Search

In this article, we describe the architecture of the LIMA (Libre Multilingual Analyzer) framework and its recent evolution with the addition of new text analysis modules based on deep neural networks. We extended the functionality of LIMA in terms of the number of supported languages while preserving existing configurable architecture and the availability of previously developed rule-based and statistical analysis components. Models were trained for more than 60 languages on the Universal Dependencies 2.5 corpora, WikiNer corpora, and CoNLL-03 dataset. Universal Dependencies allowed us to increase the number of supported languages and generate models that could be integrated into other platforms. This integration of ubiquitous Deep Learning Natural Language Processing models and the use of standard annotated collections using Universal Dependencies can be viewed as a kind of model and data interoperability, complementary to the technical interoperability between systems. [ABSTRACT FROM AUTHOR]

This paper presents a procedure for and evaluation of using a semantic similarity metric as a loss function for neural source code summarization. Code summarization is the task of writing natural language descriptions of source code. Neural code summarization refers to automated techniques for generating these descriptions using neural networks. Almost all current approaches involve neural networks as either standalone models or as part of a pretrained large language models, for example, GPT, Codex, and LLaMA. Yet almost all also use a categorical cross‐entropy (CCE) loss function for network optimization. Two problems with CCE are that (1) it computes loss over each word prediction one‐at‐a‐time, rather than evaluating a whole sentence, and (2) it requires a perfect prediction, leaving no room for partial credit for synonyms. In this paper, we extend our previous work on semantic similarity metrics to show a procedure for using semantic similarity as a loss function to alleviate this problem, and we evaluate this procedure in several settings in both metrics‐driven and human studies. In essence, we propose to use a semantic similarity metric to calculate loss over the whole output sentence prediction per training batch, rather than just loss for each word. We also propose to combine our loss with CCE for each word, which streamlines the training process compared to baselines. We evaluate our approach over several baselines and report improvement in the vast majority of conditions. [ABSTRACT FROM AUTHOR]

We demonstrate a model of chirp-velocity sensitivity in the inferior colliculus (IC) that retains the tuning to amplitude modulation (AM) that was established in earlier models. The mechanism of velocity sensitivity is sequence detection by octopus cells of the posteroventral cochlear nucleus, which have been proposed in physiological studies to respond preferentially to the order of arrival of cross-frequency inputs of different amplitudes. Model architecture is based on coincidence detection of a combination of excitatory and inhibitory inputs. Chirp-sensitivity of the IC output is largely controlled by the strength and timing of the chirp-sensitive octopus-cell inhibitory input. AM tuning is controlled by inhibition and excitation that are tuned to the same frequency. We present several example neurons that demonstrate the feasibility of the model in simulating realistic chirp-sensitivity and AM tuning for a wide range of characteristic frequencies. Additionally, we explore the systematic impact of varying parameters on model responses. The proposed model can be used to assess the contribution of IC chirp-velocity sensitivity to responses to complex sounds, such as speech. [ABSTRACT FROM AUTHOR]

In M'bahiakro, nitrate contamination of drinking well water is becoming a cause for concern and continues despite the efforts made in the town. To help monitor these waters, this study aims to predict nitrate concentrations in the M'bahiakro water table based on physico-chemical parameters measured in situ. To this end, a gradient error back-propagation (BPNN) artificial neural network (ANN) was developed to simulate nitrate concentrations using temperature (T), electrical conductivity (EC), dissolved oxygen (O2), redox potential (Eh) and well water depth as input data. The resulting dataset was divided into two parts to form the artificial neural network, where 70% of the dataset was used for training, and the remaining 30% was also divided into two equal parts: one for testing and the other for model validation. The models were configured using a constructive approach, which consists of testing each input variable individually in a reference network and combining the variables until the best intelligent model is obtained according to the chosen performance criteria. The intelligent models obtained were evaluated on the basis of the coefficient of determination (R2) closest to 1 and the lowest mean square error (MSE). The results obtained showed that the BPNN models developed using four input variables in the dry and rainy seasons provided the best results. The MSE and R2 values were around 0.01 mg/L and 95%, respectively. They are obviously more accurate since the mean square errors are low with coefficients of determination close to unity. The BPNN models thus obtained were able to reproduce satisfactorily the nitrate concentrations obtained experimentally in 19 wells in the town of M'bahiakro. However, it is essential to continue this study in order to define the time interval during which the BPNN models obtained can remain valid in terms of performance in the M'bahiakro area. [ABSTRACT FROM AUTHOR]

