Brain Informatics

This paper presents a first study to apply a computational approach to Genito-Pelvic Pain/Penetration Disorder (GPPPD) using a Fear Avoidance Model. An adaptive temporal-causal network model for fear avoidance was designed and therapeutic interventions were incorporated targeting one or two emotional states. Validation with empirical data shows that for one type of individual therapeutic intervention targeting two states can reduce pain and other complaints. For three other types of individuals, targeting two emotional states was not sufficient to reduce pain and other complaints. The computational model can address large individual differences and supports the claim that interventions for GPPPD should be multidisciplinary.

Parental influence plays an important role in the mental development of a child. In the early years of childhood, a parent acts as a role model to a child, so most of the children try to mimic their parents. In our work, we address a complex network model of a child who is influenced by a narcissistic parent from his/her childhood to his/her adolescence. This concept of mimicking in childhood is represented by social contagion. Later on, he/she can learn to develop his/her own personality based on experience and learning. This model can be used to predict the influence of a parent over the personality of a child.

Multivariate prediction of human behavior from resting state data is gaining increasing popularity in the neuroimaging community, with far-reaching translational implications in neurology and psychiatry. However, the high dimensionality of neuroimaging data increases the risk of overfitting, calling for the use of dimensionality reduction methods to build robust predictive models. In this work, we assess the ability of four dimensionality reduction techniques to extract relevant features from resting state functional connectivity matrices of stroke patients, which are then used to build a predictive model of the associated language deficits based on cross-validated regularized regression. Features extracted by Principal Component Analysis (PCA) were found to be the best predictors, followed by Independent Component Analysis (ICA), Dictionary Learning (DL) and Non-Negative Matrix Factorization. However, ICA and DL led to more parsimonious models. Overall, our findings suggest that the choice of the dimensionality reduction technique should not only be based on prediction/regression accuracy, but also on considerations about model complexity and interpretability.

Previous Experiments in the field of human behavior and game theory has shown that loss aversion has a major effect on players’ decisions in coordination problems. The overarching aim of our study was to model the effect of loss aversion on individual player behavior in divergent interest tacit coordination games. Based on a large-scale behavioral data we have designed a model predicting the total number of points players allocate to themselves as a result of increased penalty values in cases of non-coordination. Understanding the effect of loss aversion in case of divergent interest coordination problems on players’ behavior will allow us to better predict the human decision-making process and as a result, create more realistic algorithms for human-machine cooperation’s. Understanding the effect of loss aversion in the context of divergent interest tacit coordination games may enable the construction of better algorithms for human-machine interaction that could more accurately predict human decision behavior under uncertainty.

This research provides evidence highlighting that through the use of a gamma 40 Hz entrainment frequency, mood, memory and cognition can be improved with respect to a 10 participant cohort. Participants constituted towards three binaural entrainment frequency groups; the 40 Hz, 25 Hz and 100 Hz respectively. Additionally, we asked participants to attend entrainment frequency sessions twice a week for a duration of four weeks. Sessions involved the assessment of a participants cognitive abilities, mood as well as memory; where the cognitive and memory assessments occurred before and after a 5 min binaural beat stimulation. The mood assessment scores were collected from sessions 1, 4 and 8 respectively. Within the gamma 40 Hz entrainment frequency group, we observed a weak statistical significance (alpha = 0.10, p = 0.076) mean improvement in cognitive scores; elevating from 75% average to 85% average upon conclusion of the experimentation. Additionally, we observed memory score improvements at a greater significance (alpha = 0.05, p = 0.0027); elevating from an average of 87% to 95%. Moreover, we observed a similar trend across the average of all of the frequency groups for the mood results. Finally, correlation analysis revealed a stronger correlation value (0.9838) within the 40 Hz group between sessions as well as mood score compared across the entire frequency group cohort.

