Poster Presentations

Evidences suggest the brain function and gut motility signals influence one another. Many brain regions underpinning cognitive behavior regulate autonomous nervous control which influences the gut motility; alternatively, gut motility altered broadly through satiation or more specifically through certain nutrient or medication intake, has the potential to influence the brain. However, understanding the precise gut electrical instincts and their modulations on the brain rhythms has not been carefully explored during diverse cognitive brain states. Recently as a principal investigator of a pilot research study (supported by the Kavli Foundation) with N=20 healthy subjects, we for the first time explored the brain-gut interactions through electrophysiological recordings. We collected EEG activations (amplitude) and EGG (phase, sourced by the interstitial Cajal cells, operating in 0.05Hz), respectively from the brain and the gut, and performed phase-amplitude coupling analysis. The findings from the study suggested a significant, spatially selective modulation of the gut in the brain, that depend on the underlying cognitive engagement dynamics. More precisely, we found that the 1) gut-brain coupling depend on the state of hunger, where hunger evoked significant right insula activations and satiety evoked parietooccipital activations. Interestingly, the 2) modulation index also depend on the cognitive state of the participant—that is, response control or emotion processing or working memory selective brain engagements. Altogether the research opens doors to new domain of inquiry to understand the effect of the brain-gut axis on cognition and behavior.

A fundamental challenge in neuroscience is to understand how the activities across brain regions are coordinated. Genomic investigation can systematically uncover genes involved in such coordination, but current genomic studies have focused mostly on within-region analysis of healthy vs. disease states. This study performed an inter-region differential correlation (DC) analysis on post-mortem human brain RNA seq data (87 control vs. 131 Alzheimers Disease (AD) individuals) from Mount Sinai Brain Bank, and identified how AD rewires the gene-gene correlation structure across four brain regions, independent of disruptions in cellular composition. Each brain region uses a different set of DC genes while interacting with other brain regions, and partitioning such DC gene pairs into bipartite (region-region) communities revealed that synaptic signaling and transporter activities are the most altered biological processes, further uncovering genes that have previously not been identified as AD biomarkers. Thus, inter-region comparison provides a new perspective for comprehending AD aetiology.

Locomotion and navigation require the perception of surrounding space and embodied decision making to respond appropriately to an incoming stimulus. Animals are highly adept at such tasks, and the extraordinary agility and dexterity exhibited are highly desirable for legged robots. A comparative study between three methods to solve the quadruped locomotion and navigation problem was performed to develop a hierarchical controller for the quadruped platform called MePed. The three methods consisted of the use of a Multi-Layer Perceptron (MLP), a hybrid neural network combining Oscillatory and Deep Neural Networks (DNNs) and a Central Pattern Generator (CPG) parameterised by an MLP respectively for the lower level of the proposed control system. The lower level acts as a pattern generator, as seen in the spinal cord, and the higher levels perform higher brain functions. This project focuses on emulating gait when the hierarchical controller is trained using Reinforcement Learning.

Place cells are neurons in the hippocampus that fire at a high rate when an animal is in a particular location in the environment. In an experimental study, rodents move in a 3D cubic lattice structure (the lattice's face is parallel to the ground). Place fields recorded from the rats were homogeneously distributed throughout the lattice mazes and also showed some exciting properties wherein the place fields were oriented parallel to the maze axes. The fields were also elongated more in the vertical direction thus suggesting that vertical spatial encoding was less informative. To model this experiment, we simulated the rat's trajectory. After that, the low pass filter is used for averaging the path integration (PI), thus removing the high-frequency terms. The output of the PI layer is input to the Autoencoder layer that has a non-linear activation function. The Autoencoder hidden layer mimics the place cells and represents the firing of the place cell at a particular location in the trajectory. Finally, we can replicate the main results in the aligned case as observed in the experimental study, thus validating our Autoencoder-based Spatial Navigation model.

The hippocampal cells broadly categorized as spatial cells have a key role in storage of experience. We devise a computational model for understanding theta-sequences during linear motion. Theta sequences are “clear, ordered sequences” that are observed in the theta wave with segments of it reflecting the position, time during motion.