This article introduces the enhanced Runtime Analyzing and Visualization Simulator (RAVSim) v2.0, a graphical tool that not only supports SNN design and analysis but also facilitates a comprehensive comparative analysis of various SNN models. The new version of RAVSim introduces a groundbreaking feature enabling users to conduct in-depth comparisons of SNN models, enhancing understanding and aiding in model selection for specific applications. Furthermore, with the updated version of RAVSim, researchers, and developers can effortlessly generate trained model weights using a custom dataset, eliminating the need to investigate or write complicated backend code. This new feature facilitates the seamless integration of diverse datasets, streamlining the process for further analysis and exploration. Therefore, the developers can now focus on high-level tasks and gain a clear understanding of SNN without worrying about the technical complexities of weight generation. This advancement represents a significant step towards making SNNs more accessible and user-friendly, unlocking their full potential in artificial intelligence and computational neuroscience applications. Furthermore, RAVSim’s code has undergone extensive optimization and debugging, leading to a substantial ∼65% reduction in image classification simulation time compared to the previous RAVSim version. This improvement makes it easier and quicker to train models and generate weights.

Abstract Time series analysis and prediction have attained significant attention from the research community in the past few decades. However, the prediction accuracy of the models highly depends on the models’ learning process. In order to optimize resource usage, a better learning methodology, in terms of accuracy and learning time, is needed. In this context, the current research work proposes EvoLearn, a novel method to improve and optimize the learning process of neural-based models. The presented technique integrates the genetic algorithm with back-propagation to train model weights during the learning process. The fundamental idea behind the proposed work is to select the best components from multiple models during the training process to obtain an adequate model. To demonstrate the applicability of EvoLearn, the method is tested on the state-of-the-art neural models (namely MLP, DNN, CNN, RNN, and GRU), and performances are compared. Furthermore, the presented study aims to forecast two types of time series, i.e. air pollution and energy consumption time series, using the developed framework. In addition, the considered neural models are tested on two datasets of each time series type. From the performance comparison and evaluation of EvoLearn using a one-tailed paired T-test against the conventional back-propagation-based learning approach, it was found that the proposed method significantly improves the prediction accuracy.

In this experimental study, the erosion wear behavior of glass fibre-reinforced (GF) composite materials was examined according to ASTM G76–95. Pure/GF reinforced epoxy composite (EP) materials were chosen as the main test sample. Boric Acid (H3BO3), Borax (B2O3), Silicon Dioxide (SiO2), and Aluminium Oxide (Al2O3) were added to the resin as reinforcement at a rate of 15% by weight. The erosion wear rate was investigated with various impingement angles (30°, 60°, and 90°), impact velocities (≈23, 34, and 53 m/s), alumina abrasive particle sizes (≈200 and 400 μm), and fibre directions (0° and 45°). Neural network models were employed effectively to predict the influence of the reinforcements on erosive wear rate. The erosive wear rate indicated that Al2O3 added GF/EP exhibited the most anti-erosive characteristics followed by silicon dioxide GF/EP and pure GF/EP however, the anti-erosive nature of GF/EP deteriorated with the addition of Borax and Boric Acid. [ABSTRACT FROM AUTHOR]

