Context codes and the effect of noisy learning on a simplified hippocampal CA3 model

This paper investigates how noise affects a minimal computational model of the hippocampus and, in particular, region CA3. The architecture and physiology employed are consistent with the known anatomy and physiology of this region. Here, we use computer simulations to demonstrate and quantify the ability of this model to create context codes in sequential learning problems. These context codes are mediated by local context neurons which are analogous to hippocampal place-coding cells. These local context neurons endow the network with many of its problem-solving abilities. Our results show that the network encodes context on its own and then uses context to solve sequence prediction under ambiguous conditions. Noise during learning affects performance, and it also affects the development of context codes. The relationship between noise and performance in a sequence prediction is simple and corresponds to a disruption of local context neuron firing. As noise exceeds the signal, sequence completion and local context neuron firing are both lost. For the parameters investigated, extra learning trials and slower learning rates do not overcome either of the effects of noise. The results are consistent with the important role played, in this hippocampal model, by local context neurons in sequence prediction and for disambiguation across time.     

Temporal Sequence Compression by an Integrate-and-Fire Model of Hippocampal Area CA3

Cells in the rat hippocampus fire as a function of the animal's location in space. Thus, a rat moving through the world produces a statistically reproducible sequence of place cell firings. With this perspective, spatial navigation can be viewed as a sequence learning problem for the hippocampus. That is, learning entails associating the relationships among a sequence of places that are represented by a sequence of place cell firing. Recent experiments by McNaughton and colleagues suggest the hippocampus can recall a sequence of place cell firings at a faster rate than it was experienced. This speedup, which occurs during slow-wave sleep, is called temporal compression. Here, we show that a simplified model of hippocampal area CA3, based on integrate-and-fire cells and unsupervised Hebbian learning, reproduces this temporal compression. The amount of compression is proportional to the activity level during recall and to the relative timespan of associativity during learning. Compression seems to arise from an alteration of network dynamics between learning and recall. During learning, the dynamics are paced by external input and slowed by a low overall level of activity. During recall, however, external input is absent, and the dynamics are controlled by intrinsic network properties. Raising the activity level by lowering inhibition increases the rate at which the network can transition between previously learned states and thereby produces temporal compression. The tendency for speeding up future activations, however, is limited by the temporal range of associations that were present during learning.     

Reservoir Computing Properties of Neural Dynamics in Prefrontal Cortex

Primates display a remarkable ability to adapt to novel situations. Determining what is most pertinent in these situations is not always possible based only on the current sensory inputs, and often also depends on recent inputs and behavioral outputs that contribute to internal states. Thus, one can ask how cortical dynamics generate representations of these complex situations. It has been observed that mixed selectivity in cortical neurons contributes to represent diverse situations defined by a combination of the current stimuli, and that mixed selectivity is readily obtained in randomly connected recurrent networks. In this context, these reservoir networks reproduce the highly recurrent nature of local cortical connectivity. Recombining present and past inputs, random recurrent networks from the reservoir computing framework generate mixed selectivity which provides pre-coded representations of an essentially universal set of contexts. These representations can then be selectively amplified through learning to solve the task at hand. We thus explored their representational power and dynamical properties after training a reservoir to perform a complex cognitive task initially developed for monkeys. The reservoir model inherently displayed a dynamic form of mixed selectivity, key to the representation of the behavioral context over time. The pre-coded representation of context was amplified by training a feedback neuron to explicitly represent this context, thereby reproducing the effect of learning and allowing the model to perform more robustly. This second version of the model demonstrates how a hybrid dynamical regime combining spatio-temporal processing of reservoirs, and input driven attracting dynamics generated by the feedback neuron, can be used to solve a complex cognitive task. We compared reservoir activity to neural activity of dorsal anterior cingulate cortex of monkeys which revealed similar network dynamics. We argue that reservoir computing is a pertinent framework to model local cortical dynamics and their contribution to higher cognitive function.     

Adjusting neurophysiological computations in the adult olfactory bulb

The olfactory bulb receives signals from olfactory sensory neurons and conveys them to higher centers. The mapping of the sensory inputs generates a reproducible spatial pattern in the glomerular layer of the olfactory bulb for each odorant. Then, this restricted activation is transformed into highly distributed patterns by lateral interactions between relay neurons and local interneurons. Thus, odor information processing requires the spatial patterning of both sensory inputs and synaptic interactions. In other words, odor representation is highly dynamic and temporally orchestrated. Here, we describe how the local inhibitory network shapes the global oscillations and the precise synchronization of relay neurons. We discuss how local inhibitory interneurons transpose the spatial dimension into temporal patterning. Remarkably, this transposition is not fixed but highly flexible to continuously optimize olfactory information processing.     