We present a network model centrally built with oscillatory neurons as input and has multiple layers. The first layer is the Path integration (PI) layer that encodes the displacement in the preferred direction (forward or backward) by encoding a scaling factor of β and the speed. The output is fed to layers of stacked auto-encoder that are fed to a hidden layer which acts as the regressor to predict the velocity.

We are able to replicate the firing pattern observed in the case of theta sequences observed in one-dimensional motion. The output thus is helpful in observing theta sequences based on the underlying spatio-temporal cells in the model that extends the applicability of the current oscillator-based modelling framework to understand navigation and learning.

No need to clarify the fact that consuming alcohol has some serious effect on the human brain and hampers our daily lifestyle. Also, it may cause difficulties in recalling memories, instability, and even blackout. Recognizing early symptoms of substance dependence and having adequate care in the rehabilitation phase will make a big difference. The screening test for patients’ alcohol dependence was arbitrary and could misinterpret the true level of alcohol consumption in some cases. Although the paradigm of neuroimaging (EEG) showed positive outcomes of research in obtaining objective findings when assessing and diagnosing intoxicated patients. This work extract features from EEG brain signals and then optimizes the collection of features using mutual information, feature importance, LASSO regularization, and the RFE method step by step. The optimized features set is then considered for detecting AUD (Alcohol Use Disorder) patients and healthy persons. Super Learning approaches were adopted for the classification task. This is accomplished by bagging and boosting results from a set of machine learning models for classification. The findings reveal that the ensemble method of feature optimization accompanied by the hybrid super-learning classification provides better performance. The proposed approach has experimented with EEG data set from the UCI Machine Learning repository and the experimental results substantiate the efficacy of the approach and are also comparable to the state-of-the-art approaches.

Though text detection and recognition are considered as the classical problems, they managed to thrive and now stand as one of the most involving and demanding problems. Unlike English, recognition engines for Indian scripts are harder to develop, because of the scarcity of annotated data, complex strokes, and a myriad number of characters. Having a motivation to build a strong recognition system for Indian scripts, we created a general pipeline for data synthesis, detection, preprocessing and annotation, and trained the recognition models for Telugu and Tamil. The whole pipeline can digitize both printed and handwritten words at a document level with CER of below 1% and 5% respectively. Further, the pipeline can be effortlessly extended to other Indian scripts. We also developed a robust non-script specific text detection pipeline to locate text, enhance quality and maintain spatial relevance in hard layouts and deteriorated text regions.

The vascular network is in general considered as an irrigation network that ensures a timely

supply of oxygen and nutrients to the neural tissues. The neurons are most often considered

as the major information processing units in the brain. But recent studies have shown that

vascular network also exhibits tuned response. Tuned response to input stimuli is made

possible by means of competitive learning, which suggests that the vessels also might be

having trainable lateral connections. To explore the possibility of trainable lateral connections

and tuned responses in vessels, we modeled the neurovascular network by interconnecting

two laterally connected self-organizing networks. The study showed that the vessels should

ideally display lateral connectivity with an ON center OFF surround architecture made

possible by trainable connections. Moreover, the study also showed that the representation of

the input stimulus by means of map formation in the neural network gets transitioned from a

columnar map to a salt and pepper map depending on the radius of the perfusion field of the

vessels.

This poster discusses the contradiction of our understanding that the Central Nervous System is damaged in the neurodegenerative diseases. Currently we accept that the neurons in the brain gets degenerated and these cannot be regenerated. But ample of evidences contradict this understanding.

Also, neuromuscular junctions and spinal nerve (Central Nervous System – Peripheral Nervous System junctions) are highly vulnerable to damage caused by physical stress and pressure. Whereas the brain is highly protected and any defect in the brain would end a person's life and not just cause a progressive disease.

Human decision making often involves exploration and exploitation, which may be impacted by anxiety. The goal of this study was to investigate the interaction between different exploration strategies (value-driven or random) and anxiety, using a computational model. 32 healthy volunteers (18 F, mean age 25) performed 2 versions of a 4-armed aversive (probabilistic shock) bandit task, in which participants were asked to choose the arm perceived to have the lowest probability of shock. In one version, the best strategy relies on purely value-driven exploration (choosing to explore only when there is a perceived better choice); whereas, in the other, it involves random exploration. The performance metrics include: 1) accuracies and 2) persistent exploitation. While there was no significant difference in accuracies between versions, persistent exploitation showed a significant difference between versions (paired t-test, p = 0.0005). Additionally, the persistent exploitation difference score showed a strong correlation with the state STAI score (R = -0.55, p = 0.0035). Participants exhibited higher persistent exploitation in the version favoring random exploration. Furthermore, individual differences in persistent exploitation between versions decrease with increasing state anxiety scores. Thus, state anxiety might influence exploratory behavior, with higher scores indicating an ambivalence between random and value-driven exploration.

