SOFTWARE SIMULATORS AND DATA
On this page, you may find some software simulators that utilize some solutions in complex-valued neural networks with multi-valued neurons.
You may download them and use in your research and for solving real-world problems.
There are also data sets, which were used in recently published papers, available for downloading.
SOFTWARE SIMULATORS
All software simulators available here are free only for non-commercial use (research and teaching). They must not be distributed without prior permission of the authors. The authors are not responsible for implications from the use of this software.
· Multilayer Neural Network based on Multi-Valued Neurons with batch LLS-based learning algorithm and soft margins (MLMVN-LLS-SM) - MATLAB Simulators. MLMVN is a powerful machine learning tool, which significantly outperforms many other machine learning techniques in terms of learning speed and generalization capability. It is presented in detail in [1], [2]. Its further development utilizes Soft Margins learning technique (MLMVN-SM) and it is presented in detail in [3], [4]. A batch learning algorithm for MLMVN makes it possible to adjust weights for the entire learning set instead of adjusting them for every single learning sample separately. A batch learning algorithm, which is presented in detail in [5] is much more efficient that a classical (serial) learning algorithm, which is utilized in software simulators presented on this page below.
There are two sets of Matlab functions designed without using any special toolbox. Both are suitable for solving classification problems.
One of them utilizes the MLMVN-SM n-M-1 (a network with n inputs, M neurons in a single hidden layer and a single output neuron with the discrete activation function). This set of functions and a dataset with a readme file and demos, which make it possible to repeat a corresponding simulating experiment from [5] is available in a zip archive HERE.
Another one utilizes the MLMVN-SM n-M-k (a network with n inputs, M neurons in a single hidden layer and k output neurons with the discrete activation function). This network is an efficient tool for solving multi-class classification problems. This set of functions and a dataset with a readme file and demos, which make it possible to repeat a corresponding simulating experiment from [5] is available in the zip archive HERE.
When using this software, please, refer to this webpage (https://www.igoraizenberg.com/download-software-simulators-and-data) and publications [1], [3], [5] in your corresponding publications and presentations.
· Convolutional Neural Network with Multi-Valued Neurons (CNNMVN) – Matlab Simulator. CNNMVN is a new convolutional neural network employing multi-valued neurons and their paradigm. CNNMVN and its learning algorithm are presented in detail in [6]. A set of Matlab functions utilizing CNNMVN with a single convolutional layer, a pooling layer, a single fully connected layer and an output layer is available in a zip archive HERE.
Datasets used in [6] for simulations are also included.
When using this software, please, refer to this webpage (http://www.freewebs.com/igora/Downloads.htm) and publication [6] in your corresponding publications and presentations.
· Multilayer Neural Network based on Multi-Valued Neurons (MLMVN) - MATLAB Simulator. MLMVN is a powerful machine learning tool, which significantly outperforms many other machine learning techniques in terms of learning speed and generalization capability. It is presented in detail in [1], [2]. Its further development utilizes Soft Margins learning technique (MLMVN-SM) and it is presented in detail in [3], [4]. Soft margins are used there to get a higher accuracy when solving classification problems. This simulator supports both standard MLMVN and MLMVN-SM as well as it can be used for testing. There is Matlab function designed without using any special toolbox. This software simulates a network with either continuous or discrete inputs and with either continuous or discrete output (outputs). Thus, it can be used for solving both classification and regression problems. It is important to say that it is especially efficient in solving multi-class classification problems where the number of classes is larger than two. A zip archive (67 Kb) with the function, a quasi-manual and some supportive programs and materials is available HERE.
When using this software, please, refer to this webpage (http://www.freewebs.com/igora/Downloads.htm) and publications [1], [3] in your corresponding publications and presentations.
· Multilayer Neural Network based on Multi-Valued Neurons (MLMVN) - Executable Files. This network, which significantly outperforms many other machine learning techniques in terms of learning speed and generalization capability, is presented in detail in [1], [2] There are two exe-files for Windows 8/7/Vista/XP operating systems, a Matlab-file for converting continuous data in the format appropriate for MLMVN, its software simulator and a brief user's manual that are winzipped in a single archive. One of the exe-files should be used for learning and another one for testing. This software simulates a network with either continuous or discrete inputs and with a discrete output (outputs). Thus, it can be used for solving classifications problems. It is important to say that it is especially efficient in solving multi-class classification problems where the number of classes is larger than two. A zip archive is available HERE. Many thanks to Dr. Dmitriy Paliy for his important contribution to the software design.
When using this software, please, refer to this webpage (https://www.igoraizenberg.com/download-software-simulators-and-data) and publications [1], [2] in your corresponding publications and presentations.
· Multilayer Neural Network based on Multi-Valued Neurons with Soft Margins Learning (MLMVN-SM) - Executable Files. MLMVN with Soft Margins learning (MLMVN-SM) is a further development of MLMVN. MLMVN-SM is presented in detail in [3], [4]. Soft margins are used there to get a higher accuracy when solving classification problems. There are two exe-files for Windows 8/7/Vista/XP operating systems, a Matlab-file for converting continuous data in the format appropriate for MLMVN, MLMVN-SM software simulator and a brief user's manual that are winzipped in a single archive. One of the exe-files should be used for learning and another one for testing. This software simulates a network with either continuous or discrete inputs and with a discrete output (outputs). Thus, it can be used for solving classifications problems. It is important to say that it is especially efficient in solving multi-class classification problems where the number of classes is larger than two. A zip archive (1.5 Mb) is available HERE. Many thanks to Dr. Dmitriy Paliy for his important contribution to the software design.
When using this software, please, refer to this webpage https://www.igoraizenberg.com/download-software-simulators-and-data) and publications [3], [4] in your corresponding publications and presentations.
DATA SETS
· The data sets used and weights resulted from the experiments on Multilayer Neural Network based on Multi-Valued Neurons with Soft Margins Learning (MLMVN-SM) in the papers [3], [4]. These data sets and weights sets are formatted for their use with MLMVN and MLMVN-SM software simulators available for downloading at this page (see above). A zip archive (5 Mb) is available HERE.
If you have any difficulty in using that software, which you have downloaded from this page, feel free to contact me.