Find research datasets worth reusing

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 5. The periphery of circular area around of artificial creature in each position

It is assumed in the simulation that the artificial creature in each location stands in the center of a circular area with a radius of its vision range, i.e. 2 meters. If the creature turns around itself; it only can see objects in its visual range. When a food object appears in the periphery of this circular area, a creature should find this food. Moving around, another food appears in a random place on periphery of the circular area, and this procedure repeat for 5 iterations. As shown in Figure 5 the periphery of the circular area around the artificial creature in each position is divided to 48 slots, which can be considered as a rectangular. In the simulation program it is implemented as an array of 48 × 3.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 2. Example of different axonal conduction delay between presynaptic neuron A0 and postsynaptic neuron A1 and A2

Different axonal conduction delays between every two neurons are applied as follows: For instance, consider presynaptic neuron A0 and postsynaptic neurons A1 and A2 in Figure 2. If neuron A0 fires spike S0 in t = 9 and S1 in t = 11 and with a time step equal to 0.5 ms, S0 arrives to A1 in t = 13 and to A2 in t = 14 where S1 arrives to A1 in t = 15 and to A2 in t = 16. This is implemented using counters assigned to each spike which increases with each time step in the time window. If counters arrive to axonal delay values, the effect of spikes from presynaptic neuron applies to the post synaptic neurons.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 8. Flowchart for movements of creature

Window time has a fixed length between and and is an appropriate time interval [5]. So in each time window the total number of spikes in each three neurons is compared with other three neurons and artificial creature moves toward direction that the respective neurons fired maximum number of spikes. These fixed time windows consist of 600 time-steps. Each time step is 0.5 ms. Flowchart in Figure 8 shows details.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 1. Typical neural networks SYNAPTIC

A reservoir network has been used in this paper. This structure has been shown in Figure 1. As is observable this network has two input and output layers. The neuronal network that have been used is composed of N = 150 randomly connected Izhikevich spiking neurons and different axonal conduction delays between each two neurons. Information is transferred between neurons of the networks through the links between every two neurons representing synapses. Each neuron is connected to M = 15 random neurons, so that the probability of connection is M / N = 0.15. It is noticeable that in this network, not only the connection between two neurons is random, but also the neurons type selection is random, too.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 13. Average fitness at generation progressing

Figure 13 compares the average fitness at generations in one typical program running. As can be seen from this figure, the average of fitness grows as the number of generations increases.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 7. Effect of different type of movement on the image in the vision of artificial creature.

Figure 7a, Figure 7b, Figure 7c and Figure 7d, shows effect of different type of movement on the image in the vision of artificial creature if food be on vision boundaries. As mentioned each part of the image equal 7.5 degree. Therefore 15 degree left or right rotation locomotion equivalent two parts shift toward left or right. For motion to forward direction, size of the image has been reduplicated so that each part has been become to the two similar parts. Then half of new image in right side and left side has been deleted in order to create new close image in vision. Accordingly, if food be on vision boundaries, the number of black parts of the image for the food object in ultimate location is 2, by one movement to forward direction the number of these parts become to 4, by one movement to forward direction the number of these parts become to 8 and so on. After four movements to forward all part of the image is black and the creature is succeed find the food object.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 6, One image in sight of artificial creature.

Figure 6 shows an image in sight of the creature, where the food in this image is the dark part of the vision. If the food object places in the vision edge of the artificial creature (2 meters), two squares in the image in sight of the creature becomes black and by getting the artificial creature closer to the food, more squares of image in sight become block and if all parts of the image in sight of the creature become black, the artificial creature has been successful in finding the food object. In Figure 6 each part of the image divided to 3 subparts because 3 neurons per part of image in the input layer of the artificial creature network (vision) have been considered. So vision of the artificial creature composed of 60 neurons due to: For each black part of image in sight of the artificial creature, three signals as the input signals is applied to the three respective neurons.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 9. A typical artificial life form and circular area around of it

As mentioned the reservoir neural network has been chosen for neural network of the artificial creature as main body structure. Figure 9 illustrates a typical artificial life form and circular area around of it. Maximum seeing of the artificial life form is periphery of the circular area and can't see places that have beyond of periphery of the circular area. The artificial creature by each movement is the center of a circular area so if the creature move to forward direct can see new places and some places are voided of seeing.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 12. Genetic Algorithm flowchart

Then next generation are produced by combination of the elites (15%), crossover (55%) and mutation (30%) of the initial population. Elites are the best chromosomes which are directly transferred to the next generation. Because of long chromosome length, for crossover, five points are randomly chosen in each parent as cut points. Figure 11 shows a typical crossover with two cutpoints and Figure 12 illustrates a flowchart for the proposed evolutionary model. Selections are based on Roulette Wheel selection, more detailed information can be found in.

