Gating by Memory: a Theory of Learning in the Cerebellum

This paper presents a model of learning by the cerebellar circuit. In the traditional and dominant learning model, training teaches finely graded parallel fibre synaptic weights which modify transmission to Purkinje cells and to interneurons that inhibit Purkinje cells. Following training, input in a learned pattern drives a training-modified response. The function is that the naive response to input rates is displaced by a learned one, trained under external supervision. In the proposed model, there is no weight-controlled graduated balance of excitation and inhibition of Purkinje cells. Instead, the balance has two functional states—a switch—at synaptic, whole cell and microzone level. The paper is in two parts. The first is a detailed physiological argument for the synaptic learning function. The second uses the function in a computational simulation of pattern memory. Against expectation, this generates a predictable outcome from input chaos (real-world variables). Training always forces synaptic weights away from the middle and towards the limits of the range, causing them to polarise, so that transmission is either robust or blocked. All conditions teach the same outcome, such that all learned patterns receive the same, rather than a bespoke, effect on transmission. In this model, the function of learning is gating—that is, to select patterns that trigger output merely, and not to modify output. The outcome is memory-operated gate activation which operates a two-state balance of weight-controlled transmission. Group activity of parallel fibres also simultaneously contains a second code contained in collective rates, which varies independently of the pattern code. A two-state response to the pattern code allows faithful, and graduated, control of Purkinje cell firing by the rate code, at gated times.

Superior colliculus saccade motor bursts do not dictate movement kinematics

The primate superior colliculus (SC) contains a topographic map of visual field locations, such that the anatomical location of any given active neuron defines a desired eye movement amplitude and direction. Complementing such a spatial code, SC neurons also exhibit saccade-related bursts that are tightly synchronized with movement onset. Current models suggest that such bursts, and their properties, constitute a temporal rate code that may dictate moment-to-moment movement evolution. However, a recent result demonstrated altered movement properties with minimal changes in SC motor burst strengths (Buonocore, Tian, Khademi, & Hafed, 2021). Here, I support such a dissociation between the SC temporal rate code and instantaneous movement evolution: SC burst strength varies depending on whether saccades are directed towards the upper or lower visual fields, but the movements themselves have similar kinematics. Thus, SC saccade-related motor bursts do not necessarily dictate movement kinematics, motivating investigating other possible functional roles for these bursts.

Costs and benefits of using rhythmic rate codes

Neural oscillations are one of the most well-known macroscopic phenomena observed in the nervous system, and the benefits of oscillatory coding have been the topic of frequent analysis. Many of these studies focused on communication between populations which were already oscillating, and sought to understand how synchrony and communication interact. In this paper, take an alternative approach. We focus on measuring the costs, and benefits, of moving to an from an aperiodic code to a rhythmic one. We utilize a Linear-Nonlinear Poisson model, and assume a rate code. We report that no one factor seems to predict the costs, or benefits, of translating into a rhythmic code. Instead the synaptic connection type, strength, population size, and stimulus and oscillation firing rates interact in nonlinear ways. We suggest a number of experiments that might be used to confirm these predictions.Author summaryIt’s good to oscillate, sometimes.

Does the Zebra Finch Mating Song Circuit Use Spike Times Efficiently?

Precise and reliable spike times are thought to subserve multiple possible functions, including improving the accuracy of encoding stimuli or behaviours relative to other coding schemes. Indeed, repeating sequences of spikes with sub-millisecond precision exist in nature, such as the synfire chain of spikes in area HVC of the zebra-finch mating-song circuit. Here, we analyzed what impact precise and reliable spikes have on the encoding accuracy for both the zebra-finch and more generic neural circuits using computational modelling. Our results show that neural circuits can use precisely timed spikes to encode signals with a higher-order accuracy than a conventional rate code. Circuits with precisely timed and reliably emitted spikes increase their encoding accuracy linearly with network size, which is the hallmark signature of an efficient code. This qualitatively differs from circuits that employ a rate code which increase their encoding accuracy with the square-root of network size. However, this improved scaling is dependent on the spikes becoming more accurate and more reliable with larger networks. Finally, we discuss how to test this scaling relationship in the zebra mating song circuit using both neural data and song-spectrogram-based recordings while taking advantage of the natural fluctuation in HVC network size due to neurogenesis. The zebra-finch mating-song circuit may represent the most likely candidate system for the use of spike-timing-based, efficient coding strategies in nature.

