China
Africa
Explore chapters and articles related to this topic
Processing Techniques and Analysis of Brain Sensor Data Using Electroencephalography
Published in Mridu Sahu, G. R. Sinha, Brain and Behavior Computing, 2021
What makes us humans is the essence of emotions that are characterized as intense mental activities and feelings. The emotions range between arousal and valence such as confrontation of fearful or life-threatening events for example stimulus arised from high sexual appeal increases the levels of arousal or valance [1]. Human behavior is comprised of three main components: emotions, actions, and cognitions. The interaction between them enables us to perceive, respond, and have relations with other people in our surroundings. The emotions heavily affect human behavior, intelligence, and decision making; therefore, they need to be captured either using qualitative or quantitative analysis [2]. The information processing of brain functions with the advancement in artificial intelligence (AI) and signal-processing techniques open up a plethora of opportunities in the field of computational neuroscience for making applications that can make decisions autonomously and accurately.
Computational Neuroscience and Compartmental Modeling
Published in Bahman Zohuri, Patrick J. McDaniel, Electrical Brain Stimulation for the Treatment of Neurological Disorders, 2019
Bahman Zohuri, Patrick J. McDaniel
Computational neuroscience describes the nervous system through computational models. Although this research program is grounded in mathematical modeling of individual neurons, the distinctive focus of computational neuroscience is systems of interconnected neurons. Computational neuroscience usually models these systems as neural networks. In that sense, it is a variant, offshoot, or descendant of connectionism. However, most computational neuroscientists do not self-identify as connectionists. There are several differences between connectionism and computational neuroscience: Neural networks employed by computational neuroscientists are much more biologically realistic than those employed by connectionists are. The computational neuroscience literature is filled with talk about firing rates, action potentials, tuning curves, etc. These notions play at best a limited role in connectionist research, such as most of the research canvassed in Rogers and McClelland.70Computational neuroscience is driven in large measure by knowledge about the brain, and it assigns a huge importance to neurophysiological data (e.g., cell recordings). Connectionists place much less emphasis upon such data. Their research is primarily driven by behavioral data (although more recent connectionist writings cite neurophysiological data with somewhat greater frequency).Computational neuroscientists usually regard individual nodes in neural networks as idealized descriptions of actual neurons. Connectionists usually instead regard nodes as neuron-like processing units,70 while remaining neutral about how exactly these units map onto actual neurophysiological entities.
Active exploration based on information gain by particle filter for efficient spatial concept formation
Published in Advanced Robotics, 2023
Akira Taniguchi, Yoshiki Tabuchi, Tomochika Ishikawa, Lotfi El Hafi, Yoshinobu Hagiwara, Tadahiro Taniguchi
Our challenge is task-independent and interactive knowledge acquisition, in which the robot asks the user questions. At the interface between artificial intelligence and computational neuroscience, free-energy principle (FEP)-based active inference (AIF) [5] has gained attention as an approach through which agents actively explore and acquire knowledge. The expected free energy based on AIF theoretically encompasses information gain (IG), which is commonly used in traditional active perception/learning in robotics, and provides further inspiration. The application of AIF theories in robotics has gained increasing importance [6,7]. AIF encompasses active exploration and online learning loops, which serve as the foundation for our study. In the field of robotics, studies have been conducted on active exploration for simultaneous localization and mapping (SLAM) [8], that is, active SLAM [9–11] and active perception/learning for multimodal categorization [12,13]. Our study integrated these approaches, leading to active semantic mapping [14,15]. This study differs from vision-and-language navigation (VLN) [16–18], which uses task-dependent knowledge acquisition without AIF as its theoretical foundation. We focus on active exploration inspired by AIF for grounding a spatial lexicon in a mobile robot.
Almost second-order uniformly convergent numerical method for singularly perturbed convection–diffusion–reaction equations with delay
Published in Applicable Analysis, 2023
Mesfin Mekuria Woldaregay, Gemechis File Duressa
Singularly perturbed delay differential equations (SPDDEs) are differential equations in which their highest order derivative term is multiplied by the small perturbation parameter ε and involving at least one delay term. Singularly perturbed differential equations relate unknown functions to its derivatives evaluated at the same time SPDDEs model processes for which the evaluation not only depends on the current state of the system but also includes the past history. SPDDEs have an application in the modelling of neuronal variability in computational neuroscience [10], in the study of variational problems in control theory [11] and so on. In general, when the perturbation parameter, ε tends to zero, the smoothness of the solution of the SPDDEs deteriorates and it forms a boundary layer [12]. It is well known that standard numerical methods such as finite difference method, finite element method, finite volume method and collocation method are ineffective for solving singularly perturbed problems when the perturbation parameter approaches zero [13]. In the last decade, different authors have developed numerical schemes for treating singularly perturbed parabolic differential equations with deviating arguments on the reaction term(s) [12,14–25]. The most commonly used techniques are the non-standard finite difference method on uniform mesh, the exponentially fitted operator method on uniform mesh and standard finite difference methods on fitted mesh.
Modelling and analysis of multiscale systems related to fluid dynamical problems
Published in Mathematical and Computer Modelling of Dynamical Systems, 2018
These authors discussed neuronal models based on the Hodgkin–Huxley equation in the field of computational neuroscience. The multiscale model is based on the extension of classical deterministic models to stochastic models, including molecular noise from the underlying microphysical conditions. These models take into account the firing of individual neurons related to electric fields in extracellular space and include the affect of the firing pattern of nearby neurons. These authors developed a multiscale model that combines stochastic ion channels with the propagation of an action potential along the neuronal structure. They presented a novel three-stage multiscale modelling framework: (a) on the microscale, where the ion channels are modelled by a continuous-time discrete-state Markov chain; (b) on the intermediate scale, where the current-balance and the cable equation are modelled by an ODE; and (c) on the macroscale, where the propagation of the trans-membrane current into an electrical field in extracellular space is modelled by PDEs.