15 December 2019 in Discrete and Continuous Models and Applied Computational Science
The article is an overview. We carry out the comparison of actual machine learning libraries that can be used the neural networks development. The first part of the article gives a brief description of TensorFlow, PyTorch, Theano, Keras, SciKit Learn libraries, SciPy library stack. An overview of the scope of these libraries and the main technical characteristics, such as performance, supported programming languages, the current state of development is given. In the second part of the article, a comparison of five libraries is carried out on the example of a multilayer perceptron, which is applied to the problem of handwritten digits recognizing. This problem is well known and well suited for testing different types of neural networks. The study time is compared depending on the number of epochs and the accuracy of the classifier. The results of the comparison are presented in the form of graphs of training time and accuracy depending on the number of epochs and in tabular form.
1 March 2018 in Programming and Computer Software
When modeling such phenomena as population dynamics, controllable flows, etc., a problem arises of adapting the existing models to a phenomenon under study. For this purpose, we propose to derive new...
1 April 2019 in Journal of Physics: Conference Series
In this article, we compare the support for SIMD instructions for Julia and Fortran. The comparison is carried out according to the methodology described in work of T. Kalibera, R. E. Jones. The first part of the article gives a brief description of this technique. We emphasize on the practical implementation using Python, NumPy and SciPy. The second part of the article briefly discusses the syntactic capabilities of Fortran and Julia to work with SIMD processor extensions. Specific code snippets are given. Next, the performance of Julia and Fortran is compared for arithmetic operations on arrays of small length. The results are presented in tabular and graphical form.
5 February 2022 in Programming and Computer Software
Geometric algebra is based on the works by Grassmann and Clifford. Its main objects of research are polyvectors (p-vectors) and multivectors. Polyvectors, together with the exterior product, implement the Grassmann algebra, while multivectors with the geometric product implement the Clifford algebra. Multivector algebra generalizes many operations and objects of analytic geometry and differential geometry (e.g., vector and mixed products, normal vectors and binormals, etc.) to the multidimensional case, as well as provides their geometric interpretation. Complex numbers and quaternions are isomorphic to multivectors of a special kind. This paper applies certain ideas of geometric algebra to solve problems that occur in computer geometry. For this purpose, the Grassmann.jl package for Julia is used.
1 January 2023 in Computational Mathematics and Mathematical Physics
Tensor formalism (and its special case—vector formalism) is a mathematical technique that is widely used in physical and engineering problems. Even though this formalism is fairy universal and suitable for describing many spaces, the application of other special mathematical techniques is sometimes required. For example, the problem of rotation in a 3D space is not very well described in tensor representation, and it is reasonable to use the formalism of Clifford algebra, in particular, quaternions and geometric algebra representations for its solution. In this paper, computer algebra is used to demonstrate the solution of the problem of rotation in a 3D space using both the quaternion and geometric algebra formalisms. It is shown that although these formalisms are fundamentally similar, the latter one seems to be clearer both for computations and interpretation of results.