RESUMEN
Reservoir Computing (RC) is a paradigm in artificial intelligence where a recurrent neural network (RNN) is used to process temporal data, leveraging the inherent dynamical properties of the reservoir to perform complex computations. In the realm of RC, the excitatory-inhibitory balance b has been shown to be pivotal for driving the dynamics and performance of Echo State Networks (ESN) and, more recently, Random Boolean Network (RBN). However, the relationship between b and other parameters of the network is still poorly understood. This article explores how the interplay of the balance b, the connectivity degree K (i.e., the number of synapses per neuron) and the size of the network (i.e., the number of neurons N) influences the dynamics and performance (memory and prediction) of an RBN reservoir. Our findings reveal that K and b are strongly tied in optimal reservoirs. Reservoirs with high K have two optimal balances, one for globally inhibitory networks (b < 0), and the other one for excitatory networks (b > 0). Both show asymmetric performances about a zero balance. In contrast, for moderate K, the optimal value being K = 4, best reservoirs are obtained when excitation and inhibition almost, but not exactly, balance each other. For almost all K, the influence of the size is such that increasing N leads to better performance, even with very large values of N. Our investigation provides clear directions to generate optimal reservoirs or reservoirs with constraints on size or connectivity.
RESUMEN
OBJECTIVE: Diseases such as age-related macular degeneration and retinitis pigmentosa cause the degradation of the photoreceptor layer. One approach to restore vision is to electrically stimulate the surviving retinal ganglion cells with a microelectrode array such as epiretinal implants. Epiretinal implants are known to generate visible anisotropic shapes elongated along the axon fascicles of neighboring retinal ganglion cells. Recent work has demonstrated that to obtain isotropic pixel-like shapes, it is possible to map axon fascicles and avoid stimulating them by inactivating electrodes or lowering stimulation current levels. Avoiding axon fascicule stimulation aims to remove brushstroke-like shapes in favor of a more reduced set of pixel-like shapes. APPROACH: In this study, we propose the use of isotropic and anisotropic shapes to render intelligible images on the retina of a virtual patient in a reinforcement learning environment named rlretina. The environment formalizes the task as using brushstrokes in a stroke-based rendering task. MAIN RESULTS: We train a deep reinforcement learning agent that learns to assemble isotropic and anisotropic shapes to form an image. We investigate which error-based or perception-based metrics are adequate to reward the agent. The agent is trained in a model-based data generation fashion using the psychophysically validated axon map model to render images as perceived by different virtual patients. We show that the agent can generate more intelligible images compared to the naive method in different virtual patients. SIGNIFICANCE: This work shares a new way to address epiretinal stimulation that constitutes a first step towards improving visual acuity in artificially-restored vision using anisotropic phosphenes.
Asunto(s)
Prótesis e Implantes , Retinitis Pigmentosa , Humanos , Retina/diagnóstico por imagen , Células Ganglionares de la Retina/fisiología , MicroelectrodosRESUMEN
Neural networks are powerful tools for solving complex problems, but finding the right network topology for a given task remains an open question. Biology uses neurogenesis and structural plasticity to solve this problem. Advanced neural network algorithms are mostly relying on synaptic plasticity and learning. The main limitation in reconciling these two approaches is the lack of a viable hardware solution that could reproduce the bottom-up development of biological neural networks. Here, we show how the dendritic growth of PEDOT:PSS-based fibers through AC electropolymerization can implement structural plasticity during network development. We find that this strategy follows Hebbian principles and is able to define topologies that leverage better computing performances with sparse synaptic connectivity for solving non-trivial tasks. This approach is validated in software simulation, and offers up to 61% better network sparsity on classification and 50% in signal reconstruction tasks.
