Deep learning has achieved excellent performance in computer vision tasks, like image recognition, natural language processing, etc. However, in real-world applications, special circumstances brought about by the exte...
详细信息
ISBN:
(纸本)9781665476881
Deep learning has achieved excellent performance in computer vision tasks, like image recognition, natural language processing, etc. However, in real-world applications, special circumstances brought about by the external world may create domain bias caused by distribution discrepancy between training and testing data, leading to degrading model performance. For example, when auto-driving meets hazy weather, the model performance will drop significantly. In this paper, we explore to solve this problem by utilizing modern Domain Adaptation (DA) methods, which generalizes from the source domain to the target domain by minimizing the distribution difference caused by dataset bias. We firstly propose the cross-domain haze image datasets and benchmark the five classic DA methods. the experiments show that DA methods can mitigate the negative effect of haze and significantly improves the model performance for visual recognition.
Cellular networks are growing in complexity at increasing speed and the geographical locations in which they are deployed in are getting denser. Traditional control methods fall short in providing a scalable and dynam...
Cellular networks are growing in complexity at increasing speed and the geographical locations in which they are deployed in are getting denser. Traditional control methods fall short in providing a scalable and dynamic way of adapting the network to new conditions. Distributed multiagent reinforcement learning successfully addresses scalability problems seen in centralized approaches. the question of achieving learning with constraint satisfaction in distributed systems is still left unanswered in the state-of-the-art. In this work, we aim to perform distributed multi-agent constrained reinforcement learning in order to learn a policy online while satisfying imposed constraints. We use a coordination graph to model the interactions between agents and decompose the global value function. A conservative safety critic is trained in parallel to evaluate the safety of proposed actions. Our method allows for separate training of boththe critic and the value network independently of each other, and hence offers flexibility in how and when to train the different models. the results are compared to a baseline using no safety critic. Simulations show that the agents succeed in learning a policy that can satisfy the constraints, while still maximizing the objective.
When using evidence theory to identify targets, how to generate basic probability assignment (BPA) based on the collected data is still an open issue. Based on this problem, a novel method to generate BPA based on mem...
详细信息
When using evidence theory to identify targets, how to generate basic probability assignment (BPA) based on the collected data is still an open issue. Based on this problem, a novel method to generate BPA based on membership function and principal component analysis (PCA) is proposed. First, this paper proposed a novel membership function and divide the data set into the training set and testing set, the membership model of each attribute is constructed based on the training set data. Secondly, the testing set sample is input to obtain the initial BPA of each attribute. thirdly, the contribution rate of each attribute is used by PCA. Finally, the final BPA is obtained by discounting the initial BPA according to the contribution rate. the results of the experiment have demonstrated the classification accuracy under the Iris dataset is higher than other methods, and the average recognition rate of the Iris dataset is 97.3%.
Legal similar case retrieval is becoming increasingly important in the judicial field. Traditional methods for similar case retrieval largely rely on key-word matching, which fails to deeply understand the legal seman...
详细信息
Deep reinforcement learning (DRL) allows a system to interact with its environment and take actions by training an efficient policy that maximizes self-defined rewards. In autonomous driving, it can be used as a strat...
详细信息
ISBN:
(数字)9798331509231
ISBN:
(纸本)9798331509248
Deep reinforcement learning (DRL) allows a system to interact with its environment and take actions by training an efficient policy that maximizes self-defined rewards. In autonomous driving, it can be used as a strategy for high-level decision making, whereas low-level algorithms such as the hybrid A* path planning have proven their ability to solve the local trajectory planning problem. In this work, we combine these two methods where the DRL makes high-level decisions such as lane change commands. After obtaining the lane change command, the hybrid A* planner is able to generate a collision-free trajectory to be executed by a model predictive controller (MPC). In addition, the DRL algorithm is able to keep the lane change command consistent within a chosen time-period. Traffic rules are implemented using linear temporal logic (LTL), which is then utilized as a reward function in DRL. Furthermore, we validate the proposed method on a real system to demonstrate its feasibility from simulation to implementation on real hardware.
Aiming at the problem that the spatial features of pedestrian images are not aligned in current pedestrian re-identification and the network model cannot fully express the pedestrian information due to pose changes an...
