Enhancing Business Operations with TensorFlow and Neural Network-based Data Analysis

Data-driven multi-objective optimization with neural network-based sensitivity analysis for semiconductor devices

This article aimed to determine the effectiveness of (Machine Learning, Artificial Neural Networks, Pattern Recognition and Data Analysis, and Fire Dynamics Simulations) in improving the results of arson investigations in the United Arab Emirates (UAE). The research relied on the descriptive method. The study community consisted of all members of the teams specialized in fire investigations. The research sample was taken using a simple random method that included (74) individuals from the research community. The researcher relied on the questionnaire as the main instrument for collecting data from the study sample. The results of the study showed that artificial intelligence techniques (Machine Learning, Artificial Neural Networks, Pattern Recognition and Data Analysis, and Fire Dynamics Simulations) can be effective in improving the results of arson investigations. The study recommended providing training programs to investigators on how to make full use of AI techniques in the investigation process.

A common way to advance our understanding of brain processing is to decode behavior from recorded neural signals. In order to study the neural correlates of learning a task, we would like to decode behavior across the entire timespan of learning, which can take multiple recording sessions across many days. However, decoding across sessions is hindered due to a high amount of session-to-session variability in neural recordings. Here, we propose utilizing multidimensional neural signals from Localized semi-non negative matrix factorization processing (LocaNMF) with high behavioral correlations across sessions, as well as a novel data augmentation method and region-based converter, to optimally align neural recordings. We apply our method to widefield calcium activity across many sessions while a mouse learns a decision-making task. We first decompose each session's neural activity into region-based spatial and temporal components that can reconstruct the data with high variance. Next, we perform data augmentation of the neural data to smooth the variability across trials. Finally, we design a region-based neural converter across sessions that transforms one session's neural signals into another while preserving its dimensionality. We test our approach by decoding the mouse's behavior in the decision-making task, and find that our method outperforms approaches that use purely anatomical information while analyzing neural activity across sessions. By preserving the high dimensionality in the neural data while converting neural activity across sessions, our method can be used towards further analyses of neural data across sessions and the neural correlates of learning.

An integrative analysis platform for multiple neural spike train data

Objective. Neural decoding is an important tool in neural engineering and neural data analysis. Of various machine learning algorithms adopted for neural decoding, the recently introduced deep learning is promising to excel. Therefore, we sought to apply deep learning to decode movement trajectories from the activity of motor cortical neurons. Approach. In this paper, we assessed the performance of deep learning methods in three different decoding schemes, concurrent, time-delay, and spatiotemporal. In the concurrent decoding scheme where the input to the network is the neural activity coincidental to the movement, deep learning networks including artificial neural network (ANN) and long-short term memory (LSTM) were applied to decode movement and compared with traditional machine learning algorithms. Both ANN and LSTM were further evaluated in the time-delay decoding scheme in which temporal delays are allowed between neural signals and movements. Lastly, in the spatiotemporal decoding scheme, we trained convolutional neural network (CNN) to extract movement information from images representing the spatial arrangement of neurons, their activity, and connectomes (i.e. the relative strengths of connectivity between neurons) and combined CNN and ANN to develop a hybrid spatiotemporal network. To reveal the input features of the CNN in the hybrid network that deep learning discovered for movement decoding, we performed a sensitivity analysis and identified specific regions in the spatial domain. Main results. Deep learning networks (ANN and LSTM) outperformed traditional machine learning algorithms in the concurrent decoding scheme. The results of ANN and LSTM in the time-delay decoding scheme showed that including neural data from time points preceding movement enabled decoders to perform more robustly when the temporal relationship between the neural activity and movement dynamically changes over time. In the spatiotemporal decoding scheme, the hybrid spatiotemporal network containing the concurrent ANN decoder outperformed single-network concurrent decoders. Significance. Taken together, our study demonstrates that deep learning could become a robust and effective method for the neural decoding of behavior.

