[ad_1]
Artificial neural networks are machine learning systems composed of a large number of connected nodes called artificial neurons. Similar to the neurons of a biological brain, these artificial neurons are the main basic units used to perform neural calculations and solve problems. Advances in neurobiology have shown the important role played by dendritic cell structures in neural computation, leading to the development of artificial neural models based on these structures.
The recently developed approach neuron model (ALNM) is a unique neural model with a dynamic dendritic structure. The ALNM can use a neural pruning function to eliminate unnecessary dendrite branches and synapses during training in order to solve a specific problem. The resulting simplified model can then be implemented in the form of a hardware logical circuit.
However, the well-known backpropagation (BP) algorithm used to form the ALMN network actually restricted the computational capacity of the neural model. "The BP algorithm was sensitive to the initial values and could easily be trapped in local minima," says author Yuki Todo of the Faculty of Electrical and Computer Engineering at Kanazawa University. "So we evaluated the capabilities of several heuristic optimization methods for the formation of ALMN network."
After a series of experiments, the state of the art material search algorithm (SMS) was selected as the most appropriate training method for the ALMN network. Six benchmark classification problems were then used to evaluate the NLA's optimization performance when it had been trained using SMS as an algorithm of the ### system. 39; learning. The results showed that the SMS offered higher training performance than BP and other heuristic convergence speed algorithms.
"An ALNM and SMS-based classifier has also been compared to several other popular classification methods," says Associate Professor Todo, "and the statistical results have confirmed the superiority of this classifier over these reference problems."
During the training process, the ALNM simplified neuronal models by synaptic pruning and dendritic pruning. The simplified structures were then replaced using logic circuits. These circuits also provided satisfactory classification accuracy for each of the reference problems. The ease of hardware implementation of these logic circuits suggests that future research will see that the ALNM and SMS protocols will be used to solve increasingly complex and large-scale problems in the real world.
Source link