Human speech function, as an incredible manifestation of human intelligence, entails intricate spatiotemporal coordination of brain networks transiently and accurately. Current investigation using neuroimaging and electrophysiological techniques laid the foundation of our understanding regarding the brain activities in the spatial, temporal, and spectral domains. However, a comprehensive view integrating these three aspects yet to be achieved by not only adopting multi- modalities of the data acquisition system but also employing algorithms to integrate them into a systematic framework. Thus, this study conducted a passive listening task using words and white noises as acoustic stimuli and utilized high-density electroencephalography (EEG) system with effective connectivity analysis to reconstruct the brain network dynamics with high temporal and spectral resolution. Besides, we introduced the high-spatial-resolution functional magnetic resonance imaging- (fMRI-) constraints into a representational similarity analysis to examine the functional performance of spatially distributed networks over time. Our results revealed that during speech perception, networks for auditory and higher cognition functioned along the ventral stream via theta and gamma oscillations and exhibited hierarchical responsive differences between word and noise conditions. Speech motor programming networks participated along the dorsal stream mainly in the beta band during a later period of speech perception. Alpha band activity served as a mediation for the dual pathway through oscillatory suppression. These functional networks progressed parallelly for the completion of the complex speech perception.

The estimation of functional connectivity from the observed Blood Oxygen Level-Dependent (BOLD) signal may not be accurate because it is an indirect measure of neuronal activity or the existing deconvolution approaches assume that hemodynamic response function (HRF), which modulates the neuronal activities, is uniform across the brain regions or across the time course. We propose a novel approach using empirical mode decomposition (EMD), to reduce the effect of HRF from estimated neuronal activity signal (NAS) obtained after blind deconvolution for a BOLD time course. The first two intrinsic mode functions (IMFs), obtained during EMD of the neuronal activity signal represent its highest oscillating modes and hence have characteristic of impulses. The sum of the first two IMFs is computed as a refined representation of neuronal activity signal to estimate resting state connectome using the framework of dictionary learning. Usefulness of the proposed method has been demonstrated using two resting state datasets (healthy control and attention deficit hyperactivity disorder) taken from ‘1000 Functional Connectomes’. For quantitative analysis, Jaccard distances are computed between spatial maps obtained using BOLD signals and refined activity signals. Results show that maps obtained using NAS are a subset of that obtained using BOLD signal and hence avoid false acceptance of active voxels, which illustrates the importance of refined NAS.

One of the major causes of death in developing nations is the Alzheimer’s Disease (AD). For the treatment of this illness, is crucial to early diagnose mild cognitive impairment (MCI) and AD, with the help of feature extraction from magnetic resonance images (MRI). This paper proposes a 4-way classification of 3D MRI images using an ensemble implementation of 3D Densely Connected Convolutional Networks (3D DenseNets) models. The research makes use of dense connections that improve the movement of data within the model, due to having each layer linked with all the subsequent layers in a block. Afterwards, a probability-based fusion method is employed to merge the probabilistic output of each unique individual classifier model. Available through the ADNI dataset, preprocessed 3D MR images from four subject groups (i.e., AD, healthy control, early MCI, and late MCI) were acquired to perform experiments. In the tests, the proposed approach yields better results than other state-of-the-art methods dealing with 3D MR images.

Recent neuroimaging evidence suggest that there exists a unique individual-specific functional connectivity (FC) pattern consistent across tasks. The objective of our study is to utilize FC patterns to identify an individual using a supervised machine learning approach. To this end, we use two previously published data sets that comprises resting-state and task-based fMRI responses. We use static FC measures as input to a linear classifier to evaluate its performance. We additionally extend this analysis to capture dynamic FC using two approaches: the common sliding window approach and the more recent phase synchrony-based measure. We found that the classification models using dynamic FC patterns as input outperform their static analysis counterpart by a significant margin for both data sets. Furthermore, sliding window-based analysis proved to capture more individual-specific brain connectivity patterns than phase synchrony measures for resting-state data while the reverse pattern was observed for the task-based data set. Upon investigating the effects of feature reduction, we found that feature elimination significantly improved results upto a point with near-perfect classification accuracy for the task-based data set while a gradual decrease in the accuracy was observed for resting-state data set. The implications of these findings are discussed. The results we have are promising and present a novel direction to investigate further.