The lanasyn black is among the most often used in manufacturing and is challenging to take out during the treatment of wastewaters from textile industry. The cloud point extraction was used for their elimination from an aqueous solution. The multivariable process parameters have been independently optimized using the central composite design and the Levenberg-Marquardt algorithm-based artificial neural network for the highest yield of the extraction of lanasyn black via the cloud point extraction. The CCD forecasts the output maximum of 97.01 % under slightly altered process parameters. Still, the ANN-LMA model predicts the extraction yield (99.98 %) using 1.04 g of KNO3, the beginning pH of solution 8.99, the initial content of lanasyn black 24.57 ppm and 0.34 mass % of Triton X-100. With the coefficients of determination of 0.997 and 0.9777, the most recent empirical verification of the model mentioned above predictions using CCD and ANN-LMA is determined to be satisfactory. [ABSTRACT FROM AUTHOR]

Time series analysis and prediction have attained significant attention from the research community in the past few decades. However, the prediction accuracy of the models highly depends on the models' learning process. In order to optimize resource usage, a better learning methodology, in terms of accuracy and learning time, is needed. In this context, the current research work proposes EvoLearn, a novel method to improve and optimize the learning process of neural-based models. The presented technique integrates the genetic algorithm with back-propagation to train model weights during the learning process. The fundamental idea behind the proposed work is to select the best components from multiple models during the training process to obtain an adequate model. To demonstrate the applicability of EvoLearn, the method is tested on the state-of-the-art neural models (namely MLP, DNN, CNN, RNN, and GRU), and performances are compared. Furthermore, the presented study aims to forecast two types of time series, i.e. air pollution and energy consumption time series, using the developed framework. In addition, the considered neural models are tested on two datasets of each time series type. From the performance comparison and evaluation of EvoLearn using a one-tailed paired T-test against the conventional back-propagation-based learning approach, it was found that the proposed method significantly improves the prediction accuracy. [ABSTRACT FROM AUTHOR]

This paper discusses a real-time approach for capturing and recognizing sign language gestures by efficient vision techniques and deep neural models. The key data constraint of limited availability of comprehensive sign corpora is tackled through configurable accumulation of hand image samples from continuous video. Configurable interfaces enabled collection of diverse sign samples spanning the alphabet as well as additional signs like “Good”, “Bad”, “Nice”, “Little” and “Stop”. The presented customizable interface enables triggering scheduled collection protocols, while focusing bounding box extraction algorithms only on active signing areas alleviates storage needs. Novel touch-less tracking by homegrown computer vision algorithms also promotes inclusion. Created samples receive augmentation including generative and projective transformations promoting variability and reduced bias. The models trained thereafter demonstrate state-of-the-art performance on internal benchmarks that surpass previous academic attempts in the domain. Qualitative assessments by independent native interpreters provide encouraging indicators on real-world viability. This expandable architecture via parameterized logging protocols, paired with customizable assembly of training data shows promise in transitioning sign recognition from controlled settings to unconstrained environments. Easy replicability also enables rapid upgrading with new vocabulary and concepts. Future efforts include conversion of identified gestures into both text and voice modalities ensuring multi-format accessibility by diverse demographic groups. Overall, this work presents an end-to-end ecosystem tackling the problem of sign language gesture recognition using bespoke computer vision and adaptive machine learning techniques for accessibility and inclusion. [ABSTRACT FROM AUTHOR]

This paper tackles the influence of the functionalization of polystyrene and graphene oxide (GO) composites on the flammability characteristics. A microscale combustion calorimeter (MCC) was used to experimentally determine the heat release capacity (HRC), the specific heat release rate (HRR) and the total heat released (THR). Neural models were designed that correlate the THR with a number of parameters related to the composition and type of flame retardant used, the heating rate, the amount of residue, the HRC, the peak heat release rate (PHRR), the temperature at the peak pyrolysis rate (TPHRR) and the time elapsed until the occurrence of the peak heat release rate (Time). The best results in the training, validation and testing stages were achieved with the neural model with 9 neurons in the input layer, 40 neurons in the hidden layer and one neuron in the output layer. This model was incorporated into an optimization procedure, based on a genetic algorithm, to establish the values of the input parameters used in the training of the neural networks, in order to generate a minimum THR value, which is the output parameter. Since the synthesis of polystyrene particles with different GO concentrations is costly, this research helps to reduce the number of experimental tests and allows to determine the best GO concentration by means of neural models and genetic algorithms. [ABSTRACT FROM AUTHOR]