CHAPTER 18 - Power and Cross Spectra

Originally invented to describe magnetism, the Ising model has proven to be useful in many other applications, as, for example, inference problems in computer science, socioeconomic physics, the analysis of neural data [1,2,3] and modeling of neural networks (binary neurons). Despite its simplicity, there exists no general solution to the Ising model, i.e. the partition function is unknown in the case of an interacting system. Mean field theory is often used as an approximation being exact in the noninteracting case and for infinite dimensions. A correction term to the mean field approximation of Gibb's free energy (the effective action) of the Ising model was given by Thouless, Anderson and Palmer (TAP) [4] as a “fait accompli” and was later derived by different methods in [5,6,7], where also higher order terms were computed.We present a diagrammatic derivation (Feynman diagrams) of these correction terms and embed the problem in the language of field theory. Furthermore, we show how the iterative construction of the effective action used in the Ising case generalizes to arbitrary non-Gaussian theories.

BSMART: a Matlab/C toolbox for analysis of multichannel neural time series.

Full text Figures and data Side by side Abstract eLife digest Introduction Results Discussion Materials and methods References Decision letter Author response Article and author information Metrics Abstract Neurons in higher cortical areas, such as the prefrontal cortex, are often tuned to a variety of sensory and motor variables, and are therefore said to display mixed selectivity. This complexity of single neuron responses can obscure what information these areas represent and how it is represented. Here we demonstrate the advantages of a new dimensionality reduction technique, demixed principal component analysis (dPCA), that decomposes population activity into a few components. In addition to systematically capturing the majority of the variance of the data, dPCA also exposes the dependence of the neural representation on task parameters such as stimuli, decisions, or rewards. To illustrate our method we reanalyze population data from four datasets comprising different species, different cortical areas and different experimental tasks. In each case, dPCA provides a concise way of visualizing the data that summarizes the task-dependent features of the population response in a single figure. https://doi.org/10.7554/eLife.10989.001 eLife digest Many neuroscience experiments today involve using electrodes to record from the brain of an animal, such as a mouse or a monkey, while the animal performs a task. The goal of such experiments is to understand how a particular brain region works. However, modern experimental techniques allow the activity of hundreds of neurons to be recorded simultaneously. Analysing such large amounts of data then becomes a challenge in itself. This is particularly true for brain regions such as the prefrontal cortex that are involved in the cognitive processes that allow an animal to acquire knowledge. Individual neurons in the prefrontal cortex encode many different types of information relevant to a given task. Imagine, for example, that an animal has to select one of two objects to obtain a reward. The same group of prefrontal cortex neurons will encode the object presented to the animal, the animal's decision and its confidence in that decision. This simultaneous representation of different elements of a task is called a 'mixed' representation, and is difficult to analyse. Kobak, Brendel et al. have now developed a data analysis tool that can 'demix' neural activity. The tool breaks down the activity of a population of neurons into its individual components. Each of these relates to only a single aspect of the task and is thus easier to interpret. Information about stimuli, for example, is distinguished from information about the animal's confidence levels. Kobak, Brendel et al. used the demixing tool to reanalyse existing datasets recorded from several different animals, tasks and brain regions. In each case, the tool provided a complete, concise and transparent summary of the data. The next steps will be to apply the analysis tool to new datasets to see how well it performs in practice. At a technical level, the tool could also be extended in a number of different directions to enable it to deal with more complicated experimental designs in future. https://doi.org/10.7554/eLife.10989.002 Introduction In many state of the art experiments, a subject, such as a rat or a monkey, performs a behavioral task while the activity of tens to hundreds of neurons in the animal's brain is monitored using electrophysiological or imaging techniques. The common goal of these studies is to relate the external task parameters, such as stimuli, rewards, or the animal's actions, to the internal neural activity, and to then draw conclusions about brain function. This approach has typically relied on the analysis of single neuron recordings. However, as soon as hundreds of neurons are taken into account, the complexity of the recorded data poses a fundamental challenge in itself. This problem has been particularly severe in higher-order areas such as the prefrontal cortex, where neural responses display a baffling heterogeneity, even if animals are carrying out rather simple tasks (Brody et al., 2003; Machens et al., 2010; Hernández et al., 2010; Mante et al., 2013; Rigotti et al., 2013). Traditionally, this heterogeneity has often been neglected. In neurophysiological studies, it is common practice to pre-select cells based on particular criteria, such as responsiveness to the same stimulus, and to then average the firing rates of the pre-selected cells. This practice eliminates much of the richness of single-cell activities, similar to imaging techniques with low spatial resolution, such as MEG, EEG, or fMRI. While population averages can identify some of the information that higher-order areas process, they ignore much of the fine structure of the single cell responses (Wohrer et al., 2013). Indeed, most neurons in higher cortical areas will typically encode several task parameters simultaneously, and therefore display what has been termed mixed selectivity (Rigotti et al., 2013; Pagan and Rust, 2014; Park et al., 2014; Raposo et al., 2014). Instead of looking at single neurons and selecting from or averaging over a population of neurons, neural population recordings can be analyzed using dimensionality reduction methods (for a review, see Cunningham and Yu, 2014). In recent years, several such methods have been developed that are specifically targeted to electrophysiological data, working on the level of single spikes (Pfau et al., 2013), accommodating different time scales of latent