Restoring movements after hemiparesis caused by stroke is an ongoing challenge in the field of rehabilitation. In this study, we use a convolutional neural network, trained to perform bilateral reaching movements in 3D space, to map patient characteristics with rehabilitation parameters. The network was designed with bilateral symmetry to reflect the bilaterality of cerebral hemispheres with the two halves joined by cross-connections, analogous to Corpus Callosum. 3 patient characteristics were introduced in the network – lesion size, recovery stage and structural integrity of cross-connections. Similarly, 3 parameters were used to administer rehabilitation to the network – movement complexity, hand selection mode (unimanual vs bimanual), and extent of plasticity. Upon analysis, we found that the regardless of patient characteristic the network improved better with increase in movement complexity. Level of plasticity, however, depended very much on the network’s structural integrity. Thus, understanding patient characteristics is necessary for efficient post-stroke rehabilitation.

Various aspects of cognitive abilities include the ability to process goal oriented information, perform selective attention, perform response inhibition, suppress irrelevant information that act as distractors, store and retrieve information etc. Much research has focused on studying these individual components in isolation, whereas multiple aspects of cognition are altered in a range of neuropsychiatric disorders. Hence it is important to study multiple cognitive abilities within the same individual. A scalable mobile platform (BrainE) is developed by NEAT labs that assays several essential aspects of cognition in this direction.

In our study we develop an agent computational model of decision making which can assess the various cognitive abilities such as selective attention, response inhibition, distractor processing and working memory properties using a unified framework consisting of a deep neural network (DNN) with convolutional auto encoder (CAE) for stimulus representation that represents the visual system, hidden layer consisting of neurons that maps the properties of medium spiny neurons representing the Orbito Frontal Cortex(OFC), the dorsolateral pre-frontal cortex(DLPFC) and the inferior frontal gyrus (IFG), a layer equivalent of Ventral tegmental area (VTA) representing the reward system, Q-Learning based stimulus to action mapping and action selection circuitry, Anterior cingulate cortex(ACC), which facilitates the action selection and a coupled oscillatory network representing frontopolar cortex(FPC), which facilitates the tuning of exploratory parameter.

We successfully modelled the selective attention and response inhibition using Go-Green Task-where participants have to favourably respond only to the presentation of green coloured rockets, distractor processing using the Middle Fish task ( participants have to select Left/Right based on the direction of the middle fish ignoring the directions of surrounding fishes) and working memory processing using the Lost star task (where participants are initially presented with a perceptual image and they have to detect the missing star when a probe is presented). The Model results were found to be comparable with the Experimental results based on the Reaction time, Speed, Accuracy, Efficiency and Consistency. By tuning the threshold of the ACC race model (thr) and lateral connectivity strengths of FPC subsystem , we were able to simulate different subject profiles. This model has the potential to be extended to act as a proxy for real patients, which will facilitate us to extensively test diverse scenarios and come up with therapeutic intervention for disease conditions.

A special class of hippocampal neurons broadly known as the spatial cells, whose subcategories include place cells, grid cells and head direction cells, are considered to be the building blocks of the brain’s map of the spatial world. We present a general, deep learning-based modeling framework that describes the emergence of the spatial cell responses and can also explain behavioral responses that involve a combination of path integration and vision. The first layer of the model consists of Head Direction (HD) cells that code for preferred direction of the agent. The second layer is the path integration (PI) layer with oscillatory neurons: displacement of the agent in a given direction modulates the frequency of these oscillators. Principal Component Analysis (PCA) of the PI cell responses showed emergence of cells with grid-like spatial periodicity. We show that the response of these cells could be described by Bessel functions. The output of PI layer is used to train stack of autoencoders. Neurons of both the layers exhibit responses resembling grid cells and place cells. The paper concludes by suggesting a wider applicability of the proposed modeling framework beyond the two simulated behavioral studies.