Previous research has shown that empathy, a fundamental component of human social functioning, is engaged when listening to music. Neuroimaging studies of empathy processing in music have, however, been limited. fMRI analysis methods based on graph theory have recently gained popularity as they are capable of illustrating global patterns of functional connectivity, which could be very useful in studying complex traits such as empathy. The current study examines the role of trait empathy, including cognitive and affective facets, on whole-brain functional network centrality in 36 participants listening to music in a naturalistic setting. Voxel-wise eigenvector centrality mapping was calculated as it provides us with an understanding of globally distributed centres of coordination associated with the processing of empathy. Partial correlation between Eigenvector centrality and measures of empathy showed that cognitive empathy is associated with higher centrality in the sensorimotor regions responsible for motor mimicry while affective empathy showed higher centrality in regions related to auditory affect processing. Results are discussed in relation to various theoretical models of empathy and music cognition.

Classification and prediction of posttraumatic stress disorder (PTSD) based on brain imaging measures is important because it could aid in PTSD diagnosis and clinical management of PTSD. The goal of the present study was to test the effectiveness of using cortical morphological measures (i.e. volume, thickness, and surface area) to classify PTSD cases and controls on 3571 individuals from the ENIGMA-Psychiatric Genomics Consortium PTSD Workgroup, the largest PTSD neuroimaging dataset to date. We constructed 6 feature sets from different demographic variables (age and sex) and cortical morphological measures and used four machine learning algorithms for classification: logistic regression, random forest, support vector machine, and multi-layer perceptron. We found that classifiers trained using only cortical morphological measures (any one of volume, thickness, or surface area) performed better than classifiers trained using only demographic variables. Among all 6 feature sets, combining demographic variables and all three cortical morphological measures yielded the best prediction accuracy, with area under the receiver operating characteristic curve (ROC AUC) scores ranging from 0.615 for logistic regression to 0.648 for random forest. These findings suggest that using cortical morphological measures only has modest prediction power for PTSD classification. Future studies that wish to produce clinically and practically significant findings should consider using whole brain morphological measures, as well as incorporating other neuroimaging modalities and relevant clinical and behavioral symptoms.

Unmanned segmentation of brain tumors is one of the hardest tasks to be solved in Computer Vision. In this work, we focus on Convolutional Neural Network model to segment tumorous cells in MRI brain scans. The inputs to the network are multi-channel MR image intensity information extracted from patches around each point to be predicted. The pre-processing steps are employed to precise the magnetic field bias and then intensity values are normalized using Z-score technique. The training was done for both HGG and LGG and the network was optimized with SGD in which the gradients are calculated using Nesterov Accelerated Gradient. The obtained results are promising for the complete tumor, the core tumor and the enhancing tumor segmentation. The propounded model achieved a dice score of 0.86, 0.62 and 0.65 for complete, core and enhancing tumor.

Neuroscience is a vast field of study, important for its role in human body and brain disorders. Neuroscientists tend to ask complicated questions that need a complex set of actions and multiple resources. A question resolution approach in this field should be able to address these issues.This study uses an ontology-based approach and creates codes that mirror the internal structure of questions, called templates, to translate questions to machine-understandable language. This research uses ontologies to expand queries, disambiguate terms, integrate resources, explore brain structures and create templates.

Neuronal signals allow us to understand how the brain operates and this process requires sophisticated processing of the acquired signals, which is facilitated by machine learning-based methods. However, these methods require large amount of data to first train them on the patterns present in the signals and then employ them to identify patterns from unknown signals. This data acquisition process involves expensive and complex experimental setups which are often not available to all – especially to the computational researchers who mainly deal with the development of the methods. Therefore, there is a basic need for the availability of open access datasets which can be used as benchmark towards novel methodological development and performance comparison across different methods. This would facilitate newcomers in the field to experiment and develop novel methods and achieve more robust results through data aggregation. In this scenario, this paper presents a curated list of available open access datasets of invasive neuronal signals containing a total of more than 25 datasets.

Epilepsy is one of the most recurrent brain disorders worldwide and mainly affects children. As a diagnostic support, the electroencephalogram is used, which is relatively easy to apply but requires a long time to analyze. Automatic EEG analysis presents difficulties both in the construction of the database and in the extracted characteristics used to build models. This article a machine learning-based methodology that uses a visual word bag of raw EEG images as input to identify images with abnormal signals. The performance introduces of the algorithms was tested using a proprietary pediatric EEG database. Accuracy greater than 95% was achieved, with calculation times less than 0.01 s per image. Therefore, the paper demonstrates the feasibility of using machine learning algorithms to directly analyze EEG images.