In this study, the density for binary and ternary mixtures of benzyl alcohol, n-hexanol and water was determined experimentally in a composition range influenced by the miscibility of the components, four temperatures, 293.15, 303.15, 313.15 and 323.15 K, respectively and atmospheric pressure (0.1 MPa). The excess molar volume was calculated in correlation with composition, normalized temperature and refractive index, easily to determine experimentally using a small amount of substance. The 108 data sets were mixed, then randomly divided into 93 for training and 15 for validation stages. For the modelling purpose, artificial neural networks (ANN), with one or two layers of hidden neurons, were built. The best performance was obtained with the ANN neuronal model (4:8:4:1). The standard deviation calculated in the training stage is ± 0.0059 cm³ mol-1, and ± 0.070 cm³ mol-1 in the validation stage. The results obtained with neural models were also compared with those provided by the regression algorithms: k-Nearest Neighbor, Random Forest, Support Vector Machines and Linear Regression. The performance of the Random Forest model with 1000 trees is significantly better than that of the neural model (4:8:4:1). In the validation stage, the correlation coefficient for the Random Forest model was 0.9265, while for the neural model (4:8:4:1) it was 0.6595. [ABSTRACT FROM AUTHOR]

The study of neural cell behavior is one of the exciting fields that has attracted scientists' attention since 1890. Accordingly, many efforts have been made to model the neural cell specifying different neuron behaviors and characteristics. These studies mainly focus on two primary targets: finding new computing paradigms and providing better models and tools for medical applications. A critical model for studying biological neurons' behavior is the Hodgkin–Huxley neuron model (HH for short), presented in 1952. In this paper, the model is realized and verified using only one multiplier with some simplification. By comparing the original and simplified HH models, the RMSE and NRMSE are 1.83 and 0.016 for neuron membrane output potential, respectively, at the injected current of 20 uA. The dynamical behaviors of the model are also studied to ensure stability. The proposed model is synthesized and implemented on an FPGA platform, where the outputs are examined, and the results are presented as proof of concept. [ABSTRACT FROM AUTHOR]

As key interfaces for the disabled, optimal prosthetics should elicit natural sensations of skin touch or proprioception, by unambiguously delivering the multimodal signals acquired by the prosthetics to the nervous system, which still remains challenging. Here, a bioinspired temperature‐pressure electronic skin with decoupling capability (TPD e‐skin), inspired by the high‐low modulus hierarchical structure of human skin, is developed to restore such functionality. Due to the bionic dual‐state amplifying microstructure and contact resistance modulation, the MXene TPD e‐skin exhibits high sensitivity over a wide pressure range and excellent temperature insensitivity (91.2% reduction). Additionally, the high‐low modulus structural configuration enables the pressure insensitivity of the thermistor. Furthermore, a neural model is proposed to neutrally code the temperature‐pressure signals into three types of nerve‐acceptable frequency signals, corresponding to thermoreceptors, slow‐adapting receptors, and fast‐adapting receptors. Four operational states in the time domain are also distinguished after the neural coding in the frequency domain. Besides, a brain‐like machine learning‐based fusion process for frequency signals is also constructed to analyze the frequency pattern and achieve object recognition with a high accuracy of 98.7%. The TPD neural system offers promising potential to enable advanced prosthetic devices with the capability of multimodality‐decoupling sensing and deep neural integration. [ABSTRACT FROM AUTHOR]

In order to achieve a more efficient allocation of healthcare resources in the near future, it is crucial to understand the patterns and causes of excess mortality and hospitalizations. Neonatal mortality still poses a significant challenge, particularly in developed nations where the mortality rates are already low and healthcare resources are generally available to most of the population. Furthermore, the low mortality rates mean that the data available for modeling are often very limited, restricting the modeling methods that can be used. It is also important that the chosen methods allow for explainable, non-black-box models that can be interpreted by healthcare professionals. Considering these challenges, the work hereby presented thoroughly analyzed the time series of the neonatal mortality rates in Portugal between 2014 and 2019 in terms of trend and seasonal patterns. The applicability and performance of different data-based methods were also analyzed. Furthermore, the mortality rates were studied in terms of their relation to environmental variables, such as temperature and air pollution indicators, with the goal of establishing causal relations between such variables and excess mortality. The preliminary results show that ARMA, neural and fuzzzy models are able to forecast the studied mortality rates with good performance. In particular, neural models have the best predictive performance, while fuzzy models are well suited to obtain interpretable models with acceptable performance. [ABSTRACT FROM AUTHOR]

