A5100404784- Reinforcement Learning in Robotics
- Energy Efficient Wireless Sensor Networks
- Face and Expression Recognition
- Bayesian Modeling and Causal Inference
- Face recognition and analysis
- Artificial Intelligence in Games
- Elevator Systems and Control
- Machine Learning and Algorithms
- Emotion and Mood Recognition
- Multi-Agent Systems and Negotiation
- Fault Detection and Control Systems
- Anomaly Detection Techniques and Applications
- Distributed Sensor Networks and Detection Algorithms
- Target Tracking and Data Fusion in Sensor Networks
- Multimedia Communication and Technology
- Data Stream Mining Techniques
- Mobile Ad Hoc Networks
- Security in Wireless Sensor Networks
- IoT-based Smart Home Systems
- Embedded Systems Design Techniques
- Internet of Things and AI
- Gaze Tracking and Assistive Technology
- Robotic Path Planning Algorithms
- Data-Driven Disease Surveillance
- Adversarial Robustness in Machine Learning
Peng Cheng Laboratory
2022-2023
China Mobile (China)
2022
Shenzhen Polytechnic
2006-2017
Wuhan University of Technology
2009
Yunnan University
2009
Southeast University
2008
A good skin detector that is capable of capturing tones under different conditions important for human-machine interaction applications. In a general situation, detectors, such as probability maps or Gaussian mixture models, achieve acceptable segmentation results. However, the false positive rate increases significantly when are in shadow skin-like background objects similar illumination. this paper, we propose novel feature learning algorithm based on stacked autoencoders, which deep...
The sensor scheduling for energy-efficient target tracking with high performance in wireless networks (WSNs) is a dilemma problem. By analyzing the intrinsic relationship between and energy consumption, we cast problem of WSN as optimal policy partially observable Markov decision process (POMDP), propose dynamic cluster members (DCMS) algorithm to solve tradeoff consumption. First, exploit an election method, based on mixed weights signal strength residual node, choose head node. Then, seem...
Learning and planning in partially observable Markov decision processes (POMDPs) is computationally intractable real-time system. In order to address this problem, paper proposes a belief policy reuse (BPR) method avoid repeated computation. Firstly, the evaluation mechanism based on Kullback-Leibler divergence presented as similarity metric between beliefs belief-policy library. If current similar any of past ones, library reused. Otherwise, BPR exploits Monte-Carlo particle explore new...
abstract In this paper, we propose a novel posterior belief clustering (PBC) algorithm to solve the tradeoff between target tracking performance and sensors energy consumption in wireless sensor networks. We model under dynamic uncertain environment using partially observable Markov decision processes (POMDPs), transform optimization of into yielding optimal value function POMDPs. analyze error class continuous beliefs by Kullback–Leibler (KL) divergence, cluster these one based on KL...
This paper presents a hybrid agent architecture, which includes reactive architecture and deliberative the former enhances reactivity of agents, later improves intelligence agents. The also introduces novel decision algorithm to make accomplish tasks promptly artificially. had successfully applied RoboCup simulation system, results show they are effective
Bayesian reinforcement learning provides an elegant solution to the optimal tradeoff between exploration and exploitation of uncertainty in learning. Unfortunately, size parameters grows exponentially with problem horizon. In this paper, we propose a novel Monte Carlo tree search for approach using compact factored representation, solve online. At first, exploit representation describe states transition function reduce parameters. Then, can formulate model-based as partially observable...
Learning the enormous number of parameters is a challenging problem in model-based Bayesian reinforcement learning. In order to solve problem, we propose factored learning (F-BRL) approach. F-BRL exploits representation describe states reduce parameters. Representing conditional independence relationships between state features using dynamic networks, adopts inference method learn unknown structure and networks simultaneously. A point-based online value iteration approach then used for...
An experimentally feasible scheme for teleporting an arbitrary and unknown entangled state is proposed. In this paper, we use a cluster as the quantum channel, do not need any joint Bell-state measurement (BSM). Our based on Josephson charge qubits, successful probability fidelity of teleportation can both reach unity. Moreover, current be realized within experimental technology.
This paper introduces the application research on gas pipeline leakage repair based Multi-Agent. For characteristics of Multi - Agent system and collection information exchange between these Agents were described in detail. The core is scheduling algorithm Multi-Agent cooperation time. So, cases modeling optimization scheme are established as a result theoretical foundation. Finally, examples operation verified by simulation platform AnyLogic software.
Abstract The use of the internet things (IoT) is steadily increasing in a wide range applications. Integration IoT, computer vision, and artificial intelligence can improve people's daily life various domains such as smart homes, cities, industries. There are large number face recognition attribute scenarios reality, industry commonly decomposes these tasks, with three models responsible for handling detection, recognition, recognition. multi‐model approach requires lot computational...
Major concern of energy constrained WSN technology is to design efficiency and low latency communication protocol, multilayer clustering the whole system nodes an efficient solution prolong network lifetime decrease event. By using layer rotation, all has been sufficiently used. In simulation, our proposed protocol shows better than LEACH (Low Energy Adaptive Clustering Hierarchy) furthermore this have in most test scenario.
Benefiting from deep learning, the accuracy of face expression recognition tasks based on convolutional neural networks has been greatly improved. However, traditional SoftMax activation function lacks ability to discriminate between classes. To solve this problem, industry proposed several functions softmax, such as A-softmax, LMCL, etc. We investigate geometric significance weights a fully connected layer and consider class centers. By extracting feature vector samples extending...
With the application of 5G technology in field education, construction smart campus has set off a wave digital transformation. At same time, traditional is also facing exponential growth number Internet Things devices, servers, and terminals, which makes it difficult to achieve flat management. In view current difficulties campus, this paper proposes architecture based on blockchain technology. Unlike architecture, combines characteristics decentralization, high confidentiality, data sharing...
Partially Observable Markov Decision Processes (POMDP) provides piecewise-linear a natural and principled framework for sequential decision-making under uncertainty. However, large-scale POMDP suffers from the exponential growth of belief points policy trees space. We present new point-based incremental pruning algorithm based on piecewise linearity convexity value function. Instead reasoning about whole space when cross-sums in construction, our uses to perform approximate by generating...
Bayesian reinforcement learning has turned out to be an effective solution the optimal tradeoff between exploration and exploitation. However, in practical applications, parameters with exponential growth are main impediment for online planning learning. To overcome this problem, we bring factored representations, model-based learning, together a new approach. Firstly, exploit representation describe states reduce size of parameters, adopt inference method learn unknown structure...
The online planning and learning in partially observable Markov decision processes are often intractable because belief states space has two curses: dimensionality history. In order to address this problem, paper proposes a point-based Monte Carto approach POMDPs. This involves performing value backup at specific reachable points, rather than over the entire simplex, speed up computation processes. Then Carlo tree search algorithm is exploited share of actions across each subtree so as...