The study of the characteristics of hand tremors of the patients suffering from Parkinson’s disease (PD) offers an effective way to detect and assess the stage of the disease’s progression. During the semi-quantitative evaluation, neurologists label the PD patients with any of the (0–4) Unified Parkinson’s Diseases Rating Scale (UPDRS) score based on the intensity and prevalence of these tremors. This score can be bolstered by some other modes of assessment as like gait analysis to increase the reliability of PD detection. With the availability of conventional smartphones with a built-in accelerometer sensor, it is possible to acquire the 3-axes tremor and gait data very easily and analyze them by a trained algorithm. Thus we can remotely examine the PD patients from their homes and connect them to trained neurologists if required. The objective of this study was to investigate the usability of smartphones for assessing motor impairments (i.e. tremors and gait) that can be analyzed from accelerometer sensor data. We obtained 98.5% detection accuracy and 91% UPDRS labeling accuracy for 52 PD patients and 20 healthy subjects. The result of this study indicates a great promise for developing a remote system to detect, monitor, and prescribe PD patients over long distances. It will be a tremendous help for the older population in developing countries where access to a trained neurologist is very limited. Also, in a pandemic situation like COVID-19, patients from developed countries can be benefited from such a home-oriented PD detection and monitoring system.

Neurophysiological recordings, particularly neuronal signals recorded using multi-site neuronal probes or multielectrode arrays, are often contaminated with unwanted signals or artifacts from external or internal sources. Almost all types of neuronal signals including electroencephalogram (EEG), electrocorticogram (ECoG), local field potentials (LFP), and spikes very often suffer greatly from these artifacts and require extensive amount of processing to get rid of them. Despite considerable efforts in developing sophisticated methods to detect and remove these artifacts, it often appears a challenging task due to the inherent similar spatio-temporal properties of the artifacts and the recorded signals. In such cases, the incorporation of another modality can facilitate and improve the detection of these artifacts, and remove them. This paper focuses on the EEG signal and empirically analyses the role played by the addition of a new modality (e.g., cardiac signals, muscular signals, ocular signals, and motion signals) in detecting artifacts from EEG signals.

Fall is one of the most serious clinical problems faced by the elderly people. Elder people with neurodegenerative disorders like Parkinson disease often fall. This leads to the damage of physical condition and also mental condition. Therefore, elderly people should be taken care of all the time. However, it is not possible to take care of them every moment. Therefore, an automatic fall detection system is required to track elderly at any time. An automated fall detection system will provide timely assistance and hence, it will reduce medical care costs significantly. The recent developments in motion- sensor technologies have allowed the efficient use of wearable sensors in the overall treatment of the elderly. The paper presents a machine learning framework consisting of data collection, preprocessing of data, feature extraction and machine learning classifiers. They comprise C4.5, Random Forest, RepTree, and LMT (Logistic Model Tree). Dataset used in this research has been collected by using 3-axis accelerometer sensors which are mounted on a person’s waist. Features have been extracted from this dataset which are used by these classifiers. C4.5 gives the highest accuracy which is 97.36% in comparison to other classifiers.

Fall is deemed to be one of the critical problems for the elderly patient having neurological disorders as it may cause injury or death. It turns to be a public health concern and attracts researchers to detect fall using sensing devices wearable, portable, and imaging. With the availability of low cost pervasive sensing elements, advancement of ubiquitous computing and better understanding of machine learning approaches, researchers have employing various machine learning approaches in detecting fall from the sensor data. In this paper, we have proposed a recurrent neural network (RNN)-based framework for detecting fall/daily activity of a patient having a neurological disorder using Internet of things and then manage the patient by referring to doctor. If an anomaly is detected in the daily activity and notify caregiver/family member if fall is detected. The RNN based fall detection model fused knowledge from both the smartphone/wearable and camera installed on the wall and ceiling. The proposed RNN is trained with open-labeled and UR data-sets and is compared with the support vector machine and random forest for these two data-sets. The performance evaluation shows the proposed method is effecting and outperforms its counterparts.