The abundance of digital documents offers a valuable chance to gain insights into public opinion, social structure, and dynamics. However, the scale and volume of these digital collections makes manual analysis approaches extremely costly and not scalable. In this paper, we study the potential of using automated methods from natural language processing and machine learning, in particular weak supervision strategies, to understand how news influence decision making in society. Besides proposing a weak supervision solution for the task, which replaces manual labeling to a certain extent, we propose an improvement of a recently published economic index. This index is known as economic policy uncertainty (EPU) index and has been shown to correlate to indicators such as firm investment, employment, and excess market returns. In summary, in this paper, we present an automated data efficient approach based on weak supervision and deep learning (BERT + WS) for identification of news articles about economical uncertainty and adapt the calculation of EPU to the proposed strategy. Experimental results reveal that our approach (BERT + WS) improves over the baseline method centered in keyword search, which is currently used to construct the EPU index. The improvement is over 20 points in precision, reducing the false positive rate typical to the use of keywords. [ABSTRACT FROM AUTHOR]

Topic models extract commonly occurring latent topics from textual data. Statistical models such as Latent Dirichlet Allocation do not produce dense topic embeddings readily integratable into neural architectures, whereas earlier neural topic models are yet to fully take advantage of the discrete nature of the topic space. To bridge this gap, we propose a novel neural topic model, Discrete-Variational-Inference-based Topic Model (DVITM), which learns dense topic embeddings homomorphic to word embeddings via discrete variational inference. The model also views words as mixtures of topics and digests embedded input text. Quantitative and qualitative evaluations empirically demonstrate the superior performance of DVITM compared to important baseline models. In the end, case studies on text generation from a discrete space and aspect-aware item recommendation are presented to further illustrate the power of our model in downstream tasks. [ABSTRACT FROM AUTHOR]

The paper is devoted to the review of the coherence resonance phenomenon in excitable neural networks. In particular, we explain how coherence can be measured and how noise affects neural activity. According to our research, intrinsic brain noise, which affects neural activity at the microscopic level, has a positive effect at the macroscopic level related to brain connectivity. Namely, it coordinates responses of different brain areas and forces their interaction to efficiently process sensory information. We find that brain noise can be altered as a result of attention and cognitive training to optimize the efficiency of information processing. Numerous experimental and theoretical studies provide substantial evidence for beneficial effects of internal brain noise on cognitive performance. Furthermore, coherence resonance in the brain response to a cognitive task not only increases neural activity in certain brain areas, but also provides pathways for neural communication between distant brain areas. In addition, the study of coherent resonance allows finding optimal parameters for better performance and efficient control of brain–computer interfaces. [ABSTRACT FROM AUTHOR]

Although having robust construction, the three-phase induction motor is frequently exposed to electrical, mechanical, and thermal stress, which, over time, may result in failure. This study addresses the application of signal processing tools and artificial neural networks to perform a multiple-fault diagnosis, considering different coupled load levels. The strategy consists of converting abc-referenced stator currents to dq rotating coordinate system by Clarke and Park transforms. After that, wavelet transform is employed to decompose the projected signals and the standard deviation values of the detail coefficients are calculated. Subsequently, we compare the performances of self-organizing maps, multilayer perceptrons, and radial basis function classifiers using four result metrics: accuracy, precision, sensitivity, and specificity rates. The conclusion is that all classifiers performed well for low levels of coupled load. However, for levels close to the nominal motor value, only the perceptron was able to discriminate the tested conditions. [ABSTRACT FROM AUTHOR]