Real time computation: zooming in on population codes

Information processing in nervous systems intricately combines computation at the neuronal and network levels. Many computations may be envisioned as sequences of signal processing steps along some pathway. How can information encoded by single cells be mapped onto network population codes, and how do different modules or layers in the computation synchronize their communication and computation? These fundamental questions are particularly severe when dealing with real time streams of inputs. Here we study this problem within the context of a minimal signal perception task. In particular, we encode neuronal information by externally applying a space- and time-localized stimulus to individual neurons within a network. We show that a pulse-coupled recurrent neural network can successfully handle this task in real time, and obeys three key requirements: (i) stimulus dependence, (ii) initial-conditions independence, and (iii) accessibility by a readout mechanism. In particular, we suggest that the network's overall level of activity can be used as a temporal cue for a robust readout mechanism. Within this framework, the network can rapidly map a local stimulus onto a population code that can then be reliably read out during some narrow but well defined window of time.     

An olfactory neuronal network for vapor recognition in an artificial nose

Odorant sensitivity and discrimination in the olfactory system appear to involve extensive neural processing of the primary sensory inputs from the olfactory epithelium. To test formally the functional consequences of such processing, we implemented in an artificial chemosensing system a new analytical approach that is based directly on neural circuits of the vertebrate olfactory system. An array of fiber-optic chemosensors, constructed with response properties similar to those of olfactory sensory neurons, provide time-varying inputs to a computer simulation of the olfactory bulb (OB). The OB simulation produces spatiotemporal patterns of neuronal firing that vary with vapor type. These patterns are then recognized by a delay line neural network (DLNN). In the final output of these two processing steps, vapor identity is encoded by the spatial patterning of activity across units in the DLNN, and vapor intensity is encoded by response latency. The OB-DLNN combination thus separates identity and intensity information into two distinct codes carried by the same output units, enabling discrimination among organic vapors over a range of input signal intensities. In addition to providing a well-defined system for investigating olfactory information processing, this biologically based neuronal network performs better than standard feed-forward neural networks in discriminating vapors when small amounts of training data are used. Received: 30 June 1997 / Accepted in revised form: 12 January 1998     

Linking dynamics of the inhibitory network to the input structure

Networks of inhibitory interneurons are found in many distinct classes of biological systems. Inhibitory interneurons govern the dynamics of principal cells and are likely to be critically involved in the coding of information. In this theoretical study, we describe the dynamics of a generic inhibitory network in terms of low-dimensional, simplified rate models. We study the relationship between the structure of external input applied to the network and the patterns of activity arising in response to that stimulation. We found that even a minimal inhibitory network can generate a great diversity of spatio-temporal patterning including complex bursting regimes with non-trivial ratios of burst firing. Despite the complexity of these dynamics, the network’s response patterns can be predicted from the rankings of the magnitudes of external inputs to the inhibitory neurons. This type of invariant dynamics is robust to noise and stable in densely connected networks with strong inhibitory coupling. Our study predicts that the response dynamics generated by an inhibitory network may provide critical insights about the temporal structure of the sensory input it receives.     

Unsupervised learning of visual features through spike timing dependent plasticity

Spike timing dependent plasticity (STDP) is a learning rule that modifies synaptic strength as a function of the relative timing of pre- and postsynaptic spikes. When a neuron is repeatedly presented with similar inputs, STDP is known to have the effect of concentrating high synaptic weights on afferents that systematically fire early, while postsynaptic spike latencies decrease. Here we use this learning rule in an asynchronous feedforward spiking neural network that mimics the ventral visual pathway and shows that when the network is presented with natural images, selectivity to intermediate-complexity visual features emerges. Those features, which correspond to prototypical patterns that are both salient and consistently present in the images, are highly informative and enable robust object recognition, as demonstrated on various classification tasks. Taken together, these results show that temporal codes may be a key to understanding the phenomenal processing speed achieved by the visual system and that STDP can lead to fast and selective responses.     

Application of the recommendation architecture to telecommunications network management.

The recommendation architecture has been proposed as a system architecture which can enable a system to learn to perform a complex combination of interrelated functions. The capability of a system with the recommendation architecture to learn to manage complex telecommunication backbone networks has been investigated. A network model with a number of nodes and links and carrying realistic but randomly generated traffic was used as the target for the management system. Traffic data taken from the model was used as input to the recommendation architecture system. The traffic data was organized into inputs once every 5 minutes, and the management system organized these inputs into a hierarchy of repetition similarity. It was demonstrated that the outputs of this hierarchy provided information on the condition of the network. This output information was a compressed version of the inputs which correlated with major network conditions.     