Emotion regulation plays a major role in everyday life, as it enables individuals to modulate their emotions. Several strategies, for regulating emotions, can be used individually or simultaneously, such as suppression, rumination, acceptance, problem-solving, self-criticism, and experiential avoidance. This paper presents a temporal causal network model that simulates the employment of these seven emotion regulation strategies by a person experiencing varying intensity of anxiety. Simulation results are reported for both, the high and low, emotional intensity where the level of activation of these strategies vary with the intensity of negative emotions.

Anxiety is common in youth with autism spectrum disorder (ASD), causing unique lifelong challenges that severely limit everyday opportunities and reduce quality of life. Given the detrimental consequences and long-term effects of pervasive anxiety for childhood development and the covert nature of mental states, brain-computer interfaces (BCIs) represent a promising method to identify maladaptive states and allow for individualized and real-time mitigatory action to alleviate anxiety. Here we investigated the effects of slow paced breathing entrainment during stress induction on the perceived levels of anxiety in neurotypical adolescents and adolescents with autism, and propose a multi-class long short-term recurrent neural net (LSTM RNN) deep learning classifier capable of identifying anxious states from ongoing electroencephalography (EEG) signals. The deep learning classifier used was able to discriminate between anxious and non-anxious classes with an accuracy of 90.82% and yielded an average accuracy of 93.27% across all classes. Our study is the first to successfully apply an LSTM RNN classifier to identify anxious states from EEG. This LSTM RNN classifier holds promise for the development of neuroadaptive systems and individualized intervention methods capable of detecting and alleviating anxious states in both neurotypical adolescents and adolescents with autism.

Alcoholism involves psychological and biological components where multiple risk factors come into play. Assessment of the psychiatric emergency is a challenging issue for clinicians working with alcohol-dependent patients. Identifying alcoholics from healthy controls from their EEG signals can be effective in this scenario. In this research, we have applied two instance-based classifiers and three neural network classifier to classify Electroencephalogram data of alcoholics and normal person. For data preprocessing, we have applied discrete wavelet transform, Principal component analysis and Independent component analysis. After successful implementation of the classifiers, an accuracy of 95% is received with Bidirectional Long Short-Term Memory. Finally, comparing the performance of the two categories of algorithms, we have found that neural networks have higher potentiality against instance-based classifiers in the classification of EEG signals of alcoholics.

When the world is suffering from the deadliest consequences of COVID-19, people with autism find themselves in the worst possible situation. The patients of autism lack social skills, and in many cases, show repetitive behavior. Many of them need outside support throughout their life. During the COVID-19 pandemic, as many of the places are in lockdown conditions, it is very tough for them to find help from their doctors and therapists. Suddenly, the caregivers and parents of the ASD patients find themselves in a strange situation. Therefore, we are proposing an artificial intelligence-based system that uses sensor data to monitor the patient’s condition, and based on the emotion and facial expression of the patient, adjusts the learning method through exciting games and tasks. Whenever something goes wrong with the patient’s behavior, the caregivers and the parents are alerted about it. We then presented how this AI-based system can help them during COVID-19 pandemic. This system can help the parents to adjust to the new situation and continue the mental growth of the patients.

Alzheimer patient’s routine care at the onset of a catastrophe like coronavirus disease 2019 (COVID-19) pandemic is interrupted as healthcare is providing special attention to the patient having severe acute respiratory syndrome coronavirus 2 (SARS-COV-2) or COVID-19 infection. In order to decrease the spread of the disease, government has shut down regular services at the hospital, and advised all vulnerable people to stay at home and maintain social distance (of 3 fts) which hampered the routine care and rehabilitation therapy of elderly patient having a chronic disease like Alzheimer. On the other hand, the artificial intelligence (AI)-based internet of healthcare things allows clinicians to monitor physiological conditions of patients in real-time and machine learning models can able to detect any anomaly in the patient’s condition. Besides, the advancement in Information and Communication Technology enable us to provide special distance care (such as medication and therapy) by dedicated medical teams or special therapists. This paper discusses the effect of COVID-19 on patient care of Alzheimer’s Disease (AD) and how AI-based IoT can help special care of AD patients at home. Finally, we have outlined some recommendations for Family and Caregiver, Volunteer and Social Care which will help to develop the Government policy.