BRAIN Journal-Artificial Neuron Modelling Based on Wave Shape-Figure 1. Typical feedforward artificial neural network processing unit

In Figure 1, each input X is weighted by a separate weight value w. These are then summed together to give a total input value for the neuron. This total can then be passed through a function, to transform it into the desired output value. This is then compared to the actual output value d; where the error or differences between the two sets is measured and used to correct the weight values, to bring the two sets of values closer. One way to update the weights is after each individual pattern is presented and processed. Another option is to process the error after the presentation of the whole dataset, as part of a batch update.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 4: Pseudo code for applying different axonal conduction delay between presynaptic neurons and postsynaptic neurons

In each time step, axonal conduction delays between presynaptic neurons and postsynaptic neurons are examined whether they are equal to the elements of array I_S, to apply the respective spikes. The pseudo code for applying different axonal conduction delay between presynaptic neurons and postsynaptic neurons are shown in Figure 4.

BRAIN Journal-Evolving Spiking Neural Networks for Control of Artificial Creatures-Figure 15. Movements a typical successful creature to find food

Figure 15 (a, b, c, d, e, f, g, h) illustrates movements of a typical successful creature for finding one food object. We believe that this study can be a step forward in understanding the morphology of artificial creatures. Also this paper suggests more complex artificial life examination by adding different part to these networks similar to different segments of the brain such as: vision, locomotion, hippocampus and communication in a future work. Because of the different axonal conduction delay between every two neurons in the neural network of artificial creatures in this paper, our next study is to enhance the artificial lives by STDP learning.

opencc-by-4.0Feb 2022View details →

Real-Time Adaptation of an Artificial Neural Network for Transfemoral Amputees Using a Powered Prosthesis

This dataset contains the data used in our manuscript titled "Real-Time Adaptation of an Artificial Neural Network for Transfemoral Amputees Using a Powered Prosthesis". Data structure is explained in the README.txt file located at the top-level of the dataset. Please contact corresponding author Richard B. Woodward for any questions.

BRAIN Journal-Brain-Like Artificial Intelligence for Automation-Figure 12. Affinities and Differences of Neuro-Symbolic Networks in Comparison to Classical Neural Networks

After having briefly illustrated the basic function principle of neuro-symbolic networks, this section aims at reviewing their affinities and differences to standard neural networks like for example multi-layer perceptrons (MLPs) [58]. A summary of these affinities and differences is given in Figure 12. The affinities concern certain functions of individual nodes of the networks. In both cases, weighted input information is summed up and an activation function is applied to this sum. In both cases, the individual nodes are interconnected to form networks. Much larger than the number of affinities between neuro-symbolic networks and neural network is however the number of differences. The first difference consists in the application domain. Neuro-symbolic networks have so far mainly been applied for complex, large-scale sensor data processing of multimodal data – an application which can so far barely be handled by neural networks.

Figure 5. Sensory score and period of storage for processed cheese-Time-Delay Artificial Neural Network Computing Models for Predicting Shelf Life of Processed Cheese

R2 was found to be 96.5 percent of the total variation as explained by sensory scores. Period of storage (days) for which the processed cheese has been in the shelf can be determined based on sensory score (Fig. 5).

Figure 4. Comparison of ASS and PSS for multilayer model R-Time-Delay Artificial Neural Network Computing Models for Predicting Shelf Life of Processed Cheese

TDNN models with single and multi layers were developed taking soluble nitrogen, pH, standard plate count, yeast & mould count, spore count as input parameters, and sensory score as output parameter for predicting the shelf life of processed cheese stored at 30o C. Mean Square Error, Root Mean Square Error, Coefficient of Determination and Nash - Sutcliffo Coefficient were used in order to compare the prediction ability of the developed TDNN models. Regression equations were developed for predicting the shelf life of processed cheese, which came out as 28.25 days. Since, predicted value is close to the experimentally determined shelf life of 30 days, hence from the study it can be concluded that TDNN artificial neural network models are quite efficient in predicting shelf life of processed cheese.

Figure 2. Training pattern of TDNN models-Time-Delay Artificial Neural Network Computing Models for Predicting Shelf Life of Processed Cheese

The Neural Network Toolbox under MATLAB software was used for developing the TDNN models. Training pattern of TDNN models is presented in Fig.2.

Figure 1. Inputs and output parameters for TDNN models-Time-Delay Artificial Neural Network Computing Models for Predicting Shelf Life of Processed Cheese

The data consisted of 36 samples, which were divided into two subsets, i.e., 30 used for training the network and 6 for testing the TDNN models. Soluble nitrogen, pH, standard plate count, yeast & mould count, and spore count were taken as input parameters, and sensory score as output parameter for developing TDNN single and multilayer models (Fig.1).

Figure 6. Performance plot for NASDAQ index (RNN)-Comparative study of Financial Time Series Prediction by Artificial Neural Network with Gradient Descent Learning

Figure 5 we can see that the mse curve reaches the In performance goal but it does not decrease in that good manner,but in Figure 6 the mse is reduces widely. By analyzing all these results one can say that RNN is better choice than Feedforward MLP in prediction purpose.

Figure 3. Comparison of ASS and PSS single layer model-Time-Delay Artificial Neural Network Computing Models for Predicting Shelf Life of Processed Cheese