RESUMEN
Reservoir computing provides a time and cost-efficient alternative to traditional learning methods. Critical regimes, known as the "edge of chaos," have been found to optimize computational performance in binary neural networks. However, little attention has been devoted to studying reservoir-to-reservoir variability when investigating the link between connectivity, dynamics, and performance. As physical reservoir computers become more prevalent, developing a systematic approach to network design is crucial. In this article, we examine Random Boolean Networks (RBNs) and demonstrate that specific distribution parameters can lead to diverse dynamics near critical points. We identify distinct dynamical attractors and quantify their statistics, revealing that most reservoirs possess a dominant attractor. We then evaluate performance in two challenging tasks, memorization and prediction, and find that a positive excitatory balance produces a critical point with higher memory performance. In comparison, a negative inhibitory balance delivers another critical point with better prediction performance. Interestingly, we show that the intrinsic attractor dynamics have little influence on performance in either case.
RESUMEN
The goal of the Acoustic Question Answering (AQA) task is to answer a free-form text question about the content of an acoustic scene. It was inspired by the Visual Question Answering (VQA) task. In this paper, based on the previously introduced CLEAR dataset, we propose a new benchmark for AQA, namely CLEAR2, that emphasizes the specific challenges of acoustic inputs. These include handling of variable duration scenes, and scenes built with elementary sounds that differ between training and test set. We also introduce NAAQA, a neural architecture that leverages specific properties of acoustic inputs. The use of 1D convolutions in time and frequency to process 2D spectro-temporal representations of acoustic content shows promising results and enables reductions in model complexity. We show that time coordinate maps augment temporal localization capabilities which enhance performance of the network by â¼ 17 percentage points. On the other hand, frequency coordinate maps have little influence on this task. NAAQA achieves 79.5% of accuracy on the AQA task with â¼ four times fewer parameters than the previously explored VQA model. We evaluate the performance of NAAQA on an independent data set reconstructed from DAQA. We also test the addition of a MALiMo module in our model on both CLEAR2 and DAQA. We provide a detailed analysis of the results for the different question types. We release the code to produce CLEAR2 as well as NAAQA to foster research in this newly emerging machine learning task.
RESUMEN
This study proposes voltage-dependent-synaptic plasticity (VDSP), a novel brain-inspired unsupervised local learning rule for the online implementation of Hebb's plasticity mechanism on neuromorphic hardware. The proposed VDSP learning rule updates the synaptic conductance on the spike of the postsynaptic neuron only, which reduces by a factor of two the number of updates with respect to standard spike timing dependent plasticity (STDP). This update is dependent on the membrane potential of the presynaptic neuron, which is readily available as part of neuron implementation and hence does not require additional memory for storage. Moreover, the update is also regularized on synaptic weight and prevents explosion or vanishing of weights on repeated stimulation. Rigorous mathematical analysis is performed to draw an equivalence between VDSP and STDP. To validate the system-level performance of VDSP, we train a single-layer spiking neural network (SNN) for the recognition of handwritten digits. We report 85.01 ± 0.76% (Mean ± SD) accuracy for a network of 100 output neurons on the MNIST dataset. The performance improves when scaling the network size (89.93 ± 0.41% for 400 output neurons, 90.56 ± 0.27 for 500 neurons), which validates the applicability of the proposed learning rule for spatial pattern recognition tasks. Future work will consider more complicated tasks. Interestingly, the learning rule better adapts than STDP to the frequency of input signal and does not require hand-tuning of hyperparameters.
RESUMEN
Despite decades of research, the features of an input audio stimulus that are encoded in an electroencephalogram (EEG) are still not clearly identified. We wish to investigate whether a frequency-band coupling model that estimates the cortical neural activity from EEGs can capture the important features of an input audio stimulus. To do so, EEG recordings were acquired from 8 subjects during a listening task where the vowels a, i and u were randomly presented. The neural activity was estimated from the EEG using a frequency-band coupling model that combined the EEG's phase in the delta band (2 Hz-4 Hz) and its amplitude in the gamma band (30 Hz-100 Hz). To investigate if the estimated neural activity could capture relevant features of an input audio stimulus, we fitted a generalized linear model (GLM) to the estimated neural activity and applied a statistical relative deviance metric to evaluate how important is the input audio stimulus in the estimated neural activity. We demonstrate that the input audio stimulus is the main component explaining the estimated neural activity and that other aspects such as the contribution of the surrounding network dynamics do not contribute significantly to the estimated neural activity. These results confirm that the features of the EEG used in the coupling model, namely the phase of the delta band and the power of the gamma band, do encode relevant aspects of an input audio signal. This non-invasive approach could be used, for example, to study how the presence of spectro-temporal features in the estimated neural activity is modified depending on different listening conditions or types of input sounds.