variables (Yu et al., 2009), or accounting for the dynamics of the population response (Buesing et al., 2012a; 2012b; Churchland et al., 2012). However, these approaches reduce the dimensionality of the data without taking task parameters, i.e., sensory and motor variables controlled or monitored by the experimenter, into account. Consequently, mixed selectivity remains in the data even after the dimensionality reduction step. The problem can be addressed by dimensionality reduction methods that are informed by the task parameters (Machens et al., 2010; Machens, 2010; Brendel et al., 2011; Mante et al., 2013; Raposo et al., 2014). We have previously introduced a dimensionality reduction technique, demixed principal component analysis (dPCA) (Brendel et al., 2011), that emphasizes two goals. It aims to find a decomposition of the data into latent components that (a) are easily interpretable with respect to the experimentally controlled and monitored task parameters; and (b) preserve the original data as much as possible, ensuring that no valuable information is thrown away. Here we present a radically modified version of this method, and illustrate that it works well on a wide variety of experimental data. The new version of the method has the same objectives as the older version (Brendel et al., 2011), but is more principled, more flexible, and has an analytical solution, meaning that it does not suffer from any numerical optimization problems. Furthermore, the new mathematical formulation highlights similarities to and differences from related well-known methods such as principal component analysis (PCA) and linear discriminant analysis (LDA). The dPCA code is available at http://github.com/machenslab/dPCA for Matlab and Python. Results Existing approaches We illustrate the classical approaches to analyzing neural activity data from higher-order areas in Figure 1. To be specific, we consider recordings from the prefrontal cortex (PFC) of monkeys performing a somatosensory working memory task (Romo et al., 1999; Brody et al., 2003). In this task, monkeys were required to discriminate two vibratory stimuli presented to the fingertip. The stimuli F1 and F2 were separated by a 3 s delay, and the monkeys had to report which stimulus had a higher frequency by pressing one of the two available buttons (Figure 1a). Figure 1 Download asset Open asset Existing approaches to population analysis, illustrated with recordings from monkey PFC during a somatosensory working memory task (Romo et al., 1999). (a) Cartoon of the paradigm, adapted from Romo and Salinas (2003). Legend shows 12 experimental conditions. (b) Average per-condition firing rates (PSTHs) for four exemplary neurons out of N=832. Colors refer to stimulus frequencies F1 and line styles (dashed/solid) refer to decision, see legend in (a). (c) Fraction of cells, significantly (p<0.05, two-way ANOVA) tuned to stimulus and decision at each time point. (d) Left: Distribution of stimulus tuning effect sizes across neural population at F1 period (black arrow in c). Significantly tuned neurons are shown in dark gray. Right: Same for decision at F2 period (gray arrow in c). (e) The average of zero-centered PSTHs over all significantly tuned neurons (for neurons with negative effect size, the sign of PSTHs was flipped). Arrows mark time-points that were used to select the significant cells. (f) Fraction of cells, significantly (p<0.05, linear regression) tuned to stimulus and decision at each time point. (g) Distribution of regression coefficients of neural firing rates to stimulus (during F1 period) and decision (during F2 period). (h) Stimulus and decision components produced by the method of Mante et al. (2013). Briefly, neural PSTHs are weighted by the regression coefficients. (i) Fraction of variance captured by the first 20 principal components. (j) Distributions of weights used to produce the first six principal components (weights are elements of the eigenvectors of the N×N covariance matrix). (k) First six principal components (projections of the full data onto the eigenvector directions). https://doi.org/10.7554/eLife.10989.003 When we focus on the neural representation of the stimulus F1 and the decision, we have to take 12 experimental conditions into account (six possible values of F1 and two possible decisions). For each of these conditions, we can average each neuron's spike trains over trials and then smooth the resulting time series in order to estimate the neuron's time-dependent firing rate (also known as peri-stimulus time histogram or PSTH). We find that the PSTHs of many neurons are tuned to the stimulus F1, the decision, or both (Figure 1b; so-called mixed selectivity), and different neurons generally show different tuning. Our goal is to characterize and summarize the tuning of all N recorded neurons. The most standard and widespread approach is to resort to a statistical test (e.g. a two-way analysis of variance or ANOVA), in order to check whether the firing rate of a neuron depends significantly on the frequency F1 or on the monkey's decision. Such a test can be run for each neuron and each time point, in which case the population tuning over time is often summarized as the fraction of cells significantly tuned to stimulus or decision at each time point (p<0.05, Figure 1c). In addition to providing such a 'summary statistics', this approach is also used to directly visualize the population activity. For that purpose, one selects the subset of neurons significantly tuned to stimulus or decision (e.g. by focusing on a particular time point, Figure 1d) and then averages their PSTHs. The resulting 'population average' is shown in Figure 1e, where we also took the sign of the effect size into account. The population average is generally thought to demonstrate the 'most typical' firing pattern among the cells encoding the corresponding parameter. Importantly, this method yields one single population average or 'component' for each parameter. Each such component can be understood as a linear combination (or a linear readout) of the individual PSTHs, with all Ns significant neurons for a parameter having the same weights ± 1/Ns and all others having weight zero. In a related approach, the firing rates of each neuron at each time point are linearly regressed on stimulus and decision (Figure 1f) (Brody et al., 2003). Mante et al. (2013) suggested to use the regression coefficients of all N neurons (Figure 1g) as weights to form linear combinations of PSTHs representing stimulus and decision tuning (Figure 1h). This approach, which the authors call 'targeted dimensionality reduction' (TDR), also yields one component per task parameter: in our example, we obtain one component for the stimulus and one for the decision (Figure 1h; see Materials and methods for