more memory capacity than a single-hidden-layer network with the same number of hidden layer neurons, suggesting the importance of the network depth for memory ca

An open problem in neuroscience is to explain the functional role of oscillations in neural networks, contributing, for example, to perception, attention, and memory. Cross-frequency coupling (CFC) is associated with information integration across populations of neurons [1]. Impaired CFC is linked to neurological disease [2]. It is unclear what role CFC has in information processing and brain functional connectivity. We construct a model of CFC which predicts a computational role for observed $\theta - \gamma$ oscillatory circuits in the hippocampus and cortex [3]. Our model predicts that the

It seems we see the world in full detail. However, the eye is not a camera nor is the brain a computer. Incredible metabolic constraints render us unable to encode more than a fraction of information available in each glance. Instead, our illusion of stable and complete perception is accomplished by parsimonious representation relying on natural order inherent in the surrounding environment. I will begin by discussing previous behavioral work from our lab demonstrating one such strategy by which the visual system represents average properties of Gestalt-grouped sets of individual objects, warp

Contrary to typical artificial neural network (ANN) design, biological neurons are not identical. Neurons differ substantially in their physiological properties. Heterogeneity has been hypothesized to increase the dimensionality of the neural dynamics, which improves the encoding properties of a network [1], promotes robustness and stability [2], and maximizes information flow in large networks [3]. We aim to show the functional effect of heterogeneity in rate-based recurrent neural networks. To vary the degree of heterogeneity, we introduce neuron types, with each neuron type having its own

A central question in systems neuroscience is what features of neural networks determine their memory capacity and whether these features are optimized in the brain. Here, we provide an analytical estimate of the memory capacity for a general class of network models. Our derivation extends previous theoretical results [1], which assumed homogeneous connectivity and coding levels (i.e., cell activation probabilities in the memory patterns), to models with arbitrary network architectures (i.e., different constraints on the arrangement of connections between cells) and heterogeneous coding levels

The brain stores a large number of memories, which are recalled through external stimuli or spontaneously during memory replay, a process essential for memory consolidation. Memory storage has traditionally been attributed to excitatory synaptic plasticity and the effect of inhibitory plasticity on memory capacity and recall remains underexplored. To elucidate how inhibitory plasticity affects memory capacity and recall, we compare two inhibitory plasticity rules in spiking autoassociative networks: a homeostatic plasticity rule and a rule that maintains firing rate heterogeneity. Our results

The self-ordered spatial search (SOSS) task is a touchscreen paradigm probing visuospatial working memory (SWM), that can be applied in non-human primate research in a preclinical translational setting. We examined the performance of the animals in the task on control baseline days and after a transient cognitive impairment induced by scopolamine, a muscarinic acetylcholine-receptor antagonist.Fourteen adult male rhesus monkeys performed the SOSS task daily, in 1-hour sessions with 300-500 trials each. In every trial the animals were shown a set of 4-8 identical cyan squares on a touchscreen,