Vol 44, No 3 (2018)

The task of displacement estimation for frames of a video sequence is considered. A new convolutional neural network architecture for the optical flow problem is proposed. The method is based on learning the regularization operator for a fast optimization method. The proposed method has low computational complexity and memory footprint at test time. The neural network architecture is based on unrolling iterations of a fast primal-dual method as layers of a convolutional neural network. Iterations of the optimization method are represented as convolutions with filters that are trained on ground truth data by backpropagation. A real-time implementation using graphics processing units is proposed. Experimental results demonstrate an improved quality of the optical flow field as compared to the optimization method based on a fixed regularization operator.

A new look at the problem of constructing a scheduler in the case of a group of strictly periodic tasks is proposed. The structure of the system of periods is represented in terms of graph theory. A criterion for the existence of a conflict-free schedule based on this representation is obtained, and generic schemes of algorithms for constructing such a schedule are described. The proposed approach is illustrated by building schedules for a number of strictly periodic tasks.

In this paper, we present a Big Data analysis paradigm related to smart cities using cloud computing infrastructures. The proposed architecture follows the MapReduce parallel model implemented using the Hadoop framework. We analyse two case studies: a quality-of-service assessment of public transportation system using historical bus location data, and a passenger-mobility estimation using ticket sales data from smartcards. Both case studies use real data from the transportation system of Montevideo, Uruguay. The experimental evaluation demonstrates that the proposed model allows processing large volumes of data efficiently.

A representation of algorithms extracted from binary code by reverse engineering is discussed. Both intermediate representations designed for automatic analysis and final representations passed to the end user are considered. The two main tasks of reverse engineering—automatic detection of exploitable vulnerabilities and discovery of undocumented features— are analyzed. The basic scheme of the system implementing the automatic detection of exploitable vulnerabilities is presented and the key properties of the intermediate representation designed for solving this problem using an efficient generation of a system of equations for an SMT solver are described. The workflow for discovering undocumented features is described. These steps are the localization of the algorithm, its representation in the form that is convenient for analysis, and investigation of its properties. To automate the first phase, a combined static and dynamic representation is constructed, which includes OS-level events and calls to library functions; they serve as anchor points used by the analyst for the algorithm localization. The further support of localization uses code slicing and navigation algorithms. Once the algorithm is localized, the further work goes in two directions: interactive construction of a compact annotated representation of the algorithm by a flowchart and automated investigation of the algorithm properties aimed at determining declared and undeclared data flows. The representation of the algorithm is based on the construction of simplified models of functions taking into account input and output buffers and on the automatic detection of data dependences between buffers of various function calls. The overall scenario of the analyst' work with such a flowchart in the context of discovering undocumented features is described; this scenario is based on annotating the declared data flows and on the automatic detection of undeclared data flows. In conclusion, an example of the resulting representation is discussed and the directions of further research are discussed.