Motor-related signals support localization invariance for stable visual perception

Our ability to perceive a stable visual world in the presence of continuous movements of the body, head, and eyes has puzzled researchers in the neuroscience field for a long time. We reformulated this problem in the context of hierarchical convolutional neural networks (CNNs)—whose architectures have been inspired by the hierarchical signal processing of the mammalian visual system—and examined perceptual stability as an optimization process that identifies image-defining features for accurate image classification in the presence of movements. Movement signals, multiplexed with visual inputs along overlapping convolutional layers, aided classification invariance of shifted images by making the classification faster to learn and more robust relative to input noise. Classification invariance was reflected in activity manifolds associated with image categories emerging in late CNN layers and with network units acquiring movement-associated activity modulations as observed experimentally during saccadic eye movements. Our findings provide a computational framework that unifies a multitude of biological observations on perceptual stability under optimality principles for image classification in artificial neural networks.     

Whole-Brain Mapping of Inputs to Projection Neurons and Cholinergic Interneurons in the Dorsal Striatum

The dorsal striatum integrates inputs from multiple brain areas to coordinate voluntary movements, associative plasticity, and reinforcement learning. Its projection neurons consist of the GABAergic medium spiny neurons (MSNs) that express dopamine receptor type 1 (D1) or dopamine receptor type 2 (D2). Cholinergic interneurons account for a small portion of striatal neuron populations, but they play important roles in striatal functions by synapsing onto the MSNs and other local interneurons. By combining the modified rabies virus with specific Cre- mouse lines, a recent study mapped the monosynaptic input patterns to MSNs. Because only a small number of extrastriatal neurons were labeled in the prior study, it is important to reexamine the input patterns of MSNs with higher labeling efficiency. Additionally, the whole-brain innervation pattern of cholinergic interneurons remains unknown. Using the rabies virus-based transsynaptic tracing method in this study, we comprehensively charted the brain areas that provide direct inputs to D1-MSNs, D2-MSNs, and cholinergic interneurons in the dorsal striatum. We found that both types of projection neurons and the cholinergic interneurons receive extensive inputs from discrete brain areas in the cortex, thalamus, amygdala, and other subcortical areas, several of which were not reported in the previous study. The MSNs and cholinergic interneurons share largely common inputs from areas outside the striatum. However, innervations within the dorsal striatum represent a significantly larger proportion of total inputs for cholinergic interneurons than for the MSNs. The comprehensive maps of direct inputs to striatal MSNs and cholinergic interneurons shall assist future functional dissection of the striatal circuits.     

A model linking immediate serial recall,the Hebb repetition effect and the learning of phonological word forms

We briefly review the considerable evidence for a common ordering mechanism underlying both immediate serial recall (ISR) tasks (e.g. digit span, non-word repetition) and the learning of phonological word forms. In addition, we discuss how recent work on the Hebb repetition effect is consistent with the idea that learning in this task is itself a laboratory analogue of the sequence-learning component of phonological word-form learning. In this light, we present a unifying modelling framework that seeks to account for ISR and Hebb repetition effects, while being extensible to word-form learning. Because word-form learning is performed in the service of later word recognition, our modelling framework also subsumes a mechanism for word recognition from continuous speech. Simulations of a computational implementation of the modelling framework are presented and are shown to be in accordance with data from the Hebb repetition paradigm.     

Mixed Signal Learning by Spike Correlation Propagation in Feedback Inhibitory Circuits

The brain can learn and detect mixed input signals masked by various types of noise, and spike-timing-dependent plasticity (STDP) is the candidate synaptic level mechanism. Because sensory inputs typically have spike correlation, and local circuits have dense feedback connections, input spikes cause the propagation of spike correlation in lateral circuits; however, it is largely unknown how this secondary correlation generated by lateral circuits influences learning processes through STDP, or whether it is beneficial to achieve efficient spike-based learning from uncertain stimuli. To explore the answers to these questions, we construct models of feedforward networks with lateral inhibitory circuits and study how propagated correlation influences STDP learning, and what kind of learning algorithm such circuits achieve. We derive analytical conditions at which neurons detect minor signals with STDP, and show that depending on the origin of the noise, different correlation timescales are useful for learning. In particular, we show that non-precise spike correlation is beneficial for learning in the presence of cross-talk noise. We also show that by considering excitatory and inhibitory STDP at lateral connections, the circuit can acquire a lateral structure optimal for signal detection. In addition, we demonstrate that the model performs blind source separation in a manner similar to the sequential sampling approximation of the Bayesian independent component analysis algorithm. Our results provide a basic understanding of STDP learning in feedback circuits by integrating analyses from both dynamical systems and information theory.     