Asunto(s)
Percepción Auditiva , Electroencefalografía , HumanosRESUMEN
In the cortex, demarcated unimodal sensory regions often respond to unforeseen sensory stimuli and exhibit plasticity. The goal of the current investigation was to test evoked responses of primary visual cortex (V1) neurons when an adapting auditory stimulus is applied in isolation. Using extracellular recordings in anesthetized cats, we demonstrate that, unlike the prevailing observation of only slight modulations in the firing rates of the neurons, sound imposition in isolation entirely shifted the peaks of orientation tuning curves of neurons in both supra- and infragranular layers of V1. Our results suggest that neurons specific to either layer dynamically integrate features of sound and modify the organization of the orientation map of V1. Intriguingly, these experiments present novel findings that the mere presentation of a prolonged auditory stimulus may drastically recalibrate the tuning properties of the visual neurons and highlight the phenomenal neuroplasticity of V1 neurons.
Asunto(s)
Estimulación Acústica/métodos , Neuronas/fisiología , Orientación Espacial/fisiología , Estimulación Luminosa/métodos , Corteza Visual/fisiología , Animales , Gatos , Femenino , MasculinoRESUMEN
V1 is fundamentally grouped into columns that descend from layers II-III to V-VI. Neurons inherent to visual cortex are capable of adapting to changes in the incoming stimuli that drive the cortical plasticity. A principle feature called orientation selectivity can be altered by the presentation of non-optimal stimulus called 'adapter'. When triggered, LGN cells impinge upon layer IV and further relay the information to deeper layers via layers II-III. Using different adaptation protocols, neuronal plasticity can be investigated. Superficial neurons in area V1 are well acknowledged to exhibit attraction and repulsion by shifting their tuning peaks when challenged by a non-optimal stimulus called 'adapter'. Layers V-VI neurons in spite of partnering layers II-III neurons in cortical computation have not been explored simultaneously toward adaptation. We believe that adaptation not only affects cells specific to a layer but modifies the entire column. In this study, through simultaneous multiunit recordings in anesthetized cats using a multichannel depth electrode, we show for the first time how layers V-VI neurons (1000-1200 µm) along with layers II-III neurons (300-500 µm) exhibit plasticity in response to adaptation. Our results demonstrate that superficial and deeper layer neurons react synonymously toward adapter by exhibiting similar behavioral properties. The neurons displayed similar amplitude of shift and maintained equivalent sharpness of Gaussian tuning peaks before and the following adaptation. It appears that a similar mechanism, belonging to all layers, is responsible for the analog outcome of the neurons' experience with adapter.
Asunto(s)
Plasticidad Neuronal , Neuronas/fisiología , Corteza Visual/fisiología , Potenciales de Acción , Adaptación Fisiológica , Animales , Gatos , Femenino , Masculino , Estimulación Luminosa , Percepción VisualRESUMEN
Gamma oscillations are ubiquitous in brain and are believed to be inevitable for information processing in brain. Here, we report that distinct bands (low, 30-40Hz and high gamma, 60-80Hz) of stimulus-triggered gamma oscillations are systematically linked to the orientation selectivity index (OSI) of neurons in the cat primary visual cortex. The gamma-power is high for the highly selective neurons in the low-gamma band, whereas it is high for the broadly selective neurons in the high-gamma band. We suggest that the low-gamma band is principally implicated in feed-forward excitatory flow, whereas the high-gamma band governs the flow of this excitation.
