Relieving the Incompatibility of Network Representation and Classification for Long-Tailed Data Distribution

In the real-world scenario, data often have a long-tailed distribution and training deep neural networks on such an imbalanced dataset has become a great challenge. The main problem caused by a long-tailed data distribution is that common classes will dominate the training results and achieve a very low accuracy on the rare classes. Recent work focuses on improving the network representation ability to overcome the long-tailed problem, while it always ignores adapting the network classifier to a long-tailed case, which will cause the “incompatibility” problem of network representation and network classifier. In this paper, we use knowledge distillation to solve the long-tailed data distribution problem and fully optimize the network representation and classifier simultaneously. We propose multiexperts knowledge distillation with class-balanced sampling to jointly learn high-quality network representation and classifier. Also, a channel activation-based knowledge distillation method is also proposed to improve the performance further. State-of-the-art performance on several large-scale long-tailed classification datasets shows the superior generalization of our method.

The list algorithms accuracy characteristics improvement based on Krohn’s algorithm and its modifications

The research of algorithms for uniform loading of devices for homogeneous information processing systems is a very important science-intensive task. An experimental approach was chosen for the research. This is primarily due to the fact that the analytical solution of the distribution problem gives solutions that are far from reality, since it is unable to take into account many factors that affect the computing machine during its operation. The aim of this research is to improve the accuracy characteristics of the list algorithms through the use of heuristic algorithms, such as Krohn’s algorithm and its modifications. This made it possible to obtain a more even distribution of tasks among executive devices, which can be networked workstations, processors or processor cores. The work uses list algorithms, such as the Critical Path algorithm and Pashkeev’s algorithm, as well as heuristic algorithms - Krohn’s algorithm and its modifications. The main idea of the research is to obtain the best suboptimal solution by improving the quality of the resulting distribution. In this case, with the help of the list algorithms, the initial distribution is formed, and its refinement is carried out through the application of the Krohn’s algorithm and its modifications. In fact, in the work, a number of symbiotic algorithms are examined and analyzed. For this, many computational experiments were carried out and a large amount of output data were collected, on the basis of which conclusions were drawn about the effectiveness of the solution obtained for each symbiotic group and for all groups as a whole.

Using the Hopfield Neural Network to Select a Behaviour Strategy for the Group of Mobile Robots

The use of the Hopfield neural network for the task distribution problem solving in teams of mobile robots performing monosyllabic operations in a single workspace is considered. The study is a continuation of earlier works in which the same problem was solved by the authors using other heuristic algorithms – swarm and genetic. This article presents the problem statement and the model of the working space, distinguishes the goals of robotic operation. The quality indicator is the total distance traveled by each of the robots in the group. To enable the original problem to be solved using the Hopfield neural network, a graph representation of the Hopfield is made by switching from the VRP to the TSP problem. The results of computational experiments confirming the effectiveness of the chosen approach for choosing a strategy of behavior of a group of mobile robots are shown.

Research on application of single index membership function based on power coefficient method

The power coefficient method can determine the satisfactory value and the unallowable value of the index, quantify multiple targets, and then determine the power coefficient value of each target. Combining the distribution characteristics of the four types of membership functions, this paper innovatively proposes to solve the regional distribution problem of single-index membership functions by means of an improved efficiency coefficient method, and combined with empirical research, the application of this mathematical method in practical engineering.