Relative spike timing in stochastic oscillator networks of the Hermissenda eye

The role of relative spike timing on sensory coding and stochastic dynamics of small pulse-coupled oscillator networks is investigated physiologically and mathematically, based on the small biological eye network of the marine invertebrate Hermissenda. Without network interactions, the five inhibitory photoreceptors of the eye network exhibit quasi-regular rhythmic spiking; in contrast, within the active network, they display more irregular spiking but collective network rhythmicity. We investigate the source of this emergent network behavior first analyzing the role of relative input to spike–timing relationships in individual cells. We use a stochastic phase oscillator equation to model photoreceptor spike sequences in response to sequences of inhibitory current pulses. Although spike sequences can be complex and irregular in response to inputs, we show that spike timing is better predicted if relative timing of spikes to inputs is accounted for in the model. Further, we establish that greater noise levels in the model serve to destroy network phase-locked states that induce non-monotonic stimulus rate-coding, as predicted in Butson and Clark (J Neurophysiol 99:146–154, 2008a; J Neurophysiol 99:155–165, 2008b). Hence, rate-coding can function better in noisy spiking cells relative to non-noisy cells. We then study how relative input to spike–timing dynamics of single oscillators contribute to network-level dynamics. Relative timing interactions in the network sharpen the stimulus window that can trigger a spike, affecting stimulus encoding. Also, we derive analytical inter-spike interval distributions of cells in the model network, revealing that irregular Poisson-like spike emission and collective network rhythmicity are emergent properties of network dynamics, consistent with experimental observations. Our theoretical results generate experimental predictions about the nature of spike patterns in the Hermissenda eye.     

Implications of diffusion and time-varying morphogen gradients for the dynamic positioning and precision of bistable gene expression boundaries

The earliest models for how morphogen gradients guide embryonic patterning failed to account for experimental observations of temporal refinement in gene expression domains. Following theoretical and experimental work in this area, dynamic positional information has emerged as a conceptual framework to discuss how cells process spatiotemporal inputs into downstream patterns. Here, we show that diffusion determines the mathematical means by which bistable gene expression boundaries shift over time, and therefore how cells interpret positional information conferred from morphogen concentration. First, we introduce a metric for assessing reproducibility in boundary placement or precision in systems where gene products do not diffuse, but where morphogen concentrations are permitted to change in time. We show that the dynamics of the gradient affect the sensitivity of the final pattern to variation in initial conditions, with slower gradients reducing the sensitivity. Second, we allow gene products to diffuse and consider gene expression boundaries as propagating wavefronts with velocity modulated by local morphogen concentration. We harness this perspective to approximate a PDE model as an ODE that captures the position of the boundary in time, and demonstrate the approach with a preexisting model for Hunchback patterning in fruit fly embryos. We then propose a design that employs antiparallel morphogen gradients to achieve accurate boundary placement that is robust to scaling. Throughout our work we draw attention to tradeoffs among initial conditions, boundary positioning, and the relative timescales of network and gradient evolution. We conclude by suggesting that mathematical theory should serve to clarify not just our quantitative, but also our intuitive understanding of patterning processes.     

Balanced synaptic input shapes the correlation between neural spike trains

Stimulus properties, attention, and behavioral context influence correlations between the spike times produced by a pair of neurons. However, the biophysical mechanisms that modulate these correlations are poorly understood. With a combined theoretical and experimental approach, we show that the rate of balanced excitatory and inhibitory synaptic input modulates the magnitude and timescale of pairwise spike train correlation. High rate synaptic inputs promote spike time synchrony rather than long timescale spike rate correlations, while low rate synaptic inputs produce opposite results. This correlation shaping is due to a combination of enhanced high frequency input transfer and reduced firing rate gain in the high input rate state compared to the low state. Our study extends neural modulation from single neuron responses to population activity, a necessary step in understanding how the dynamics and processing of neural activity change across distinct brain states.     