Asunto(s)
Ritmo Gamma , Neuronas/fisiología , Corteza Visual/fisiología , Potenciales de Acción , Animales , Gatos , Estimulación LuminosaRESUMEN
Visual neurons coordinate their responses in relation to the stimulus; however, the complex interplay between a stimulus and the functional dynamics of an assembly still eludes neuroscientists. To this aim, we recorded cell assemblies from multi-electrodes in the primary visual cortex of anaesthetized cats in response to randomly presented sine-wave drifting gratings whose orientation tilted in 22.5° steps. Cross-correlograms revealed the functional connections at all the tested orientations. We show that a cell-assembly discriminates between orientations by recruiting a 'salient' functional network at every presented orientation, wherein the connections and their strengths (peak-probabilities in the cross-correlogram) change from one orientation to another. Within these assemblies, closely tuned neurons exhibited increased connectivity and connection-strengths compared with differently tuned neurons. Minimal connectivity between untuned neurons suggests the significance of neuronal selectivity in assemblies. This study reflects upon the dynamics of functional connectivity, and brings to the fore the importance of a 'signature' functional network in an assembly that is strictly related to a specific stimulus. It appears that an assembly is the major 'functional unit' of information processing in cortical circuits, rather than the individual neurons.
Asunto(s)
Neuronas/fisiología , Corteza Visual/fisiología , Percepción Visual/fisiología , Potenciales de Acción , Animales , Gatos , Femenino , Masculino , Estimulación LuminosaRESUMEN
Neural correlations (noise correlations and cross-correlograms) are widely studied to infer functional connectivity between neurons. High noise correlations between neurons have been reported to increase the encoding accuracy of a neuronal population; however, low noise correlations have also been documented to play a critical role in cortical microcircuits. Therefore, the role of noise correlations in neural encoding is highly debated. To this aim, through multi-electrodes, we recorded neuronal ensembles in the primary visual cortex of anaesthetized cats. By computing cross-correlograms, we divulged the functional network (microcircuit) between neurons within an ensemble in relation to a specific orientation. We show that functionally connected neurons systematically exhibit higher noise correlations than functionally unconnected neurons in a microcircuit that is activated in response to a particular orientation. Furthermore, the mean strength of noise correlations for the connected neurons increases steeply than the unconnected neurons as a function of the resolution window used to calculate noise correlations. We suggest that neurons that display high noise correlations in emergent microcircuits feature functional connections which are inevitable for information encoding in the primary visual cortex.
Asunto(s)
Red Nerviosa/fisiología , Neuronas/fisiología , Estimulación Luminosa/métodos , Corteza Visual/fisiología , Vías Visuales/fisiología , Animales , Gatos , ElectricidadRESUMEN
BACKGROUND: Within sensory systems, neurons are continuously affected by environmental stimulation. Recently, we showed that, on cell-pair basis, visual adaptation modulates the connectivity strength between similarly tuned neurons to orientation and we suggested that, on a larger scale, the connectivity strength between neurons forming sub-networks could be maintained after adaptation-induced-plasticity. In the present paper, based on the summation of the connectivity strengths, we sought to examine how, within cell-assemblies, functional connectivity is regulated during an exposure-based adaptation. RESULTS: Using intrinsic optical imaging combined with electrophysiological recordings following the reconfiguration of the maps of the primary visual cortex by long stimulus exposure, we found that within functionally connected cells, the summed connectivity strengths remain almost equal although connections among individual pairs are modified. Neuronal selectivity appears to be strongly associated with neuronal connectivity in a "homeodynamic" manner which maintains the stability of cortical functional relationships after experience-dependent plasticity. CONCLUSIONS: Our results support the "homeostatic plasticity concept" giving new perspectives on how the summation in visual cortex leads to the stability within labile neuronal ensembles, depending on the newly acquired properties by neurons.
