V Srinivasa Chakravarthy

We develop a sequential Q-learning model using a recurrent neural network to count objects in images using attentional search. The proposed model, which is based on visual attention, scans images by making a sequence of attentional jumps or saccades. By integrating the information gathered by the sequence of saccades, the model counts the number of targets in the image. The model consists primarily of two modules: the Classification Network and the Saccade Network. Whereas the Classification network predicts the number of target objects in the image, the Saccade network predicts the next saccadic jump. When the probability of the best predicted class crosses a threshold, the model halts making saccades and outputs its class prediction. Correct prediction results in positive reward, which is used to train the model by Q-learning. We achieve an accuracy of 92.1% in object counting. Simulations show that there is a direct relation between the number of glimpses required and the number of objects present to achieve a high accuracy in object counting.

Parkinson's disease (PD) is a neurodegenerative disorder caused by loss of dopaminergic neurons in substantia nigra pars compacta (SNc). Although the exact cause of cell death is not clear, the hypothesis that metabolic deficiency is a key factor has been gaining attention in recent years. In the present study, we investigated this hypothesis using a multi-scale computational model of the subsystem of the basal ganglia comprising the subthalamic nucleus (STN), globus pallidus externa (GPe), and SNc. The proposed model is a multiscale model in that interaction among the three nuclei are simulated using more abstract Izhikevich neuron models, while the molecular pathways involved in cell death of SNc neurons are simulated in terms of detailed chemical kinetics. Simulation results obtained from the proposed model showed that energy deficiencies occurring at cellular and network levels could precipitate the excitotoxic loss of SNc neurons in PD. At the subcellular level, the models show how calcium elevation leads to apoptosis of SNc neurons. The therapeutic effects of several neuroprotective interventions are also simulated in the model. From neuroprotective studies, it was clear that glutamate inhibition and apoptotic signal blocker therapies were able to halt the progression of SNc cell loss when compared to other therapeutic interventions, which only slowed down the progression of SNc cell loss.

Parkinson’s disease (PD) is the second most prominent neurodegenerative disease around the world. Although it is known that PD is caused by the loss of dopaminergic cells in substantia nigra pars compacta (SNc), the decisive cause of this inexorable cell loss is not clearly elucidated. We hypothesize that “Energy deficiency at a sub-cellular/cellular/systems level can be a common underlying cause for SNc cell loss in PD.” Here, we propose a comprehensive computational model of SNc cell, which helps us to understand the pathophysiology of neurodegeneration at the subcellular level in PD. The aim of the study is to see how deficits in the supply of energy substrates (glucose and oxygen) lead to a deficit in adenosine triphosphate (ATP). The study also aims to show that deficits in ATP are the common factor underlying the molecular-level pathological changes, including alpha-synuclein aggregation, reactive oxygen species formation, calcium elevation, and dopamine dysfunction. The model suggests that hypoglycemia plays a more crucial role in leading to ATP deficits than hypoxia. We believe that the proposed model provides an integrated modeling framework to understand the neurodegenerative processes underlying PD.

Although there is a plethora of modeling literature dedicated to the object recognition processes of the ventral (“what”) pathway of primate visual systems, modeling studies on the motion-sensitive regions like the Medial superior temporal area (MST) of the dorsal (“where”) pathway are relatively scarce. Neurons in the MST area of the macaque monkey respond selectively to different types of optic flow sequences such as radial and rotational flows. We present three models that are designed to simulate the computation of optic flow performed by the MST neurons. Model-1 and model-2 each composed of three stages: Direction Selective Mosaic Network (DSMN), Cell Plane Network (CPNW) or the Hebbian Network (HBNW), and the Optic flow network (OF). The three stages roughly correspond to V1-MT-MST areas, respectively, in the primate motion pathway. Both these models are trained stage by stage using a biologically plausible variation of Hebbian rule. The simulation results show that, neurons in model-1 and model-2 (that are trained on translational, radial, and rotational sequences) develop responses that could account for MSTd cell properties found neurobiologically. On the other hand, model-3 consists of the Velocity Selective Mosaic Network (VSMN) followed by a convolutional neural network (CNN) which is trained on radial and rotational sequences using a supervised backpropagation algorithm. The quantitative comparison of response similarity matrices (RSMs), made out of convolution layer and last hidden layer responses, show that model-3 neuron responses are consistent with the idea of functional hierarchy in the macaque motion pathway. These results also suggest that the deep learning models could offer a computationally elegant and biologically plausible solution to simulate the development of cortical responses of the primate motion pathway.

A large-scale model of brain dynamics, as it is manifested in functional neuroimaging data, is presented in this study. The model is built around a general trainable network of Hopf oscillators, the dynamics of which are described in the complex domain. It was shown earlier that when a pair of Hopf oscillators are coupled by power coupling with a complex coupling strength, it is possible to stabilize the normal phase difference at a value related to the angle of the complex coupling strength. In the present model, the magnitudes of the complex coupling weights are set using the Structural Connectivity information obtained from Diffusion Tensor Imaging (DTI). The complex-valued outputs of the oscillator network are transformed by a complex-valued feedforward network with a single hidden layer. The entire model is trained in 2 stages: in the 1 st stage, the intrinsic frequencies of the oscillators in the oscillator network are trained, whereas in the 2 nd stage, the weights of the feedforward network are trained using the complex backpropagation algorithm. The Functional Connectivity Matrix (FCM) obtained from the network’s output is compared with empirical Functional Connectivity Matrix, a comparison that resulted in a correlation of 0.99 averaged over 5 subjects.