We have designed a novel oscillatory based architecture which is different than already existed rate code and spike code model, this model is more biologically plausible where ensemble of neuron activity is measured. In this network architecture multiple interconnected complex hopf oscillators are used to learn intrinsic property like frequency, normalized phase and amplitude of ensemble neuronal activity or EEG activity. Our network is organized into two stages of training (Biswas et al.,2021). In 1st stage the underlying dominating frequency components of the cortical activity hidden in the input EEG. Magnitude and absolute phase relationship of the individual frequency components is also learnt in 2nd phase. Here we increase the duration (20sec) and frequency range (0.1 to 20Hz) of the signal and we analyzed all five sleep stages which is the extension of our last publication. By changing bifurcation parameter and coefficient of interconnected weight oscillator we are able to predict signal more accurately.

Ref: Biswas, D., Pallikkulath, S., and Chakravarthy, V. S. (2021). A Complex-Valued Oscillatory Neural Network for Storage and Retrieval of Multidimensional Aperiodic Signals. Front. Comput. Neurosci. 15, 1–24. doi:10.3389/fncom.2021.551111.

Neurons generate precise sequences of activity during the sensory-motor processing occurring during an organism's navigation in the environment. Evolution has produced a representation manifold that can be linked to position, an inertial compass and internal coordinates. We propose a novel approach to model the properties of place cells. Two important properties of place cells are population coding and phase precession. In our model, we encode an animal's trajectory using a minimum uncertainty state wavefunction Ψ, which captures the dynamics of a population of place cells. Ψ has minimum uncertainty and its absolute value is peaked on the trajectory. The time dependent Hamiltonian of our system is determined by the trajectory. After fine-tuning of parameters, our simulations show that our wavefunction captures the phenomenology of population coding and phase precession for any random, bounded 2D trajectory.

Sign Language is a potential tool for communication in the hearing and speech-impaired community. As individual words cannot be communicated accurately using the sign language gestures, fingerspelling is adopted to spell out names and places. Due to rich vocabulary and diversity in Indic scripts, and the abugida nature of Indic scripts that distinguish them from a prominent world script like the Roman script, it is cumbersome to use American Sign Language (ASL) convention for finger-spelling in Indian languages. Moreover, due to the existence of 10 major scripts in India, it is a futile task to develop separate fingerspelling convention for each individual Indic script based on the geometry of the characters. Therefore, we propose a novel and unified fingerspelling system known as Mudrabharati for Indic scripts in general. The gestures of Mudrabharati are constructed based on the phonetics of Indian scripts and not the geometry of the glyphs that compose the individual characters. Unlike ASL that utilizes just one hand, Mudrabharati uses both the hands - one for consonants and the other for vowels; swarayukta aksharas (Consonant-Vowel Combinations) are gestured by using both the hands. An Artificial Intelligence (AI) based recognition system for Mudrabharati that returns the character in Devanagari and Tamil scripts is developed.

Multiple sclerosis (MS) is a chronic autoimmune, inflammatory neurological disease, widely associated with Grey and white matter degradation due to demyelination of axons. Thus exposing the underlying causes of this condition can lead to a novel treatment approach for Multiple Sclerosis. The total RNA microarray processed data from GEO for Multiple sclerotic patients was comprehensively analyzed to find out underlying differences between Grey Matter lesions, Normal appearing Grey Matter and Control Grey matter at the transcriptomics level. Thus, in the current study, we performed various bioinformatics analyses on transcriptional profiles of a total of 184 Total-RNA-seq samples including 105 NAMG, 37 GML, and 42 Controls that were obtained from the NCBI-Bio project (PRJNA543111). Conclusively, our study depicted significant differences in the gene expression patterns between GML and CG samples. As a result, 20736 genes (padj. value <0.05, log2 fold change ±1.5) were found to be significantly differentially expressed among these conditions based on differential gene expression analysis. This study reveals the genes like OR10A7, OR11L1, OR2AG1, OR2C3, ORT25, OR4D6, OR52E4, 0R5D16, OR5K1, OR7D4, OR8K1, and ORAI3 as the key features that may substantially contribute to the loss of Olfactory senses in MS patients. Our study also proposes the involvement of Protein Kinase-A in the pathogenesis of Multiple sclerosis. Eventually, the results presented here reveal new insights into MS and its relation with the biological pathways especially Olfactory transduction and Hematopoietic stem cell lineage pathways.