Asunto(s)
Adaptación Fisiológica/fisiología , Red Nerviosa/fisiología , Plasticidad Neuronal/fisiología , Neuronas/fisiología , Reconocimiento Visual de Modelos/fisiología , Corteza Visual/fisiología , Animales , Gatos , Fenómenos Electrofisiológicos , Femenino , Masculino , Imagen ÓpticaRESUMEN
Visual processing in the cortex involves various aspects of neuronal properties such as morphological, electrophysiological and molecular. In particular, the neural firing pattern is an important indicator of dynamic circuitry within a neuronal population. Indeed, in microcircuits, neurons act as soloists or choristers wherein the characteristical activity of a 'soloist' differs from the firing pattern of a 'chorister'. Both cell types correlate their respective firing rate with the global populational activity in a unique way. In the present study, we sought to examine the relationship between the spike shape (thin spike neurons and broad spike neurons) of cortical neurons recorded from V1, their firing levels and their propensity to act as soloists or choristers. We found that thin spike neurons, which exhibited higher levels of firing, generally correlate their activity with the neuronal population (choristers). On the other hand, broad spike neurons showed lower levels of firing and demonstrated weak correlations with the assembly (soloists). A major consequence of the present study is: estimating the correlation of neural spike trains with their neighboring population is a predictive indicator of spike waveforms and firing level. Indeed, we found a continuum distribution of coupling strength ranging from weak correlation-strength (attributed to low-firing neurons) to high correlation-strength (attributed to high-firing neurons). The tendency to exhibit high- or low-firing is conducive to the spike shape of neurons. Our results offer new insights into visual processing by showing how high-firing rate neurons (mostly thin spike neurons) could modulate the neuronal responses within cell-assemblies.
Asunto(s)
Potenciales de Acción , Neuronas/fisiología , Corteza Visual/fisiología , Animales , GatosRESUMEN
Neuronal assemblies typically synchronise within the gamma oscillatory band (30-80 Hz) and are fundamental to information processing. Despite numerous investigations, the exact mechanisms and origins of gamma oscillations are yet to be known. Here, through multiunit recordings in the primary visual cortex of cats, we show that the strength of gamma power (20-40 and 60-80 Hz) is significantly stronger between the functionally connected units than between the unconnected units within an assembly. Furthermore, there is increased frequency coherence in the gamma band between the connected units than between the unconnected units. Finally, the higher gamma rhythms (60-80 Hz) are mostly linked to the fast-spiking neurons. These results led us to postulate that gamma oscillations are intrinsically generated between the connected units within cell assemblies (microcircuits) in relation to the stimulus within an emergent '50-ms temporal window of opportunity'.
Asunto(s)
Potenciales de Acción/fisiología , Ritmo Gamma/fisiología , Red Nerviosa/fisiología , Neuronas/fisiología , Periodicidad , Corteza Visual/citología , Animales , Gatos , Electroencefalografía , Vías Nerviosas/fisiología , Orientación/fisiología , Estimulación Luminosa , Estadística como AsuntoRESUMEN
Cortical organization rests upon the fundamental principle that neurons sharing similar properties are co-located. In the visual cortex, neurons are organized into orientation columns. In a column, most neurons respond optimally to the same axis of an oriented edge, that is, the preferred orientation. This orientation selectivity is believed to be absolute in adulthood. However, in a fully mature brain, it has been established that neurons change their selectivity following sensory experience or visual adaptation. Here, we show that after applying an adapter away from the tested cells, neurons whose receptive fields were located remotely from the adapted site also exhibit a novel selectivity in spite of the fact that they were not adapted. These results indicate a robust reconfiguration and remapping of the orientation domains with respect to each other thus removing the possibility of an orientation hole in the new hypercolumn. These data suggest that orientation columns transcend anatomy, and are almost strictly functionally dynamic.
Asunto(s)
Orientación , Corteza Visual/fisiología , Animales , Gatos , Electrofisiología , Femenino , Masculino , Neuronas , Estimulación LuminosaRESUMEN
We study the impact of different encoding models and spectro-temporal representations on the accuracy of Bayesian decoding of neural activity recorded from the central auditory system. Two encoding models, a generalized linear model (GLM) and a generalized bilinear model (GBM), are compared along with three different spectro-temporal representations of the input stimuli: a spectrogram and two bio-inspired representations, i.e. a gammatone filter bank (GFB) and a spikegram. Signal to noise ratios between the reconstructed and original representations are used to evaluate the decoding, or reconstruction accuracy. We experimentally show that the reconstruction accuracy is best with the spikegram representation and worst with the spectrogram representation and, furthermore, that using a GBM instead of a GLM significantly increases the reconstruction accuracy. In fact, our results show that the spikegram reconstruction accuracy with a GBM fitting yields an SNR that is 3.3 dB better than when using the standard decoding approach of reconstructing a spectrogram with GLM fitting.