While affordance detection and Human-Object interaction (HOI) detection tasks are related, the theoretical foundation of affordances makes it clear that the two are distinct. In particular, researchers in affordances make distinctions between J. J. Gibson's traditional definition of an affordance, “the action possibilities” of the object within the environment, and the definition of a telic affordance, or one defined by conventionalized purpose or use. We augment the HICO-DET dataset with annotations for Gibsonian and telic affordances and a subset of the dataset with annotations for the orientation of the humans and objects involved. We then train an adapted Human-Object Interaction (HOI) model and evaluate a pre-trained viewpoint estimation system on this augmented dataset. Our model, AffordanceUPT, is based on a two-stage adaptation of the Unary-Pairwise Transformer (UPT), which we modularize to make affordance detection independent of object detection. Our approach exhibits generalization to new objects and actions, can effectively make the Gibsonian/telic distinction, and shows that this distinction is correlated with features in the data that are not captured by the HOI annotations of the HICO-DET dataset.

Brain-inspired Spiking Neural Network (SNN) is opening new possibilities towards human-level intelligence, by leveraging its nature of spatiotemporal information encoding and processing that bring both learning effectiveness and energy efficiency. Although substantial advances in SNN studies have been made, highly effective SNN learning algorithms are still urged, driven by the challenges of coordinating spiking spatiotemporal dynamics. We therefore propose a novel algorithm, SpikeBASE, denoting Spiking learning with Backward Adaption of Synaptic Efflux, to globally, supervisedly, and comprehensively coordinate the synaptic dynamics including both synaptic strength and responses. SpikeBASE can learn synaptic strength by backpropagating the error through the predefined synaptic responses. More importantly, SpikeBASE enables synaptic response adaptation through backpropagation, to mimic the complex dynamics of neural transmissions. Further, SpikeBASE enables multi-scale temporal memory formation by supporting multi-synaptic response adaptation. We have evaluated the algorithm on a challenging scarce data learning task and shown highly promising performance. The proposed SpikeBASE algorithm, through comprehensively coordinating the learning of synaptic strength, synaptic responses, and multi-scale temporal memory formation, has demonstrated its effectiveness on end-to-end SNN training. This study is expected to greatly advance the learning effectiveness of SNN and thus broadly benefit smart and efficient big data applications. [ABSTRACT FROM AUTHOR]

Wind turbine power curve (WTPC) modeling from measured data is essential to predict the power generation from wind farms. Polynomial regression is commonly the first choice for this purpose, but there are other more powerful alternatives based on neural networks and fuzzy algorithms, for instance. Despite the existence of previous applications of such learning algorithms to WTPC modeling, a critical analysis of their performances has not yet been carried out while taking into into account both quantitative and quantitative aspects. Quantitative figures of merit include the root-mean-square error (RMSE) and R-squared ( R 2 ), whereas qualitative approaches are often based on simple visual inspection. In this context, this work reports the results of a comprehensive performance comparison involving the estimation of WTPC. The study comprises three neural-network-based models, that is, multilayer perceptron (MLP), radial basis function (RBF), and extreme learning machine (ELM); as well two fuzzy-logic-based models, that is, Takagi-Sugeno-Kang (TSK) and adaptive network fuzzy inference system (ANFIS). Using two real-world challenging data sets, it is possible to evaluate how the models perform concerning the accuracy of the curve fitting, sensitivity to parameter initialization, and occurrence of pathological solutions. Relevant issues, such as hyperparameter settings and data normalization are also addressed. The obtained results confirm the fact that the model selection should not rely only on quantitative performance indices. Thus, it is reasonable to state that the design of general-purpose modeling tools such as the ones evaluated in this work should incorporate domain-specific knowledge to provide good accuracy associated with reliable results. [ABSTRACT FROM AUTHOR]

