Artificial intelligence on transport

Main article: Artificial Intelligence (AI), Artificial intelligence (AI)

2020: DVFU and MIPT develop mathematical algorithms for the solution of transport tasks and work with data

On April 28, 2020 it became known that scientists of Far Eastern Federal University (DVFU) together with colleagues from the Moscow Institute of Physics and Technology (MIPT) develop mathematical methods of convex optimization for the accelerated solution of the broadest spectrum of problems of economy, science, many applied directions of human activity. Scientists announced the results in the book "Numerical Methods of Convex Optimization" of Springer publishing house.

According to the company, algorithms are adaptive, i.e. in the course of work will recognize all necessary parameters, and are economical, their work requires rather small amount of memory. Use these algorithms, it is reasonable, for example, for modeling of traffic flows, fight against traffic jams and optimization of routes of cargo transport, calculation of a fare, ranging of web pages, the solution of the return tasks when it is required to understand the reasons which generated some effects.

At the heart of rough or convex optimization the principle of decomposition lies. It means that the big task can often be separated into much small which coordinate then among themselves using the special coordinating task. For April, 2020 it is relevant for work with Big Data. In the modern world often there is a need to process, transfer the data measured by gigabytes and more and also to solve very complex problems on their basis. Naive direct approach, even using the fastest supercomputers, will demand for the solution of such tasks hundreds and thousands of years. The mathematics accelerates these processes so that they purchase practical sense. Evgeny Nurminsky, professor of School of natural sciences of DVFU told

The scientist told that for a classical task of the solution of a linear equation system modern algorithms are repeatedly more effective than traditional methods which labor input is approximately equal to a cube from quantity of variables.

If in a problem of 5 variables, then you spend 125 transactions for its solution and, say, 1 second of time. If variables 50, then you need 125 thousand transactions and about 15 minutes. Provide that variables 5000. It is necessary to solve a problem by a traditional method about 30 years. These methods will reduce this time to 40 seconds. Of course, it is possible to spend tens of billion rubles or dollars, to build the supercomputer value about pyramid of Cheops and energy consumption of the ice breaker which nevertheless will solve your problem in a day. But whether it is better to select a thousand part of this amount for talented pupils who will make much more? Of course, and the supercomputer will not prevent! Evgeny Nurminsky, professor of School of natural sciences of DVFU added

On the basis of algorithms it is possible to create a method to process the "heavy" image so that at the exit it required 10 times less place, than on an input, but saved 95 percent of initial properties. At the same time, such picture approximately cannot be distinguished from initial.

If to speak very ordinary, optimization helps to dig less both literally, and figuratively. For example, it is necessary to decide how to dig out the underground passage connecting four points at the intersection so that it was possible to get from any input to any exit on other side of the road. It would seem, it is necessary to draw a square and to dig through tunnels on two of its diagonals. The mathematics says to us what for smaller labor costs should be dug differently. Projecting very big mechanical constructions, methods of convex optimization can be considered how to receive the smallest lot of these constructions without loss in durability. Other example — convex optimization helps to define the optimal method of collection of fares on paid roads leading to minimization of total losses of net surfers on the way. Alexander Gasnikov, the associate professor of mathematical bases of management of MIPT explained

The scientist noted that problems of optimization have a direct bearing on life, i.e. the nature often speaks mathematical language and to understand its device, it is necessary to solve a problem of optimization.

In difficult (nonconvex) tasks we, however, in most cases cannot receive the ideal solution, but is frequent it and it is not required. In practice often quite arrange the suboptimal results received with some error, but for reasonable time. It is applicable, for example, to many problems of deep learning. Alexander Gasnikov, the associate professor of mathematical bases of management of MIPT added

The book "Numerical Methods of Convex Optimization" opens traditional and more modern methods of convex optimization. Work is intended for students, teachers, scientists and practicians whose sphere of activity is connected with convex optimization. Scientists of DVFU and MIPT told about the methods of convex optimization in separate chapter "Subgradiyentny methods of the solution of a problem of convex optimization with small costs of memory".

Publishers collected under one cover of the leading scientists who develop area of convex optimization. The book can be useful to all who wish to obtain up-to-date information about the state of affairs and tools which this field of mathematics contains.

Algorithms of convex optimization can be applied to the correct creation of models of the real world, and in such areas as collecting, processing and data transmission, machine learning and artificial intelligence, engineering sciences, economy and business, computer chemistry, physics and medicine.

Robotics