Asunto(s)
Colículos Inferiores/fisiología , Modelos Neurológicos , Teorema de Bayes , Electrodos , Modelos Lineales , Neuronas/fisiología , Procesamiento de Señales Asistido por Computador , Relación Señal-RuidoRESUMEN
This article investigates the cross-modal correspondences between musical timbre and shapes. Previously, such features as pitch, loudness, light intensity, visual size, and color characteristics have mostly been used in studies of audio-visual correspondences. Moreover, in most studies, simple stimuli e.g., simple tones have been utilized. In this experiment, 23 musical sounds varying in fundamental frequency and timbre but fixed in loudness were used. Each sound was presented once against colored shapes and once against grayscale shapes. Subjects had to select the visual equivalent of a given sound i.e., its shape, color (or grayscale) and vertical position. This scenario permitted studying the associations between normalized timbre and visual shapes as well as some of the previous findings for more complex stimuli. One hundred and nineteen subjects (31 females and 88 males) participated in the online experiment. Subjects included 36 claimed professional musicians, 47 claimed amateur musicians, and 36 claimed non-musicians. Thirty-one subjects have also claimed to have synesthesia-like experiences. A strong association between timbre of envelope normalized sounds and visual shapes was observed. Subjects have strongly associated soft timbres with blue, green or light gray rounded shapes, harsh timbres with red, yellow or dark gray sharp angular shapes and timbres having elements of softness and harshness together with a mixture of the two previous shapes. Color or grayscale had no effect on timbre-shape associations. Fundamental frequency was not associated with height, grayscale or color. The significant correspondence between timbre and shape revealed by the present work allows designing substitution systems which might help the blind to perceive shapes through timbre.
RESUMEN
The interest in brain-like computation has led to the design of a plethora of innovative neuromorphic systems. Individually, spiking neural networks (SNNs), event-driven simulation and digital hardware neuromorphic systems get a lot of attention. Despite the popularity of event-driven SNNs in software, very few digital hardware architectures are found. This is because existing hardware solutions for event management scale badly with the number of events. This paper introduces the structured heap queue, a pipelined digital hardware data structure, and demonstrates its suitability for event management. The structured heap queue scales gracefully with the number of events, allowing the efficient implementation of large scale digital hardware event-driven SNNs. The scaling is linear for memory, logarithmic for logic resources and constant for processing time. The use of the structured heap queue is demonstrated on a field-programmable gate array (FPGA) with an image segmentation experiment and a SNN of 65,536 neurons and 513,184 synapses. Events can be processed at the rate of 1 every 7 clock cycles and a 406×158 pixel image is segmented in 200 ms.
Asunto(s)
Potenciales de Acción/fisiología , Computadores , Modelos Neurológicos , Redes Neurales de la Computación , Neuronas/fisiología , Procesamiento de Señales Asistido por Computador/instrumentación , Simulación por Computador , HumanosRESUMEN
Many studies explored mechanisms through which the brain encodes sensory inputs allowing a coherent behavior. The brain could identify stimuli via a hierarchical stream of activity leading to a cardinal neuron responsive to one particular object. The opportunity to record from numerous neurons offered investigators the capability of examining simultaneously the functioning of many cells. These approaches suggested encoding processes that are parallel rather than serial. Binding the many features of a stimulus may be accomplished through an induced synchronization of cell's action potentials. These interpretations are supported by experimental data and offer many advantages but also several shortcomings. We argue for a coding mechanism based on a sparse synchronization paradigm. We show that synchronization of spikes is a fast and efficient mode to encode the representation of objects based on feature bindings. We introduce the view that sparse synchronization coding presents an interesting venue in probing brain encoding mechanisms as it allows the functional establishment of multi-layered and time-conditioned neuronal networks or multislice networks. We propose a model based on integrate-and-fire spiking neurons.