The contemporary world’s emerging issue is how the mental health and falling of a senior citizen with a neurological disorder can be maintained living at their homes as the number of aged people is increasing with the rising of life expectancy. With the advancement of the Internet of Things (IoT) and big data analytics, several works had been done on smart home health care systems that deal with in house monitoring for fall detection. Despite so much work, the challenges remain for not considering emotional care in the fall detection system for the old ones. As a remedy to the problems mentioned above, we propose an emotion aware fall monitoring framework using IoT, Artificial Intelligence (AI) Algorithms, and Big data analytics, which will deal with emotion recognition of the aged people, predictions about health conditions, and real-time fall monitoring. In the case of an emergency, the proposed framework alerts about a situation of urgency to the predefined caregiver. A smart ambulance or mobile clinic will reach the older adult’s location at minimum time.

Detecting emotions from the speech is one of the emergent research fields in the area of human information processing. Expressing emotion is a very difficult task for a person with neurological disorder. Hence, a Speech Emotion Recognition (SER) system may solve this by ensuring a barrier-less communication. Various research has been carried out in the area of SER. Therefore, the main objective of this research is to develop a system that can recognize emotion from the speech of a neurologically disordered person. Since convolutional neural network (CNN) is an effective method, it has been considered to develop the system. The system uses tonal properties like MFCCs. RAVDESS audio speech and song databases for training and testing. In addition, a custom local dataset developed to support further training and testing. The performance of the proposed system compared with the traditional machine learning models as well as with the pre-trained CNN models including VGG16 and VGG19. The results demonstrate that the CNN model proposed in this research performed better than the mentioned machine learning techniques. This system enables one tohhhhhh classify eight emotions of neurologically disordered person including calm, angry, fearful, disgust, happy, surprise, neutral and sad.

Cognitive performance dictates how an individual perceives, records, maintains, retrieves, manipulates, uses and expresses information and are provided in any task that the person is involved in, let it be from the simplest to the most complex. Therefore, it is imperative to identify how a person is cognitively engaging specially in tasks such as information acquisition and studying. Given the surge in online education system, this even becomes more important as the visual feedback of student engagement is missing from the loop. To address this issue, the current study proposes a pipeline to detect cognitive performance by analyzing electroencephalogram (EEG) signals using bidirectional multilayer long-short term memory (BML-LSTM). Tested on an EEG brainwave dataset from 10 students while they watched massive open online course video clips, the obtained results using BML-LSTM show an accuracy $${>}95\%$$ in detecting cognitive performance which outperforms all previous methods applied on the same dataset.

Brain-Computer Interface (BCI) has been a hot topic and an emerging technology in this decade. It is a communication tool between humans and systems using electroencephalography (EEG) to predicts certain aspects of cognitive state, such as attention or emotion. There are many types of sensors created to acquire the brain signal for different purposes. For example, the wet electrode is to obtain good quality, and the dry electrode is to achieve a wearable purpose. Hence, this paper investigates a comparative study of wet and dry systems using two cognitive tasks: attention experiment and music-emotion experiment. In attention experiments, a 3-back task is used as an assessment to measure attention and working memory. Comparatively, the music-emotion experiments are conducted to predict the emotion according to the user’s questionnaires. The proposed model is constructed by combining a shallow convolutional neural network (Shallow ConvNet) and a long short-term memory (LSTM) network to perform the feature extraction and classification tasks, respectively. This study further proposes transfer learning that focuses on utilizing knowledge acquired for the wet system and applying it to the dry system.

Mammalian hippocampus is involved in short-term formation of declarative memories. We employed a bio-inspired neural model of hippocampal CA1 region consisting of a zoo of excitatory and inhibitory cells. Cells’ firing was timed to a theta oscillation paced by two distinct neuronal populations exhibiting highly regular bursting activity, one tightly coupled to the trough and the other to the peak of theta. To systematically evaluate the model’s recall performance against number of stored patterns, overlaps and ‘active cells per pattern’, its cells were driven by a non-specific excitatory input to their dendrites. This excitatory input to model excitatory cells provided context and timing information for retrieval of previously stored memory patterns. Inhibition to excitatory cells’ dendrites acted as a non-specific global threshold machine that removed spurious activity during recall. Out of the three models tested, ‘model 1’ recall quality was excellent across all conditions. ‘Model 2’ recall was the worst. The number of ‘active cells per pattern’ had a massive effect on network recall quality regardless of how many patterns were stored in it. As ‘active cells per pattern’ decreased, network’s memory capacity increased, interference effects between stored patterns decreased, and recall quality improved. Key finding was that increased firing rate of an inhibitory cell inhibiting a network of excitatory cells has a better success at removing spurious activity at the network level and improving recall quality than increasing the synaptic strength of the same inhibitory cell inhibiting the same network of excitatory cells, while keeping its firing rate fixed.