details). Both of these approaches are supervised, meaning that they are informed by the task parameters. At the same time, they do not seek to faithfully represent the whole dataset and are prone to losing some information about the neural activities. Indeed, the two components from Figure 1e explain only 23% of the total variance of the population firing rates and the two components from Figure 1h explain only 22% (see Materials and methods). Consequently, a naive observer would not be able to infer from the components what the original neural activities looked like. While such supervised approaches can be extended in various ways to produce more components and capture more variance, a more direct way to avoid this loss of information is to resort to unsupervised methods such as principal component analysis (PCA). This method extracts a set of principal components (PCs) that are linear combinations of the original PSTHs, just as the population averages above. However, the weights to form these linear combinations are chosen so as to maximize the amount of explained variance (first six components explain 69% of variance, see Figure 1i–k). The principal components can be thought of as 'building blocks' of neural activity: PSTHs of actual neurons are given by linear combinations of PCs, with the first PCs being more informative than the later ones. However, since PCA is an unsupervised method, information about stimuli and decisions is not taken into account, and the resulting components can retain mixed selectivity and therefore fail to highlight neural tuning to the task parameters. The most striking observation when comparing supervised and unsupervised approaches is how different the results look. Indeed, PCA paints a much more complex picture of the population activity, dominated by strong temporal dynamics, with several stimulus- and decision-related components. At the same time, none of the methods can fully demix the stimulus and decision information: even the supervised methods show decision-related activity in the stimulus components and stimulus-related activity in the decision components (Figure 1e,h). Demixed principal component analysis (dPCA) To address these problems, we developed a modified version of PCA that not only compresses the data, but also demixes the dependencies of the population activity on the task parameters. We will first explain that these two goals generally constitute a trade-off, then suggest a solution to this trade-off for a single task parameter, and then generalize to multiple task parameters. The trade-off between demixing and compression is illustrated in Figure 2, where we compare linear discriminant analysis (LDA, Figure 2a,b), PCA (Figure 2c,d), and dPCA (Figure 2e–h). We will first focus on a single task parameter and seek to reduce the activity of N=2 neurons responding to three different stimuli. For each stimulus, the joint activity of the two neurons traces out a trajectory in the space of firing rates as time progresses (Figure 2b). The aim of 'demixing' in this simplified case is to find a linear mapping (decoder) of the neural activity that separates the different stimuli (Figure 2a) and ignores the time-dependency. We can use LDA in order to determine a projection of the data that optimally separates the three stimuli. However, LDA will generally not preserve the 'geometry' of the original neural activity: firing patterns for stimuli 1 and 2 are close to each other and far away from stimulus 3, whereas in the LDA projection all three stimuli are equally spaced (Figure 2b). More generally, decoding is always prone to distorting the data and therefore tends to impede a proper reconstruction of the original data from the reduced description. Figure 2 Download asset Open asset Linear dimensionality reduction. (a) Linear discriminant analysis maps the firing rates of individual neurons onto a latent component that allows us to decode a task parameter of interest. Shades of grey inside each neuron show the proportion of variance due to the various task parameters (e.g. stimulus, decision, and time), illustrating mixed selectivity. In contrast, the LDA component is maximally demixed. (b) At any moment in time, the population firing rate of N neurons is represented by a point in the N-dimensional space; here N=2. Each trial is represented by a trajectory in this space. Colors indicate different stimuli and dot sizes represent time. The LDA component for stimulus is given by the projection onto the LDA axis (black line); projections of all points are shown along this line. All three stimuli are clearly separated, but their geometrical relation to each other is lost. (c) Principal component analysis linearly maps the firing rates into a few principal components such that a second linear transformation can reconstruct the original firing rates. (d) The same set of points as in (b) is projected onto the first PCA axis. However, the stimuli are no longer separated. Rather, the points along the PCA axis have complex dependencies on stimulus and time (mixed selectivity). The PCA axis minimizes the distances between the original points and their projections. (e) Demixed principal component analysis also compresses and decompresses the firing rates through two linear transformations. However, here the transformations are found by both minimizing the reconstruction error and enforcing a demixing constraint on the latent variables. (f) The same set of points as in (b) projected onto the first dPCA decoder axis. The three stimuli are clearly separated (as in LDA), but some information about the relative distances between classes is preserved as well (as in PCA). (g) The same data as in (b) linearly decomposed into the time effect, the stimulus effect, and the noise. (h) The dPCA projection from (f) has to be mapped onto a different axis, given by the dPCA encoder, in order to reconstruct the stimulus class means (large colored circles). The decoder and encoder axes together minimize the reconstruction error between the original data and the stimulus class means. https://doi.org/10.7554/eLife.10989.004 The aim of compression is to find a linear mapping (decoder) that reduces the dimensionality and preserves the original data as much as possible (Figure 2c,d). Using PCA, we determine a projection of the data that minimizes the reconstruction error between the projections and the original points. In contrast to LDA, PCA seeks to preserve the geometry of the neural activity, and thereby yields the most faithful reduction of the data (Figure 2d). However, the PCA projection does not properly separate the stimuli and mixes the time-dependency with the stimulus-dependency. The wildly different projection axes for LDA (Figure 2b) and PCA (Figure 2d) seem to suggest that the goals of demixing