Sign Language (SL) is a potential tool for communication in the hearing and speech-impaired community. As individual words cannot be communicated accurately using the SL gestures, fingerspelling is adopted to spell out names of people 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 fingerspelling in Indian languages. Moreover, due to the existence of 10 major scripts in India, it is a futile task to develop a separate fingerspelling convention for each individual Indic script based on the geometry of the characters. In this paper, we propose a novel and unified fingerspelling system known as Mudrabharati for Indic scripts. 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.

Bipolar disorder is characterized by mood swings—oscillations between manic and depressive states. The swings (oscillations) mark the length of an episode in a patient’s mood cycle (period), and can vary from hours to years. The proposed modeling study uses decision making framework to investigate the role of basal ganglia network in generating bipolar oscillations. In this model, the basal ganglia system performs a two-arm bandit task in which one of the arms (action responses) leads to a positive outcome, while the other leads to a negative outcome. We explore the dynamics of key reward and risk related parameters in the system while the model agent receives various outcomes. Particularly, we study the system using a model that represents the fast dynamics of decision making, and a module to capture the slow dynamics that describe the variation of some meta-parameters of fast dynamics over long time scales. The model is cast at three levels of abstraction: (1) a two-dimensional dynamical system model, that is a simple two variable model capable of showing bistability for rewarding and punitive outcomes; (2) a phenomenological basal ganglia model, to extend the implications from the reduced model to a cortico-basal ganglia setup; (3) a detailed network model of basal ganglia, that incorporates detailed cellular level models for a more realistic understanding. In healthy conditions, the model chooses positive action and avoids negative one, whereas under bipolar conditions, the model exhibits slow oscillations in its choice of positive or negative outcomes, reminiscent of bipolar oscillations. Phase-plane analyses on the simple reduced dynamical system with two variables reveal the essential parameters that generate pathological ‘bipolar-like’ oscillations. Phenomenological and network models of the basal ganglia extend that logic, and interpret bipolar oscillations in terms of the activity of dopaminergic and serotonergic projections on the cortico-basal ganglia network dynamics. The network’s dysfunction, specifically in terms of reward and risk sensitivity, is shown to be responsible for the pathological bipolar oscillations. The study proposes a computational model that explores the effects of impaired serotonergic neuromodulation on the dynamics of the cortico basal ganglia network, and relates this impairment to abstract mood states (manic and depressive episodes) and oscillations of bipolar disorder.

Neurovascular coupling is typically considered as a master-slave relationship between the neurons and the cerebral vessels: the neurons demand energy which the vessels supply in the form of glucose and oxygen. In the recent past, both theoretical and experimental studies have suggested that the neurovascular coupling is a bidirectional system, a loop that includes a feedback signal from the vessels influencing neural firing and plasticity. An integrated model of bidirectionally connected neural network and the vascular network is hence required to understand the relationship between the informational and metabolic aspects of neural dynamics. In this study, we present a computational model of the bidirectional neurovascular system in the whisker barrel cortex and study the effect of such coupling on neural activity and plasticity as manifest in the whisker barrel map formation. In this model, a biologically plausible self-organizing network model of rate coded, dynamic neurons is nourished by a network of vessels modeled using the biophysical properties of blood vessels. The neural layer which is designed to simulate the whisker barrel cortex of rat transmits vasodilatory signals to the vessels. The feedback from the vessels is in the form of available oxygen for oxidative metabolism whose end result is the adenosine triphosphate (ATP) necessary to fuel neural firing. The model captures the effect of the feedback from the vascular network on the neuronal map formation in the whisker barrel model under normal and pathological (Hypoxia and Hypoxia-Ischemia) conditions.

A special class of neurons in the hippocampal formation 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 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 the 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 the emergence of cells with grid-like spatial periodicity. We show that the Bessel functions could describe the response of these cells. The output of the PI layer is used to train a stack of autoencoders. Neurons of both the layers exhibit responses resembling grid cells and place cells. The paper concludes by suggesting the wider applicability of the proposed modeling framework beyond the two simulated studies.

Artificial feedforward neural networks perform a wide variety of classification and function approximation tasks with high accuracy. Unlike their artificial counterparts, biological neural networks require a supply of adequate energy delivered to single neurons by a network of cerebral microvessels. Since energy is a limited resource, a natural question is whether the cerebrovascular network is capable of ensuring maximum performance of the neural network while consuming minimum energy? Should the cerebrovascular network also be trained, along with the neural network, to achieve such an optimum? In order to answer the above questions in a simplified modeling setting, we constructed an Artificial Neurovascular Network (ANVN) comprising a multilayered perceptron (MLP) connected to a vascular tree structure. The root node of the vascular tree structure is connected to an energy source, and the terminal nodes of the vascular tree supply energy to the hidden neurons of the MLP. The energy delivered by the terminal vascular nodes to the hidden neurons determines the biases of the hidden neurons. The “weights” on the branches of the vascular tree depict the energy distribution from the parent node to the child nodes. The vascular weights are updated by a kind of “backpropagation” of the energy demand error generated by the hidden neurons. We observed that higher performance was achieved at lower energy levels when the vascular network was also trained along with the neural network. This indicates that the vascular network needs to be trained to ensure efficient neural performance. We observed that below a certain network size, the energetic dynamics of the network in the per capita energy consumption vs. classification accuracy space approaches a fixed-point attractor for various initial conditions. Once the number of hidden neurons increases beyond a threshold, the fixed point appears to vanish, giving place to a line of attractors. The model also showed that when there is a limited resource, the energy consumption of neurons is strongly correlated to their individual contribution to the network’s performance.