Canonical vision models of the retina-to-V1 cortex pathway consist of cascades of several Linear+Nonlinear layers. In this setting, parameter tuning is the key to obtain a sensible behavior when putting all these multiple layers to work together. Conventional tuning of these neural models very much depends on the explicit computation of the derivatives of the response with regard to the parameters. And, in general, this is not an easy task. Automatic differentiation is a tool developed by the deep learning community to solve similar problems without the need of explicit computation of the analytic derivatives. Therefore, implementations of canonical visual neuroscience models that are ready to be used in an automatic differentiation environment are extremely needed nowadays. In this work we introduce a Python implementation of a standard multi-layer model for the retina-to-V1 pathway. Results show that the proposed default parameters reproduce image distortion psychophysics. More interestingly, given the python implementation, the parameters of this visual model are ready to be optimized with automatic differentiation tools for alternative goals.

Studying on EEG (Electroencephalography) data instances to discover potential recognizable patterns has been a emerging hot topic in recent years, particularly for cognitive analysis in online education areas. Machine learning techniques have been widely adopted in EEG analytical processes for non-invasive brain research. Existing work indicated that human brain can produce EEG signals under the stimulation of specific activities. This paper utilizes an optimized data analytical model to identify statuses of brain wave and further discover brain activity patterns. The proposed model, i.e. Segmented EEG Graph using PLA (SEGPA), that incorporates optimized data processing methods and EEG-based analytical for EEG data analysis. The data segmentation techniques are incorporated in SEGPA model. This research proposes a potentially efficient method for recognizing human brain activities that can be used for machinery control. The experimental results reveal the positive discovery in EEG data analysis based on the optimized sampling methods. The proposed model can be used for identifying students cognitive statuses and improve educational performance in COVID19 period.

The neuronal paradigm of studying the brain has left us with limitations in both our understanding of how neurons process information to achieve biological intelligence and how such knowledge may be translated into artificial intelligence and even its most brain-derived branch, neuromorphic computing. Overturning our assumptions of how the brain works, the recent exploration of astrocytes reveals how these long-neglected brain cells dynamically regulate learning by interacting with neuronal activity at the synaptic level. Following recent experimental studies, we designed an associative, Hopfield-type, neuronal-astrocytic network and analyzed the dynamics of the interaction between neurons and astrocytes. We show how astrocytes were sufficient to trigger transitions between learned memories in the network and derived the timing of these transitions based on the dynamics of the calcium-dependent slow-currents in the astrocytic processes. We further evaluated the proposed brain-morphic mechanism for sequence learning by emulating astrocytic atrophy. We show that memory recall became largely impaired after a critical point of affected astrocytes was reached. These results support our ongoing efforts to harness the computational power of non-neuronal elements for neuromorphic information processing.

Classification of sleep-wake is necessary for the diagnosis and treatment of sleep disorders, and EEG is normally used to assess sleep quality. Manual scoring is time-consuming and requires a sleep expert. Therefore, automatic sleep classification is essential. To accomplish this, features are extracted from the time domain, frequency domain, wavelet domain, and also from non-linear dynamics. In this study, a novel Jaya Optimization based hyper-Parameter and feature Selection (JOPS) algorithm is proposed to select optimal feature subset as well as hyper-parameters of the classifier such as KNN and SVM, simultaneously. JOPS is self-adaptive that automatically adapts to the population size. The proposed JPOS yielded the accuracy of 94.99% and 94.85% using KNN and SVM, respectively. JPOS algorithm is compared with genetic algorithm and differential evaluation-based feature selection algorithm. Finally, a decision support system is created to graphically visualize the sleep-wake state which will be beneficial to clinical staffs. Furthermore, the proposed JOPS can not only be used in sleep-wake classification but could be applied in other classification problems.