and compression are essentially incompatible in this example. However, we can achieve both goals by assuming that the reconstruction of the original data works along a separate encoder axis (Figure 2f,h). Given this additional flexibility, we first choose a decoder axis that reconciles the decoding and compression objectives. Once projected onto this axis, all three stimuli are separated from each other, as in LDA, yet their geometrical arrangement is approximately preserved, as in PCA (Figure 2f). In turn, when reconstructed along the encoder axis, the projected data still approximates the original data (Figure 2h). To define these ideas more formally, we assume that we simultaneously recorded the spike trains of N neurons. Let be our data with N in which the the firing rate or spike of the neuron for all task conditions and all trials to be i.e., with means PCA compresses the data with a decoder The resulting principal components can then be linearly through an encoder approximately the original data et al., The decoder is found by minimizing the error between the original data, and the reconstructed data, given by In the of Figure 2, the data is of size and the decoder is of size the information about task parameters does not the loss and PCA demixes these parameters. In our method, which we call demixed PCA (dPCA), we two to this classical we that the compression and steps reconstruct not the neural activity but the neural activity over trials and over some of the task parameters. In the example, the reconstruction is the of stimulus which has the same size as but in which data point is by the average neural activity for the corresponding stimulus, as shown in Figure we additional in this by the data with a linear mapping yet it with linear mapping (Figure The are chosen by minimizing the loss for each stimulus, the neural activities are projected close to the average stimulus, which allows us both to decode the stimulus and to preserve the relative distances of the neural activities. In order to see how this approach preserves all of the original data, and not just some we that the data in our both stimulus and time. The can be understood as of a linear decomposition of the full data into a that is by averaging over stimuli, and a that is by averaging over time. of the activity are captured in a (Figure In turn, we can find separate decoder and encoder axes for each of these Once more than N=2 neurons are these decoder and encoder axes constitute a dimensionality reduction that reduces the data into a few each of which properly one of the task parameters. In turn, the original neural activity can be reconstructed through linear combinations of these just as in The ideas of this can be extended to any number of task parameters. In this all datasets will have three time, stimulus, and decision, and we will the neural activities into on the and (see Figure in the Materials and Individual are given by a series of This decomposition is fully to the variance decomposition in The only is that the standard decomposition for three parameters and would have corresponding to the of and and the noise. Here we some of these as we are not in demixing (see Materials and methods). Once this decomposition is dPCA separate decoder and encoder for each by minimizing the loss Each the can be by using the solution of which can be in of (see Materials and methods). Each of each yields one demixed principal component similar to PCA, we order the components by the amount of explained that the axes corresponding to two different task parameters and are found from each other and being contrast to PCA where principal axes are all In a the loss that each set of axes the thereby proper and the data decomposition that the combination of all allows to reconstruct the original data, are a few other technical (see Materials and methods for details). We dPCA for simultaneously recorded neural activities. However, all datasets analyzed in this have been recorded across many and so to apply dPCA we have to use to any other decoding method, dPCA is prone to and so we a and to choose the parameter. The data and variance from are only if the dataset is i.e., if the same number of trials were recorded in each this is not the case, one can use a version of dPCA (Brendel et al., used the same variance decomposition but a different and loss function. The differences are out in the Materials and methods working memory task in monkey PFC We first dPCA to the dataset presented (Romo et al., 1999; Brody et al., neurons from two is for each neuron has a response pattern and many neurons show mixed selectivity are shown in Figure studies have to of these response patterns by analyzing different task such as the and (Romo et al., 1999; Brody et al., 2003; Machens et al., 2010; et al., the decision period et al., or both et al., we can summarize the features of the neural activity across the whole trial in a single (Figure Figure 3 Download asset Open asset Demixed PCA to recordings from monkey PFC during a somatosensory working memory task (Romo et al., 1999). (a) Cartoon of the paradigm, adapted from Romo and Salinas (2003). (b) Demixed principal components. first three second first three stimulus first three decision first In each the full data are projected onto the dPCA decoder axis, so that are 12 corresponding to 12 conditions (see show time during which the task parameters can be from activity with all recorded see Materials and that the across number of each component is shown in a explained are shown as (c) variance explained by PCA and dPCA Demixed PCA the same amount of variance as standard line shows an estimate of the fraction of in the data, the variance is due to in the (see Materials and methods). (d) of the individual demixed principal components. Each shows the proportion of total variance, and is out of four of different for variance, for stimulus variance, for decision variance, and for variance due to Each to be which shows how the total variance is among parameters. (e) shows dot between all of the first demixed principal mark the that are significantly and (see Materials and methods). shows between all of the first demixed principal components. of the are close to zero. as in PCA, we can of the demixed principal components (Figure as the 'building blocks' of the neural activity, in that the activity of each single neuron is a linear combination of these components. in four some are (Figure some only on stimulus F1 some only on decision and some on stimulus and decision together The components can be easily to demix the parameter which is what dPCA Indeed, the components shown in Figure are projections of the PSTHs of all neurons onto the most decoding each projection shows 12 corresponding to 12 conditions. components have all 12

