The Optimal Choice of Hybrid Convolutional Neural Network Components
Viktor Sineglazov,
Illia Boryndo
Issue:
Volume 8, Issue 2, December 2022
Pages:
12-16
Received:
4 July 2022
Accepted:
25 July 2022
Published:
4 August 2022
DOI:
10.11648/j.ajnna.20220802.11
Downloads:
Views:
Abstract: Based on the development vector of modern AI systems and extensional complexity grows of analytical and recognition tasks it is concluded that the perspective class of convolutional neural networks – hybrid convolutional neural networks. Based on research results it’s proved that this type of neural networks permits to supply less mean square error under less overall structure complexity. The generic structure of hybrid convolutional neural network was proposed. It is shown and proved that these networks must include beside traditional components (convolutional layers, pooling layers, feed-forward layers) as well as an additional supportive layers (batch normalization layer, 1x1 convolutional layer, dropout layer, etc.) to achieve best both accuracy and performance results. The important properties of additional supportive layers (blocks) have been determined and researched. Based on the architectural requirements it is considered that modern topologies of hybrid CNNs are the combination of substantive CNNs such as Squeeze-and-Excitation neural network, poly-inception neural network, residual neural network, densely connected neural network, etc. It is listed the performance testing and final accuracy results for each block used both separately and in pairs to highlight it inner parameters, advantages and limitations. It is proposed an example of hybrid convolutional neural network constructed of investigated structural blocks. Calculated average training time shorten based on each of functional blocks, their advantages and integration details.
Abstract: Based on the development vector of modern AI systems and extensional complexity grows of analytical and recognition tasks it is concluded that the perspective class of convolutional neural networks – hybrid convolutional neural networks. Based on research results it’s proved that this type of neural networks permits to supply less mean square error...
Show More
From Eating Behavior to Dialogue Using Language; Evolution of Neural Network
Issue:
Volume 8, Issue 2, December 2022
Pages:
17-23
Received:
5 November 2022
Accepted:
28 November 2022
Published:
8 December 2022
DOI:
10.11648/j.ajnna.20220802.12
Downloads:
Views:
Abstract: In this paper, logic is developed based on the view that the brain nerve circuit is composed of a combination of neural circuits with the same function. The subject of logic is animal's action; from early evolved animals that only perform feeding to more evolved animals that have ability to act as groups. In Chapter 2, reconstructed and outlined the previously published papers described about Basic Unit. Using category theory, Basic Unit is defined as an object in neural network (defined as category). By setting functions between categories, behavior of multiple categories express imitation behavior not only eating behavior. These functions enable collective actions and are the basis of individual communication. Chapter 3 describes the essential functions which human communication make superiority to non-human communication. First, a new neural network is placed on the top level of the existing neural network. The new neural network operates asynchronously with the existing neural network directly involved with the senses and motile organ. Next, a process to share events that are recognized by new neural networks among companions is presented. In other words, dialogue deploys individual knowledge to group knowledge. It is clear that the spreading of knowledge from individuals to groups is one of the most value of language. On dialogue session, there is no guarantee that the listener understands the content with just one explanation of the speaker. Speakers guess the understanding level of the listener from listener's actions and expects the following content. The contents of next dialog are opposition, misunderstandings, corrections, supplements, and etc. Ordinary, after resulting trial and error, both speakers become satisfied situation. But sometimes the dialogue ends without satisfied situation. These dialog processes are represented as changes in dialog content according to the internal state of high -level neural networks related to the reception of time series data.
Abstract: In this paper, logic is developed based on the view that the brain nerve circuit is composed of a combination of neural circuits with the same function. The subject of logic is animal's action; from early evolved animals that only perform feeding to more evolved animals that have ability to act as groups. In Chapter 2, reconstructed and outlined th...
Show More