In order to achieve a more efficient allocation of healthcare resources in the near future, it is crucial to understand the patterns and causes of excess mortality and hospitalizations. Neonatal mortality still poses a significant challenge, particularly in developed nations where the mortality rates are already low and healthcare resources are generally available to most of the population. Furthermore, the low mortality rates mean that the data available for modeling are often very limited, restricting the modeling methods that can be used. It is also important that the chosen methods allow for explainable, non-black-box models that can be interpreted by healthcare professionals. Considering these challenges, the work hereby presented thoroughly analyzed the time series of the neonatal mortality rates in Portugal between 2014 and 2019 in terms of trend and seasonal patterns. The applicability and performance of different data-based methods were also analyzed. Furthermore, the mortality rates were studied in terms of their relation to environmental variables, such as temperature and air pollution indicators, with the goal of establishing causal relations between such variables and excess mortality. The preliminary results show that ARMA, neural and fuzzzy models are able to forecast the studied mortality rates with good performance. In particular, neural models have the best predictive performance, while fuzzy models are well suited to obtain interpretable models with acceptable performance.

This article describes neural models of attention. Since attention is not a disembodied process, the article explains how brain processes of consciousness, learning, expectation, attention, resonance, and synchrony interact. These processes show how attention plays a critical role in dynamically stabilizing perceptual and cognitive learning throughout our lives. Classical concepts of object and spatial attention are replaced by mechanistically precise processes of prototype, boundary, and surface attention. Adaptive resonances trigger learning of bottom-up recognition categories and top-down expectations that help to classify our experiences, and focus prototype attention upon the patterns of critical features that predict behavioral success. These feature-category resonances also maintain the stability of these learned memories. Different types of resonances induce functionally distinct conscious experiences during seeing, hearing, feeling, and knowing that are described and explained, along with their different attentional and anatomical correlates within different parts of the cerebral cortex. All parts of the cerebral cortex are organized into layered circuits. Laminar computing models show how attention is embodied within a canonical laminar neocortical circuit design that integrates bottom-up filtering, horizontal grouping, and top-down attentive matching. Spatial and motor processes obey matching and learning laws that are computationally complementary to those obeyed by perceptual and cognitive processes. Their laws adapt to bodily changes throughout life, and do not support attention or conscious states.

Understanding how neurons perform, when they are organized in interacting networks, is a key to understanding how the brain performs complex functions. Different models that approximate the behavior of interconnected neurons have been proposed in the literature. Implementing these models to simulate neuron behavior at an appropriately detailed level to observe collective phenomena is computationally intensive. In this study we analyze the coupled Leaky Integrate-and-Fire model and report on the issues that affect performance when the model is implemented on a GPU. We conclude that the problem is heavily memory-bound. Advances in memory technology at the hardware level seem to be the deciding factor to achieve better performance on the GPU. Our results show that using an NVidia K40 GPU a modest 2x speedup can be achieved compared to a parallel implementation running on a modern multi-core CPU. However, a substantial speedup of 11.1x can be achieved using an NVidia V100 GPU, mainly due to the improvements in its memory subsystem. • Simulations of Leaky Integrate-and-Fire networks are shown to be memory-bound. • Optimizations of memory accesses are important to improve performance on the GPU. • 3D stacked memory on newer GPUs improves performance due to higher memory bandwidth. • Synchronization overhead is low. Optimizations of CUDA Graphs improve performance. • Our simulator is faster than others, despite using double precision arithmetic. [ABSTRACT FROM AUTHOR]

The ability to estimate and produce appropriately timed responses is central to many behaviors including speaking, dancing, and playing a musical instrument. A classical framework for estimating or producing a time interval is the pacemaker-accumulator model in which pulses of a pacemaker are counted and compared to a stored representation. However, the neural mechanisms for how these pulses are counted remain an open question. The presence of noise and stochasticity further complicates the picture. We present a biophysical model of how to keep count of a pacemaker in the presence of various forms of stochasticity using a system of bistable Wilson-Cowan units asymmetrically connected in a one-dimensional array; all units receive the same input pulses from a central clock but only one unit is active at any point in time. With each pulse from the clock, the position of the activated unit changes thereby encoding the total number of pulses emitted by the clock. This neural architecture maps the counting problem into the spatial domain, which in turn translates count to a time estimate. We further extend the model to a hierarchical structure to be able to robustly achieve higher counts. [ABSTRACT FROM AUTHOR]