Optimizing Artificial Neural Network-Based Models to Predict Rice Blast Epidemics in Korea

To predict rice blast, many machine learning methods have been proposed. As the quality and quantity of input data are essential for machine learning techniques, this study develops three artificial neural network (ANN)-based rice blast prediction models by combining two ANN models, the feed-forward neural network (FFNN) and long short-term memory (LSTM), with diverse input datasets, and compares their performance. The Blast_Weather_FFNN model had the highest recall score (66.3%) for rice blast prediction. This model requires two types of input data: blast occurrence data for the last 3 years and weather data (daily maximum temperature, relative humidity, and precipitation) between January and July of the prediction year. This study showed that the performance of an ANN-based disease prediction model was improved by applying suitable machine learning techniques together with the optimization of hyperparameter tuning involving input data. Moreover, we highlight the importance of the systematic collection of long-term disease data.     

Tinnitus-like “hallucinations” elicited by sensory deprivation in an entropy maximization recurrent neural network

Sensory deprivation has long been known to cause hallucinations or “phantom” sensations, the most common of which is tinnitus induced by hearing loss, affecting 10–20% of the population. An observable hearing loss, causing auditory sensory deprivation over a band of frequencies, is present in over 90% of people with tinnitus. Existing plasticity-based computational models for tinnitus are usually driven by homeostatic mechanisms, modeled to fit phenomenological findings. Here, we use an objective-driven learning algorithm to model an early auditory processing neuronal network, e.g., in the dorsal cochlear nucleus. The learning algorithm maximizes the network’s output entropy by learning the feed-forward and recurrent interactions in the model. We show that the connectivity patterns and responses learned by the model display several hallmarks of early auditory neuronal networks. We further demonstrate that attenuation of peripheral inputs drives the recurrent network towards its critical point and transition into a tinnitus-like state. In this state, the network activity resembles responses to genuine inputs even in the absence of external stimulation, namely, it “hallucinates” auditory responses. These findings demonstrate how objective-driven plasticity mechanisms that normally act to optimize the network’s input representation can also elicit pathologies such as tinnitus as a result of sensory deprivation.     

Encoding of Temporal Information by Timing,Rate, and Place in Cat Auditory Cortex

A central goal in auditory neuroscience is to understand the neural coding of species-specific communication and human speech sounds. Low-rate repetitive sounds are elemental features of communication sounds, and core auditory cortical regions have been implicated in processing these information-bearing elements. Repetitive sounds could be encoded by at least three neural response properties: 1) the event-locked spike-timing precision, 2) the mean firing rate, and 3) the interspike interval (ISI). To determine how well these response aspects capture information about the repetition rate stimulus, we measured local group responses of cortical neurons in cat anterior auditory field (AAF) to click trains and calculated their mutual information based on these different codes. ISIs of the multiunit responses carried substantially higher information about low repetition rates than either spike-timing precision or firing rate. Combining firing rate and ISI codes was synergistic and captured modestly more repetition information. Spatial distribution analyses showed distinct local clustering properties for each encoding scheme for repetition information indicative of a place code. Diversity in local processing emphasis and distribution of different repetition rate codes across AAF may give rise to concurrent feed-forward processing streams that contribute differently to higher-order sound analysis.     

Back-Propagation Operation for Analog Neural Network Hardware with Synapse Components Having Hysteresis Characteristics

To realize an analog artificial neural network hardware, the circuit element for synapse function is important because the number of synapse elements is much larger than that of neuron elements. One of the candidates for this synapse element is a ferroelectric memristor. This device functions as a voltage controllable variable resistor, which can be applied to a synapse weight. However, its conductance shows hysteresis characteristics and dispersion to the input voltage. Therefore, the conductance values vary according to the history of the height and the width of the applied pulse voltage. Due to the difficulty of controlling the accurate conductance, it is not easy to apply the back-propagation learning algorithm to the neural network hardware having memristor synapses. To solve this problem, we proposed and simulated a learning operation procedure as follows. Employing a weight perturbation technique, we derived the error change. When the error reduced, the next pulse voltage was updated according to the back-propagation learning algorithm. If the error increased the amplitude of the next voltage pulse was set in such way as to cause similar memristor conductance but in the opposite voltage scanning direction. By this operation, we could eliminate the hysteresis and confirmed that the simulation of the learning operation converged. We also adopted conductance dispersion numerically in the simulation. We examined the probability that the error decreased to a designated value within a predetermined loop number. The ferroelectric has the characteristics that the magnitude of polarization does not become smaller when voltages having the same polarity are applied. These characteristics greatly improved the probability even if the learning rate was small, if the magnitude of the dispersion is adequate. Because the dispersion of analog circuit elements is inevitable, this learning operation procedure is useful for analog neural network hardware.