详细信息
Aiming at the problem that the spatial features of pedestrian images are not aligned in current pedestrian re-identification and the network model cannot fully express the pedestrian information due to pose changes and occlusions, a method based on spatial transformation and multi-methods feature fusion is proposed. Firstly, for the pedestrian re-identification system, a processing method for enhancing the retrieval of pedestrians is provided, and the pedestrian images with more noise to be retrieved are denoised by means of side-window filtering; secondly, the spatial transformation network is improved. Channel attention and self-constrained branches are introduced to automatically align pedestrian spatial features to solve the problem of inconsistency in spatial semantic information caused by unaligned pedestrian image regions; then, multi-scale features are extracted from different deep layers of the backbone network, and coordinates attention and batch normalization of instances are integrated into different deep branches. Finally, the features of each branch are fused to obtain feature information with high representation ability. During the network training process, the dual loss function strategy is used to jointly optimize the model. Multiple experiments show that the proposed method has a higher recognition rate than other existing methods.
Traffic forecasting is an important part of the smart transportation system, and accurate traffic forecasting is crucial for urban traffic scheduling and public travel planning. the traffic forecasting problem is grea...
详细信息
Traffic forecasting is an important part of the smart transportation system, and accurate traffic forecasting is crucial for urban traffic scheduling and public travel planning. the traffic forecasting problem is greatly affected by the time dimension, and it is of great significance to investigate and summarize the related methods of time series traffic forecasting. Aiming at the problem of time series traffic forecasting, this paper focuses on the existing time series traffic forecasting models based on deep learning, and studies and analyzes the application fields and structural characteristics of different forecasting models. Finally, the current mainstream traffic prediction datasets are introduced, and the main challenges and solutions in the current traffic prediction field are discussed, which provides a reference for solving the problem of intelligent traffic prediction.
作者：
Trezubov, KirillAvksentieva, ElenaLuzhnyak, ValeriyaShulgin, IlyaITMO University
Kronverksky avenue 49 bldg. A St. Petersburg197101 Russia
St. Petersburg Institute for Informatics and Automation of the Russian Academy of Sciences 39 14th Line St. Petersburg199178 Russia
Podbelskogo road 7 Pushkin St. Petersburg196608 Russia
At the moment, devices for monitoring the physiological condition of the animal are beginning to play an increasingly important role. these devices and systems allow for operational control and monitoring of the anima...
详细信息
Hand-eye calibration methods for surgical robots are employed to derive a transformation between the robot's base motor and visual coordinate systems. Accurately completing hand-eye calibration procedures provides...
详细信息
ISBN:
(纸本)9781665476881
Hand-eye calibration methods for surgical robots are employed to derive a transformation between the robot's base motor and visual coordinate systems. Accurately completing hand-eye calibration procedures provides an important guarantee that a surgical robot will exhibit positioning and execution accuracy sufficient for assisting surgeons in successfully completing surgical procedures. To improve the accuracy of robot hand-eye calibration methods based on backpropagation neural network (BPNN) models, we propose a modified BP neural network optimized using the sparrow search algorithm for hand-eye calibration model (TSSABPNN), which can enhance population initialization by applying tent mapping. Furthermore, we also design a new sliding 3D calibration tool. the sparrow search algorithm exhibits good local exploration ability, and we introduce a tent map with ergodic characteristics to initialize the sparrow population information, which further improves the network's global search ability and convergence rate. Finally, we experimentally analyze four calibration models: TSSABP NN model, a BP NN model optimized using a genetic algorithm of simulated annealing (GASABPNN), an unoptimized BP NN model, and the traditional singular value decomposition method. the results indicate that the proposed TSSABP NN model exhibits the maximum calibration precision and best robustness and iteratively converges faster.
Despite the importance of honeybees as pollinators for the entire ecosystem and their recent decline threatening agricultural production, the dynamics of the living colony are not well understood. In our EU H2020 Robo...
详细信息
ISBN:
(数字)9798350358513
ISBN:
(纸本)9798350358520
Despite the importance of honeybees as pollinators for the entire ecosystem and their recent decline threatening agricultural production, the dynamics of the living colony are not well understood. In our EU H2020 RoboRoyale project, we aim to support the pollination activity of the honeybees through robots interacting withthe core element of the honeybee colony, the honeybee queen. In order to achieve that, we need to understand how the honeybee queen behaves and interacts withthe surrounding worker bees. To gather the necessary data, we observe the queen with a moving camera, and occasionally, we instruct the system to perform selective observations elsewhere. In this paper, we deal withthe problem of searching for the honeybee queen inside a living colony. We demonstrate that combining spatio-temporal models of queen presence with efficient search methods significantly decreases the time required to find her. this will minimize the chance of missing interesting data on the infrequent behaviors or queen-worker interactions, leading to a better understanding of the queen’s behavior over time. Moreover, a faster search for the queen allows the robot to leave her more frequently and gather more data in other areas of the honeybee colony.
暂无评论