Hybrid neural network

About: Hybrid neural network is a research topic. Over the lifetime, 1305 publications have been published within this topic receiving 18223 citations.

Sensitivity analysis of a hybrid neural network for locomotor control in the lamprey

Abstract: Neural circuitry within the spinal cord of the lamprey forms a central pattern generator (CPG) for locomotor control, which interacts with particular neurons in a reticulo-spino-reticular (RSR) loop. Some of the CPG neurons have pacemaker properties. We have used mathematical models to examine the role of N-methyl-D-aspartate (NMDA) and calcium dependent potassium (K/sub Ca/ channel currents in generating oscillatory outputs in: (1) a single neuron; (2) a hybrid neural network for the CPG with embedded pacemaker neurons; and (3) a reticular CPG network. Neurons were modeled as isopotential single compartment units with simplified biophysical properties. All three conditions showed a minimum and maximum NMDA and K/sub Ca/, conductance values for which oscillations occurred. The pacemaker neurons played an important role in the control of the locomotor rhythm by affecting the burst generation and termination mechanisms. The range over which oscillations can be obtained by altering the pacemaker properties of the neurons is affected by the architecture of the network. The pacemaker properties can provide a mechanism for frequency control by the brain with or without affecting the drive to the motor output units.

Flow prediction in vegetative channel using hybrid artificial neural network approach

Abstract: Channel flow–vegetation interaction has been extensively studied in the past few decades and many equations have been developed which essentially differ from each other in derivation and form. As the process is extremely complex, getting deterministic or analytical forms of process phenomena are too difficult. A hybrid neural network model (combining genetic algorithm with neural network), which is particularly useful in modeling processes about which adequate knowledge of the physics is limited, is presented here as a complementary tool to model channel flow–vegetation interactions in submerged vegetation conditions. The prediction capability of the model has been found to be satisfactory. The input significance of the different parameters has been analyzed in the present work in order to find out the influence of these parameters on channel flow velocity.

Deep Neural Network-Embedded Internet of Social Computing Things for Sustainability Prediction

Abstract: Social computing, exploiting utilization of advanced computational techniques to overcome typical problems in social science, has been a more visualized conception in academia. However, existing researches still suffer from two aspects of challenges: 1) lack of reliable multi-source data acquisition and management; 2) absence of high-performance algorithmic approaches. Fortunately, some newly-emerged cross-discipline technologies offer more opportunities to enhance conventional solutions. For the former, characterized by its property of information collection and integration, Internet of Things (IoT) can be introduced to produce a novel architecture named Internet of Social Computing Things (IoSCT). For the latter, specific neural network models can be set up to manipulate complicated calculation. Thus, taking the issue of sustainability prediction as objective situation, deep neural network-embedded Internet of Social Computing Things (NeSoc) is proposed in this paper. Firstly, IoSCT is put forward as bottom support platform, guaranteeing comprehensive resource involvement of social computing. Secondly, a hybrid neural network mechanism is formulated and embedded into IoSCT for centralized modeling. Finally, a series of experiments are conducted on a real-world dataset to evaluate performance of the proposed NeSoc.

Inference with Hybrid Bio-hardware Neural Networks

Abstract: To understand the learning process in brains, biologically plausible algorithms have been explored by modeling the detailed neuron properties and dynamics. On the other hand, simplified multi-layer models of neural networks have shown great success on computational tasks such as image classification and speech recognition. However, the computational models that can achieve good accuracy for these learning applications are very different from the bio-plausible models. This paper studies whether a bio-plausible model of a in vitro living neural network can be used to perform machine learning tasks and achieve good inference accuracy. A novel two-layer bio-hardware hybrid neural network is proposed. The biological layer faithfully models variations of synapses, neurons, and network sparsity in in vitro living neural networks. The hardware layer is a computational fully-connected layer that tunes parameters to optimize for accuracy. Several techniques are proposed to improve the inference accuracy of the proposed hybrid neural network. For instance, an adaptive pre-processing technique helps the proposed neural network to achieve good learning accuracy for different living neural network sparsity. The proposed hybrid neural network with realistic neuron parameters and variations achieves a 98.3% testing accuracy for the handwritten digit recognition task on the full MNIST dataset.

An ARTMAP based hybrid neural network for shift invariant Chinese character recognition

Abstract: An ARTMAP-based hybrid neural network is proposed to recognize position-shifted Chinese characters. The faster learning speed of a hybrid architectures makes practical the use of neural networks in large-scale neural computation. Four translation-invariant transformations are used to extract features of two-dimensional patterns. The results of experimentation with three different hybrid neural networks are presented. >