Optical Memory and Neural Networks

In this paper, the global Mittag-Leffler stability of fractional Hopfield neural networks (FHNNs) with \(\delta \)-inverse hölder neuron activation functions are considered. By applying the Brouwer topological degree theory and inequality analysis techniques, the proof of the existence and uniqueness of equilibrium point is addressed. By constructing the Lure’s Postnikov-type Lyapunov functions, the global Mittag-Leffler stability conditions are achieved in terms of linear matrix inequalities (LMIs). Finally, three numerical examples are given to verify the validity of the theoretical results.

To reduce random access memory (RAM) requirements and to increase speed of recognition algorithms we consider a weight discretization problem for trained neural networks. We show that an exponential discretization is preferable to a linear discretization since it allows one to achieve the same accuracy when the number of bits is 1 or 2 less. The quality of the neural network VGG-16 is already satisfactory (top5 accuracy 69%) in the case of 3 bit exponential discretization. The ResNet50 neural network shows top5 accuracy 84% at 4 bits. Other neural networks perform fairly well at 5 bits (top5 accuracies of Xception, Inception-v3, and MobileNet-v2 top5 were 87%, 90%, and 77%, respectively). At less number of bits, the accuracy decreases rapidly.

Among a number of problems in the behavior planning of an unmanned vehicle the central one is movement in difficult areas. In particular, such areas are intersections at which direct interaction with other road agents takes place. In our work, we offer a new approach to train of the intelligent agent that simulates the behavior of an unmanned vehicle, based on the integration of reinforcement learning and computer vision. Using full visual information about the road intersection obtained from aerial photographs, it is studied automatic detection the relative positions of all road agents with various architectures of deep neural networks (YOLOv3, Faster R-CNN, RetinaNet, Cascade R-CNN, Mask R-CNN, Cascade Mask R-CNN). The possibilities of estimation of the vehicle orientation angle based on a convolutional neural network are also investigated. Obtained additional features are used in the modern effective reinforcement learning methods of Soft Actor Critic and Rainbow, which allows to accelerate the convergence of its learning process. To demonstrate the operation of the developed system, an intersection simulator was developed, at which a number of model experiments were carried out.

The problem of image understanding currently attracts considerable attention of researchers, since its solution is critically important for a significant number of applied problems. Among the most critical components of this problem is the semantic segmentation of images, which provides a classification of objects on the image at the pixel level. One of the applied problems in which semantic segmentation is an essential element of the process of solving them is the information support of the behavior control systems for robotic UAVs. Among the various types of images that are used to solve such problems, it should be noted images obtained by remote sensing of the Earth’s surface. A significant number of variants of neuroarchitectures based on convolutional neural networks have been proposed to solve the semantic image segmentation problem, However, for various reasons, not all of them are suitable for working with images of the Earth’s surface obtained using remote sensing. Neuroarchitectures that are potentially suitable for solving the problem of semantic segmentation of images of the Earth’s surface are identified, a comparative analysis of their effectiveness concerning this task is carried out.

We research adaptive multidimensional signal interpolators based on switching between several interpolating functions at each signal sample. We perform the switching by decision rule, which is optimized for each signal in the parameter space of this decision rule. Algorithms for factorization and dimension reduction of decision rules are proposed. We investigate new classes of interpolating functions and systems of local features. We propose fitting procedures for adaptive interpolators. We perform the software implementation of the developed algorithms. A numerical experiment in natural multidimensional signals (video, remote sensing data and hyperspectral data) confirms the gain of the adaptive interpolator.