Selected publications
2022
- Supervised object-specific distance estimation from monocular images for autonomous driving
Yury Davydov, Wen-Hui Chen, Yu-Chen Lin
MDPI, 2022. DOI: 10.3390/s22228846.
Abstract: Accurate distance estimation is a requirement for advanced driver assistance systems (ADAS) to provide drivers with safety-related functions such as adaptive cruise control and collision avoidance. Radars and lidars can be used for providing distance information; however, they are either expensive or provide poor object information compared to image sensors. In this study, we propose a lightweight convolutional deep learning model that can extract object-specific distance information from monocular images. We explore a variety of training and five structural settings of the model and conduct various tests on the KITTI dataset for evaluating seven different road agents, namely, person, bicycle, car, motorcycle, bus, train, and truck. Additionally, in all experiments, a comparison with the Monodepth2 model is carried out. Experimental results show that the proposed model outperforms Monodepth2 by 15% in terms of the average weighted mean absolute error (MAE).
- Neural network data driven model of the process of analyzing control commands for a mobile robot in natural russian language
Alexander Sboev, Roman Rybka, Yury Davydov and Ivan Moloshnikov
In AIP Conference Proceedings, vol. 2425, no. 1, p. 340004. AIP Publishing LLC, 2022. DOI: 10.1063/5.0081608.
Abstract: The problem of controlling a robotic device in a natural environment, both of which are often very complex, seems extremely relevant. In particular, for English language this problem is highlighted in such works as [1],[2] and [3]. Here we present a neural network Data Driven model for Russian language solving multi-label classification task, that allows verbal features to be extracted from a natural text containing control commands with subsequent converting them directly into the format of control actions at the level of semiotic world of the robot. In Fig. 1 in the format of the paper [3](based on the approach presented in [4]) we present a graph that describes the possible movements of the robot and their attributes for a given model of the environment and the robot.
- A Comparison of Two Variants of Memristive Plasticity for Solving the Classification Problem of Handwritten Digits Recognition
Alexander Sboev, Yury Davydov, Roman Rybka, Danila Vlasov and Alexey Serenko
Biologically Inspired Cognitive Architectures Meeting, pp. 438-446. Springer, Cham, 2022. DOI: 10.1007/978-3-030-96993-6_48.
Abstract: Nowadays, the task of creating and training spiking neural networks (SNN) is extremely relevant due to their high energy efficiency achieved by implementing such networks via neuromorphic hardware. Especially interesting is the possibility of building SNNs based on memristors, which have properties that potentially allow them to be used as analog synapses. With that in mind, it seems relevant to study spike networks built upon plasticity rules that correspond to the experimentally observed nonlinear laws of conductivity change in memristors. Earlier it was shown that spiking neural networks trained with a biologically inspired local STDP (Spike-Timing-Dependent Plasticity) rule are capable of solving classification problems successfully. In addition, it was also demonstrated that classification problems can also be solved with spiking neural networks operating with a plasticity rule that models the change in conductivity in nanocomposite (NC) memristors. This paper presents a continuation of the study of the applicability of memristive plasticity rules on the handwritten digit recognition problem. Two types of memristive plasticity are compared: for nanocomposite and PPX memristors. It is shown that both models can successfully solve the classification problem, and the key differences between them are identified.
- Memristor-Based Spiking Neural Network with Online Reinforcement Learning
Danila Vlasov, Anton Minnekhanov, Roman Rybka, Yury Davydov, Alexander Sboev, Alexey Serenko, Alexander Ilyasov, and Vyacheslav A. Demin.
Available at SSRN 4229968, 2022. DOI: PREPRINT.
Abstract:Neural networks implemented in memristor-based hardware can provide fast and efficient in-memory computation, but traditional learning methods such as error back-propagation are hardly feasible in it. Spiking neural networks (SNNs) are highly promising in this regard, as their weights can be changed locally in a self-organized manner without demand for high-precision changes calculated with the use of information almost from the entire network. This problem is rather relevant for solving control tasks with neural-network reinforcement learning methods, as those are highly sensitive to any source of stochasticity in a model initialization, training, or decision-making procedure. This paper presents an online reinforcement learning algorithm, in which the change of connection weights is carried out after processing each environment state during interaction-with-environment data generation. Another novel feature of the algorithm is that it is applied to SNNs with memristor-based STDP-like learning rules. The plasticity functions are obtained from real memristors based on poly-p-xylylene and CoFeB-LiNbO$_3$ nanocomposite which were experimentally assembled and analyzed.The SNN is comprised of leaky integrate-and-fire neurons. Environmental states are encoded by timings of input spikes, and the control action is decoded by the first spike. The proposed learning algorithm solves the Cartpole benchmark task successfully. This result could be the first step toward implementing an real-time agent learning procedure in a continuous-time environment that can be run on neuromorphic systems with memristive synapses.
2021
- Modeling the Dynamics of Spiking Networks with Memristor-Based STDP to Solve Classification Tasks
Alexander Sboev, Danila Vlasov, Roman Rybka, Yury Davydov, Alexey Serenko, Vyacheslav Demin
MDPI, 2022, DOI: 10.3390/math9243237.
Abstract:The problem with training spiking neural networks (SNNs) is relevant due to the ultra-low power consumption these networks could exhibit when implemented in neuromorphic hardware. The ongoing progress in the fabrication of memristors, a prospective basis for analogue synapses, gives relevance to studying the possibility of SNN learning on the base of synaptic plasticity models, obtained by fitting the experimental measurements of the memristor conductance change. The dynamics of memristor conductances is (necessarily) nonlinear, because conductance changes depend on the spike timings, which neurons emit in an all-or-none fashion. The ability to solve classification tasks was previously shown for spiking network models based on the bio-inspired local learning mechanism of spike-timing-dependent plasticity (STDP), as well as with the plasticity that models the conductance change of nanocomposite (NC) memristors. Input data were presented to the network encoded into the intensities of Poisson input spike sequences. This work considers another approach for encoding input data into input spike sequences presented to the network: temporal encoding, in which an input vector is transformed into relative timing of individual input spikes. Since temporal encoding uses fewer input spikes, the processing of each input vector by the network can be faster and more energy-efficient. The aim of the current work is to show the applicability of temporal encoding to training spiking networks with three synaptic plasticity models: STDP, NC memristor approximation, and PPX memristor approximation. We assess the accuracy of the proposed approach on several benchmark classification tasks: Fisher’s Iris, Wisconsin breast cancer, and the pole balancing task (CartPole). The accuracies achieved by SNN with memristor plasticity and conventional STDP are comparable and are on par with classic machine learning approaches.