Bi-stable perception is a strong instance of cognitive self-organization, providing a research model for how ‘the brain makes up its mind.’ The complexity of perceptual bistability prevents a simple attribution of functions to areas, because many cognitive processes, recruiting multiple brain regions, are simultaneously involved. The functional magnetic resonance imaging (fMRI) evidence suggests the activation of a large network of distant brain areas. Concurrently, electroencephalographic and magnetoencephalographic (MEEG) literature shows sub second oscillatory activity and phase synchrony on several frequency bands. Strongly represented are beta and gamma bands, often associated with neural/cognitive integration processes. The spatial extension and short duration of brain activities suggests the need for a fast, large-scale neural coordination mechanism. To address the range of temporo-spatial scales involved, we systematize the current knowledge from mathematical models, cognitive sciences and neuroscience at large, from single-cell- to system-level research, including evidence from human and non-human primates. Surprisingly, despite evidence spanning through different organization levels, models, and experimental approaches, the scarcity of integrative studies is evident. In a final section of the review we dwell on the reasons behind such scarcity and on the need of integration in order to achieve a real understanding of the complexities underlying bi-stable perception processes.

Bi-stable perception is a strong instance of cognitive self-organization, providing a research model for how 'the brain makes up its mind.' The complexity of perceptual bistability prevents a simple attribution of functions to areas, because many cognitive processes, recruiting multiple brain regions, are simultaneously involved. The functional magnetic resonance imaging (fMRI) evidence suggests the activation of a large network of distant brain areas. Concurrently, electroencephalographic and magnetoencephalographic (MEEG) literature shows sub second oscillatory activity and phase synchrony on several frequency bands. Strongly represented are beta and gamma bands, often associated with neural/cognitive integration processes. The spatial extension and short duration of brain activities suggests the need for a fast, large-scale neural coordination mechanism. To address the range of temporo-spatial scales involved, we systematize the current knowledge from mathematical models, cognitive sciences and neuroscience at large, from single-cell- to system-level research, including evidence from human and non-human primates. Surprisingly, despite evidence spanning through different organization levels, models, and experimental approaches, the scarcity of integrative studies is evident. In a final section of the review we dwell on the reasons behind such scarcity and on the need of integration in order to achieve a real understanding of the complexities underlying bi-stable perception processes. [ABSTRACT FROM AUTHOR]

Artificial Neural Networks (ANN) were used for estimating the transmembrane flux in a crossflow microfiltration process with ceramic tubular membranes to clarify the vinasse. The prevision was accomplished through the training of ANN feedforward using the experimental database generated in the work of Trevisoli (2010). The results showed a good correlation between the estimated data and the experimental data of transmembrane flux. For the microfiltration process with the membrane nominal pore size of 0.8 μm, the test subset presented maximum percentage error of 5.21% and average percentage error of 1.62%. For the membrane nominal pore size of 1.2 μm, the test subset had maximum percentage error of 28.51% and average percentage error of 4.66%. Therefore, it is feasible to use the ANN technique to estimate future data, helping to study membranes in microfiltration processes.

The description of neural dynamics, in terms of precise characterizations of action potential timings and shape and voltage related measures, is fundamental for a deeper understanding of the neural code and its information content. Not only such measures serve the scientific questions posed by experimentalists but are increasingly being used by computational neuroscientists for the construction of biophysically detailed data-driven models. Nonetheless, online resources enabling users to perform such feature extraction operation are lacking. To address this problem, in the framework of the Human Brain Project and the EBRAINS research infrastructure, we have developed and made available to the scientific community the NeuroFeatureExtract, an open-access online resource for the extraction of electrophysiological features from neural activity data. This tool allows to select electrophysiological traces of interest, fetched from public repositories or from users’ own data, and provides ad hoc functionalities to extract relevant features. The output files are properly formatted for further analysis, including data-driven neural model optimization.