Dual-Factor Representation of the Environmental Context in the Retrosplenial Cortex

The retrosplenial cortex (RSC) is thought to be involved in a variety of spatial and contextual memory processes. However, we do not know how contextual information might be encoded in the RSC or whether the RSC representations may be distinct from context representations seen in other brain regions such as the hippocampus. We recorded RSC neuronal responses while rats explored different environments and discovered 2 kinds of context representations: one involving a novel rate code in which neurons reliably fire at a higher rate in the preferred context regardless of spatial location, and a second involving context-dependent spatial firing patterns similar to those seen in the hippocampus. This suggests that the RSC employs a unique dual-factor representational mechanism to support contextual memory.

Enhanced Signal Detection by Adaptive Decorrelation of Interspike Intervals

Spike trains with negative interspike interval (ISI) correlations, in which long/short ISIs are more likely followed by short/long ISIs, are common in many neurons. They can be described by stochastic models with a spike-triggered adaptation variable. We analyze a phenomenon in these models where such statistically dependent ISI sequences arise in tandem with quasi-statistically independent and identically distributed (quasi-IID) adaptation variable sequences. The sequences of adaptation states and resulting ISIs are linked by a nonlinear decorrelating transformation. We establish general conditions on a family of stochastic spiking models that guarantee this quasi-IID property and establish bounds on the resulting baseline ISI correlations. Inputs that elicit weak firing rate changes in samples with many spikes are known to be more detectible when negative ISI correlations are present because they reduce spike count variance; this defines a variance-reduced firing rate coding benchmark. We performed a Fisher information analysis on these adapting models exhibiting ISI correlations to show that a spike pattern code based on the quasi-IID property achieves the upper bound of detection performance, surpassing rate codes with the same mean rate—including the variance-reduced rate code benchmark—by 20% to 30%. The information loss in rate codes arises because the benefits of reduced spike count variance cannot compensate for the lower firing rate gain due to adaptation. Since adaptation states have similar dynamics to synaptic responses, the quasi-IID decorrelation transformation of the spike train is plausibly implemented by downstream neurons through matched postsynaptic kinetics. This provides an explanation for observed coding performance in sensory systems that cannot be accounted for by rate coding, for example, at the detection threshold where rate changes can be insignificant.

On Training Neural Network Decoders of Rate Compatible Polar Codes via Transfer Learning

Neural network decoders (NNDs) for rate-compatible polar codes are studied in this paper. We consider a family of rate-compatible polar codes which are constructed from a single polar coding sequence as defined by 5G new radios. We propose a transfer learning technique for training multiple NNDs of the rate-compatible polar codes utilizing their inclusion property. The trained NND for a low rate code is taken as the initial state of NND training for the next smallest rate code. The proposed method provides quicker training as compared to separate learning of the NNDs according to numerical results. We additionally show that an underfitting problem of NND training due to low model complexity can be solved by transfer learning techniques.

Subcircuits of deep and superficial CA1 place cells support efficient spatial coding across heterogeneous environments

The hippocampus is thought to guide navigation by forming a cognitive map of space. However, the behavioral demands for such a map can vary depending on particular features of a given environment. For example, an environment rich in cues may require a finer resolution map than an open space. It is unclear how the hippocampal cognitive map adjusts to meet these distinct behavioral demands. To address this issue, we examined the spatial coding characteristics of hippocampal neurons in mice and rats navigating different environments. We found that CA1 place cells located in the superficial sublayer were more active in cue-poor environments, and preferentially used a firing rate code driven by intra-hippocampal inputs. In contrast, place cells located in the deep sublayer were more active in cue-rich environments and expressed a phase code driven by entorhinal inputs. Switching between these two spatial coding modes was supported by the interaction between excitatory gamma inputs and local inhibition.