| viXra.org > Artificial Intelligence > 2105 |
 |
 |
| viXra.org > Artificial Intelligence > 2105 |
|
|
[6] viXra:2105.0176 [pdf] submitted on 2021-05-31 12:17:35
Authors: Abdurrahim Yilmaz, Dilanur Bayraktar, Melih Akman, Cemre Sahinoglu, Huseyin Uvet
Comments: 3 Pages.
In this paper, a detailed study on gesture classifica- tion using a dataset from Kaggle and optimizing the dataset is presented. The machine learning algorithms, which are SGD, kNN, SVM, MLP, Gaussian Naive Bayes classifier, Random Forest, LightGBM, XGBoost, and CatBoost classifiers, to conduct the research and, are used. The results are compared with each other to conclude which models perform the best in gesture classification. Except for the Gaussian Naive Bayes classifier, all methods resulted in high accuracy.
Category: Artificial Intelligence
[5] viXra:2105.0141 [pdf] submitted on 2021-05-24 21:37:09
Authors: Ruolin Jiu
Comments: 16 Pages.
A completely new learning rule of neural networks. Similar to the learning rule in the brain, completely different with gradient descent.
This learning rule, is the foundation and the key of AI memory, will open a huge growth potential for Artificial Intelligence.
Category: Artificial Intelligence
[4] viXra:2105.0138 [pdf] submitted on 2021-05-23 07:45:16
Authors: Jan Helm
Comments: 45 Pages.
This paper presents
in Part1 the basic theory of Neural Networks, and based on the standard (global) backpropagation algorithm, it introduces the local backpropagation algorithm: a layer-recurrent gradient algorithm with layer-specific target-vector.
Furthermore in Part2 , it presents calculated application examples for global backpropagation networks, local backpropagation networks and evolving cross-mutated networks.
Category: Artificial Intelligence
[3] viXra:2105.0095 [pdf] submitted on 2021-05-17 12:53:33
Authors: J Gerard Wolff
Comments: 32 Pages.
This article is about the origin, development, and benefits of the "SP System" (SPS), which means the "SP Theory of Intelligence" and its realisation in the "SP Computer Model" (SPCM). The SPS is radically different from deep neural networks (DNNs), with many advantages compared with DNNs. As will be described, the SPS provides a promising foundation for the development of human-like broad AI. The SPS was inspired in part by: evidence for the importance of information compression in human learning, perception, and cognition; and the concept of `multiple sequence alignment' in biochemistry. That latter concept led to the development of the powerful concept of SP-multiple-alignment, a concept which is largely responsible for the intelligence-related versatility of the SPS. The main advantages of the SPS are: 1) The clear potential of the SPS to solve 19 problems in AI research; 2) Versatility of the SPS in aspects of intelligence, including unsupervised learning, and several forms of reasoning; 3) Versatility of the SPS in the representation and processing of knowledge; 4) Seamless integration of diverse aspects of intelligence and diverse forms of knowledge, in any combination, a kind of integration that appears to be necessary in any artificial system that aspires to the fluidity and adaptability of the human mind; 5) Several other potential benefits and applications of the SPS. It is envisaged that the SPCM will provide the basis for the development of a first version of the {\em SP Machine}, with high levels of parallel processing and a user-friendly user interface. All software in the SP Machine would be open-source so that clones of the SP Machine may be created anywhere by individuals or groups, to facilitate further research and development of the SP System.
Category: Artificial Intelligence
[2] viXra:2105.0084 [pdf] submitted on 2021-05-14 01:08:18
Authors: Milad Keramati
Comments: 5 Pages.
In a problem facing agent, a situation can be categorized as different patterns and action can be taken based on the available information (known as method) as oppose to a simple value. Doing so will decrease the variety of situations and actions and as a result simplify the problem. Simple patterns and methods are generated at first but by detecting important patterns and methods and creating similar patterns and methods, the agent will be able to better recognize the situation it's in and find better solutions for the patterns respectively and as a result systematically broaden its knowledge over time.
By memorizing feelings (or rewards) and action result (situation) in a pattern, it's possible to make a tree of possible outcomes of an action related to a pattern and choose an action of the pattern that profit us the most by predicting future feelings and calculating the value and we know accuracy of our prediction based on similarity (or consistency) and number of results (or confidence).
I've also given my opinion and defined some standards regarding artificial intelligence, reinforcement learning, and designing agent in this paper.
Category: Artificial Intelligence
[1] viXra:2105.0033 [pdf] submitted on 2021-05-07 10:36:30
Authors: Fuyuan Xiao
Comments: 5 Pages.
In this paper, CET is generalized to quantum framework of Hilbert space in an open world, called generalized quantum evidence theory (GQET). Differ with classical GET, interference effects are involved in GQET. Especially, when a GQBBA turns into a classical GBBA, interference effects disappear, so that GQB and GQP functions of GQET degenerate to classical GBel and GPl functions of classical GET, respectively.
Category: Artificial Intelligence
| Third-Party Links: |
|