Microelectrode array recordings from single neurons generate multidimensional data (spike trains) that contains vast amount of information on underlying neural dynamics. Typically, the data analysis procedure is very time consuming, which greatly hinders the experimental throughputs. Bioinformatics community also deals with high dimensional data sets and the underlying mathematics of data analysis used in this field is very similar to that used in neural informatics. Here, we attempt to use the well-established data analysis procedure (Bayesian network inference) in Bioinformatics and utilized it to estimate the functional connectivity of cultured neural networks based on multichannel spike trains. The basic analysis procedure could be easily extended to in vivo neural spike data analysis for various neural engineering applications

The growing use of artificial intelligence (AI) in physical education (PE) has led to an urgent need to develop robust methodologies that can be used to choose the most suitable algorithms in uncertain and vague environments. This paper introduces a new hybrid decision-making (DM) model that incorporates the weighted aggregated sum product assessment (WASPAS) technique into the q-rung linear Diophantine fuzzy set (q-RLDFS) framework. The primary objective is to address the gap in the lack of structured and uncertainty-resistant methods for assessing AI models based on multiple, frequently conflicting criteria in the domain of PE. The proposed model presents a two-layer framework, where the WASPAS method enables a flexible scoring system by combining both the weighted sum model (WSM) and the weighted product model (WPM). On the other hand, the q-RLDFS framework facilitates high-order fuzzy modelling and accounts for hesitation in expert ratings, thereby making the decision results more interpretable and robust. To determine the effectiveness of the model, five AI algorithms named convolutional neural network-based motion analysis (CNN-MA), reinforcement learning based training optimizer (RL-TO), expert system for exercise prescription (ES-EP), hybrid AI tutor with natural language processing (HAI-NLP), and wearable sensor data mining algorithm (WSDMA) are evaluated against eight key criteria relevant to PE. Results revealed that CNN-MA is the most effective solution to implement, followed by RL-TO. The sensitivity and comparative analysis are thoroughly conducted to determine the validity of the model in terms of its robustness and reliability. The research offers distinct practical implications and actionable recommendations to educators, administrators, and policymakers, informing the strategic implementation of AI technologies within the PE domain. In general, the research makes a significant contribution to the adoption of AI in PE, as it provides a scalable, transparent, and practically applicable decision-support model.