Majority of stroke patients suffer from Hemiparesis, characterized by weakness and partial loss of function in one side of the body. To ensure efficient rehabilitation, proper motor assessment pre- and post-intervention is necessary. However, increasing patient to physician ratio, makes this process difficult. In this study, we propose a fully automated system that uses webcam and computer vision to assess the upper extremity function in subject. The framework used processes the localized joints, detected with the help of the webcam, frame by frame, and using a rule-based classifier returns the score for the corresponding movement. The test movements and scoring system are taken from the popular Fugl-Meyer assessment (FMA) scale. In addition to the qualitative FMA scores, the extracted features such as movement time, velocity, range of motion provides a quantitative estimate of improvement over time. The automated, cost-efficient framework of the system aids the patient to monitor their progress more frequently from their homes.

Every year, around 1.8 million Indians suffer from stroke. Out of this, majority are left disabled for life. This can be avoided with proper physical rehabilitation. However, due to low therapist to patient ratio, scarce rehabilitation centres, high costs and highly repetitive therapy leading to lack of motivation, most survivors discontinue treatment. Virtual Reality with its engaging aesthetics and rich variety is emerging as a beneficial tool for rehabilitation. As not many stroke units in India provide this facility, we propose a 'Comprehensive Rehabilitative Gaming System' (cRGS), which aims at affordable, in-house Stroke Rehabilitation, targeting upper extremity. The system comprises Virtual Environments (VE), equipped with multiple games, that are designed to be interactive and customizable to the needs of the patient. Although VR setups typically use expensive sensors to track movements, the system proposed here makes use of a webcam and Computer Vision instead, thereby making the system affordable.

Modelling human decision making through heuristic and Bayes optimal approaches is often deemed irreconcilable in nature, although recently, bounded rationality has been proposed as a bridge between the two. In this study, we investigate a binary classification task in human subjects under varying time constraints. Regression analysis of subject responses shows that with varying time constraints, performance accuracy as well as feature dependency, both show a qualitative transition. While lower time constraints encourage weighing of multiple stimulus features representing a Bayes optimal approach, higher time constraints limit feature use and can be understood as a cue discounting method of heuristic decision making. A bounded rationality model was fitted to the observations by varying information constraints and was seen to successfully capture behavioural change thus, representing both, the adaptiveness as well as the efficiency of the heuristic and optimal decision making processes respectively.

Working memory system in the brain that combines the temporary storage and manipulation of the information in the service of reasoning and the guidance of decision-making tasks. To maintain this, our brain scans the entire image piecewise by attending to only a small region of the entire big picture and part by part aggregates the entire information given in the image, with fading memory of the information represented by the parts focused at very early on and best recollection of the most recently focused regions. Similar to this, we propose a dual channel multilayered convolutional recurrent neural network architecture to solve the image reconstruction problem. We model the recurrent connections according to the architecture, consisting of a network with both Elman and Jordan layers as recurrent connections.

The neurovascular coupling involves a continuous bidirectional interaction between neural and vascular networks. Hence there is an expectation, supported by experimental evidence, that even the development of both neural and cerebrovascular systems should be interdependent. The proposed Vascular Arborization Model (VAM) is intended to capture the effect of neural activity on vascular arborization.

The VAM describes three major stages for vascular tree growth (i) The prenatal growth phase, where the vascular arborization depends on the cytoarchitecture of neurons and non-neural cells (ii) The post-natal growth phase during which the further arborization of the vasculature depends on neural activity also in addition to neural cytoarchitecture (iii) The settling phase, where the fully grown vascular tree repositions its vascular nodes to ensure minimum path length and wire length.

The VAM model was simulated to grow the vasculature in neonatal rat whisker barrel cortex under two conditions (i) Control – where the whiskers were intact and (ii) Lesioned where one row of whiskers was cauterized. The model captured a significant reduction in vascular branch density in lesioned animals compared to control animals, thus agreeing to experimental observation.

