Optical Memory and Neural Networks最新文献

The work is devoted to the development of a classifier of motor activity patterns based on electroencephalograms (EEG) for a real-time brain-computer interface (BCI), which can be used in contactless control systems. Conducted studies of various methods for classifying motor EEG images have shown that their effectiveness significantly depends on the implementation of the stages of information processing in the BCI. The most effective classification method turned out to be the support vector machine. However, its long operating time and lack of accuracy make it difficult to use for implementing real-time BCI. Therefore, a classifier was developed using an ensemble of detectors, each of which is trained to recognize its own motor EEG image. A new EEG analysis algorithm based on event functions was applied. A study of the classifier showed that it is possible to achieve detection accuracy of 98.5% with an interface delay of 230 ms.

The paper considers the effect of transcendental factors on behavior of artificial beings, agents and robots. The boundary between the rational and transcendental relies on the type of an individual or more specifically, his sensory and intellectual abilities. For agents and simple robots all, except for tangible elements of the environment and manipulations with them, is transcendental. Such transcendental factors as environmental changes and algorithm modifications determined by the programmer, supervisor, or operator have significant effects on agents and communities of agents. Hardware malfunctions (transcendental events from an agent’s point of view) can be crucial for agents. Agents can take advantages from transcendental effects if the programmer realizes a feedback. Generation of a mental copy of an agent for making new agents allows continuity and social development. For intellectual robots the boundaries of the transcendental move away because of their ability to accommodate to new environment. However, in most cases the role of the transcendental even increases with the improvement of robots because there are consciousness and growth of communication possibilities. The consciousness changes as a result of learning of transcendental information, making robots change the behavior. Robot’s communication abilities enable transcendental (along with rational) information to be received from the data base in any amount. For people living together with intelligent robots, this sort of communication can become a tool for introducing human culture in the community of robots. This in turn would result in humanization of robots and establishment of good relations between robots and human beings.

Curriculum learning in reinforcement learning utilizes a strategy that sequences simpler tasks in order to optimize the learning process for more complex problems. Typically, existing methods are categorized into two distinct approaches: one that develops a teacher (a curriculum strategy) policy concurrently with a student (a learning agent) policy, and another that utilizes selective sampling based on the student policy’s experiences across a task distribution. The main issue with the first approach is the substantial computational demand, as it requires simultaneous training of both the low-level (student) and high-level (teacher) reinforcement learning policies. On the other hand, methods based on selective sampling presuppose that the agent is capable of maximizing reward accumulation across all tasks, which may lead to complications when the primary mission is to master a specific target task. This makes those models less effective in scenarios requiring focused learning. Our research addresses a particular scenario where a teacher needs to train a new student in a new short episode. This constraint compels the teacher to rapidly master the curriculum planning by identifying the most appropriate tasks. We evaluated our framework across several complex scenarios, including a partially observable grid-world navigation environment, and in procedurally generated open-world environment Crafter.

In this paper, we explored the performance of interpretable sentiment analysis models of different combinations for the Chinese text in social media. We made experiment to study how performance varies with the change of combination of different segmentation strategies and dictionary of words or n-grams. We found that with some good combination of segmentation strategies and dictionary of words or n-grams, the result can be improved and overtake the performance of ordinary sentiment analysis model of Chinese language. This way we show the importance of selection of segmentation strategies and dictionary for the sentiment analysis of Chinese text.

In this paper, we propose a new single-photo 3D reconstruction model DiffuseVoxels focused on 3D inpainting of destroyed parts of a building. We use frustum-voxel model 3D reconstruction pipeline as a starting point for our research. Our main contribution is an iterative estimation of destroyed parts from a Gaussian noise inspired by diffusion models. Our input is twofold. Firstly, we mask the destroyed region in the input 2D image with a Gaussian noise. Secondly, we remove the noise through many iterations to improve the 3D reconstruction. The resulting model is represented as a semantic frustum voxel model, where each voxel represents the class of the reconstructed scene. Unlike classical voxel models, where each unit represents a cube, frustum voxel models divides the scene space into trapezium shaped units. Such approach allows us to keep the direct contour correspondence between the input 2D image, input 3D feature maps, and the output 3D frustum voxel model.

This research develops improved Robust Adversarial Models (RAM) to enhance Intrusion Detection Systems’ (IDS) robustness against evasion attacks. Malicious packets crafted using Scapy were infused into open-source datasets NSL-KDD and CICIDS obtained from Kaggle. Experiments involved passing this traffic through baseline IDS model such as in a free open-source IDS Snort and the improved RAM. Training processes employed perturbations using Generative Adversarial Networks (GAN), Fast Gradient Sign Methods (FGSM), and Projected Gradient Descent (PGD) against reinforcement learning of features and labels from the autoencoder model. The robust adversarial model showed 34.52% higher accuracy, 59.06% higher F1-score and 85.26% higher recall than the baseline IDS Snort model across datasets. Comparative analysis demonstrated the improved RAM’s enhanced resilience, performance, and reliability in real-world scenarios, advancing IDS models' and network infrastructures' security posture.

In this research we estimate the polarisation influence of the incident radiation on the measurement by a pyramidal aperture cantilever. The numerical modelling of the detection process was made by applying the frequency depended finite-difference time-domain method. We numerically demonstrated that the angle of incidence and the plane of inclination can affect on the measurement process by the aperture aluminum cantilever while the aperture shape has not any influence on the measurement process for both proposed types of incident light polarization: linear and circular left. Simulation results show that as the tilt angle for rotation of incident light increases the total intensity inside the cantilever decreases by about 50 and 30% for the linearly and circularly polarized light. It prooves that aperture aluminum cantilever is weakly sensitive to the longitudinal component.

In this work, spin-orbit conversion in a vector optical vortex will be considered. The polarization in such a beam corresponds to the polarization of a cylindrical vector beam, that is, it is initially linear at each point. It is shown numerically and analytically using the Richards-Wolf formalism that zones with non-zero longitudinal spin angular momentum are formed in the focal spot, i.e. zones with elliptical polarization. It has been experimentally shown that for the case when the topological charge of the optical vortex coincides with the order of the beam, the observed spin-orbit conversion is large enough to be recorded in the paraxial approximation.

The aim of this work is to develop and study the technology of automatic determination of indications for 2RT-laser treatment of AMD by SD-OCT images based on artificial intelligence methods. This is necessary to improve the accuracy and efficiency of AMD diagnosis, as well as to provide faster and more accurate treatment assignment to each patient. The U-Net architecture was chosen as the neural network architecture to extract the area of interest in the retinal OCT image. The VGG16 architecture was used as the neural network architecture for classification. These architectures are well established. As a result of training, the model showed a fairly high accuracy of 90% for segmentation and 98% for classification. Automatic localization and classification based on SD-OST images will allow the most accurate determination of indications for 2RT laser treatment. This will significantly reduce the burden on physicians and make diagnostics more accessible.

We investigate the common topological charge of a superposition of parallel identical vortex beams with an arbitrary transverse shape, either Laguerre–Gaussian beams or Bessel–Gaussian beams or some other vortex beams with rotationally symmetric intensity distribution. It is known that if all the beams in the superposition have the same phase then the common topological charge of the whole superposition equals the topological charge of each constituent beam n. We show that if the beams are located on a circle and their phases increase linearly along this circle so that the phase delay between the neighbor beams on the circle is 2πp/N with N being the number of beams and p being an integer number, then the common topological charge of the superposition is equal to n + p.