Log analysis plays a crucial role in understanding system behavior and detecting anomalies in various domains. This article presents the results of our autoencoder type neural network trained for log analysis and anomaly detection based on the reconstruction error of each entry. Furthermore, the article focuses on the challenge of text vectorization, specifically employing a bag-of-words method for representing log messages in a numerical format suitable for neural network models. The bag-of-words approach captures the frequency-based representation of log messages. The study also acknowledges the limitations of this method in capturing semantic information. It is important to note that this study represents only a starting point in the journey of log analysis and text vectorization. The field encompasses a wide range of existing working methods, and further research and exploration are necessary to uncover the full potential of log analysis techniques for various applications.

Air pollution is one of the most common concerns for human health. Air contamination has many pollutants like NO2, PM2.5, PM10, and other harmful gases, which are harmful to the living system on the earth. The quality of air gets affected by these particulate matters. This paper proposes a neural network-based prediction model for detection and prediction of air pollution. Performance measurement of Bayesian Regularization (BR) algorithm, Levenberg–Marquardt (LM) algorithm, and Scaled Conjugate Gradient (SCG) algorithm has been analyzed.

The factor evaluation for sustainability implementation in post-disaster reconstruction projects was conducted using artificial neural network (ANN)-based sensitivity analysis. The current study involves the utilization of Levenberg-Marquardt algorithm (LMA) for the ANN model development. The findings demonstrated that the governing network structure is 10-3-1 utilizing input parameters (IP) including waste management (Sl), environmental impact management (S2), support of management (S3), construction cost (S4), public health and safety (S5), user security (S6), noise pollution (S7), energy consumption (S8), public services (S9), and recycling (S10). The results revealed that the governing model (GM) has a performance of correlation (R) = 0.99813 and Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE) of 0.00110 and 0.4240/0, respectively. Moreover, a sensitivity analysis (SA) using Garson's algorithm (GA) reveals a trend of the relative value (RV) was observed to be construction cost > public services > support of management > user security > environmental impact management> noise pollution> public health and safety> energy consumption > recycling > waste management wherein the construction cost is the most influential parameter to the sustainability rating in post-disaster reconstruction projects (PDRP). The study shows how well ANN can detect the crucial elements that determine a project's sustainability rating in the post-disaster reconstruction projects.

Global or continental-scale land cover mapping with remote sensing data is limited by the spatial characteristics of satellites. Subpixel-level mapping is essential for the successful description of many land cover patterns with spatial resolution of less than /spl sim/1 km and also useful for finer resolution data. This paper presents a novel adaptive resonance theory MAP (ARTMAP) neural network-based mixture analysis model-ART mixture MAP (ART-MMAP). Compared to the ARTMAP model, ART-MMAP has an enhanced interpolation function that decreases the effect of category proliferation in ART/sub a/ and overcomes the limitation of class category in ART/sub b/. Results from experiments demonstrate the superiority of ART-MMAP in terms of estimating the fraction of land cover within a single pixel.