Spatial semantics is the study of the meaning of spatial language. We can define the objective of the study of spatial semantics as to study spatial expressions, that is, conventional specifications of the location or change of location of a given entity (Zlatev, 2012). The prepositions used to represent spatial relations are referred to as spatial prepositions. There are strong parallels between space and other semantic domains, reflected in the fact that the same expressions often take spatial, temporal, and other more abstracts meanings, as seen in expressions such as from here to there, from now to tomorrow, and from me to you (Gruber 1965; Anderson 1971; Clark 1973)

The experiment aims to study spatial prepositions using a Transformer (Vaswani et al., 2017) based computational model to understand how the network generalizes spatial relations. The model uses images from a visual genome dataset’ (Krishna et al., n.d.) containing spatial relations for training.

The phenomenon of forgetting old information while learning new information is called catastrophic forgetting/interference. The human brain overcomes this problem quite effectively, a problem that continues to challenge current deep neural network models. We propose a regularization-based model to solve the problem of catastrophic forgetting. According to the proposed training mechanism, the network parameters are constrained to vary in a direction orthogonal to the average of the error gradients corresponding to the previous tasks. We also ensure that the constraint used in parameter updating satisfies the locality principle.

The proposed model's performance is compared with Elastic Weight Consolidation on permuted MNIST, split MNIST tasks on classification tasks using fully connected networks, and Convolution-based networks. The model performance is also compared to an autoencoder on split MNIST dataset, and to complex core50 dataset on two types of classification tasks with EWC.

The proposed model gives a new view on plasticity at the neuronal level. In the proposed model, the parameter updating is controlled by the neuronal level plasticity rather than synapse level plasticity as in other standard models. The biological plausibility of the proposed model is discussed by linking the extra parameters to synaptic tagging, which represents the state of the synapse involved in Long Term Potentiation (LTP).

Understanding of neuro-cognitive network mechanisms in rest and task requires accurate spatiotemporal characterization of the relevant functional brain networks.

Here, we introduce a data-driven unsupervised approach to characterize the high dimensional dynamic functional connectivity (dFC) into dynamics of lower dimensional patterns. The present study investigates the stability of whole-brain temporal dynamics in resting state, movie watching and sensorimotor task in a healthy ageing cohort. Our primary results indicate temporal dynamics of naturalistic movie watching task is closer to resting state than the sensorimotor task. Our analysis revealed an overall trend of highest temporal stability of dFC in the sensorimotor task, followed by naturalistic movie watching task and resting state that remains similar in both young and old adults. Temporal stability of dFC in resting state is higher in younger adults than in elderly. Further on, our analysis revealed the evolution of dFC is neither random nor markovian.

Legged animal locomotion is based on periodic limb movements. The neural circuits underlying various rhythmic motor behaviors can be traced to the central pattern generator (CPG). Hence, bio-inspired robotics aims to employ CPGs to control limb movement for synchronized locomotion. A CPG can produce coordinated rhythmic output signals; ideal to be implemented by a system of coupled limit-cycle oscillators in hardware.

A single programmable relaxation oscillator can be realized by placing a two–terminal memristive device composed of Vanadium Dioxide (VO2) in series with a MOSFET and a capacitor. We consider a network of four such oscillators connected in ring topology with capacitive nearest-neighbor bidirectional coupling. The coupling capacitance CC controls the inhibitory coupling strength. We demonstrate a six-gait neuromorphic CPG by exploring the dynamics of a ring network by modulating the coupling strengths between oscillators. The network can closely generate all the primary walking gait patterns observed in quadruped animals.

Human-error is to blame for upto 90% of workplace accidents and can cost money, time, and interrupt workplace productivity. The purpose of this study is to test whether a machine learning classifier can generalize errors across EEG data from different tasks, which were, in this case, the Flanker Task and the Active Oddball Task, both available on the EEGLAB Dataset(UCSD). A Support Vector Machine was trained in Matlab on the Flanker Task, where major changes across all channels in multiple time frames were used as features and tested for their resemblance to an ERN(Error-Related Negativity). The channels Fp1, Fp2, Af7, Af8, and, Fz were the ones with the strongest error-correlation in the Flanker Task. The SVM was then able to classify over 95% of the errors in the Active Oddball Task, with a different set of defining channels. In future work, we would classify errors in more natural, real-world situations.