Kexin Liu

Kexin Liu

Auto-generated profile, based on publicly available information

Kexin Liu

Beihang University

Research Focus & Metrics

Academic Career Length

17Years (2010-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
6 May 2022 in IEEE Transactions on Neural Networks and Learning Systems
Verified
Journal Article
Closed access
TWTian Wang
Tian Wang
JCJiahui Chen
Jiahui Chen
JLJinhu Lü
Jinhu Lü
KLKexin Liu
Kexin Liu
AZAichun Zhu
Aichun Zhu
Baochang ZhangBaochang Zhang
Baochang Zhang
Modeling the spatiotemporal relationship (STR) of traffic data is important yet challenging for existing graph networks. These methods usually capture features separately in temporal and spatial dimensions or represent the spatiotemporal data by adopting multiple local spatial-temporal graphs. The first kind of method mentioned above is difficult to capture potential temporal-spatial relationships, while the other is limited for long-term feature extraction due to its local receptive field. To handle these issues, the Synchronous Spatio-Temporal grAph Transformer (S²TAT) network is proposed for efficiently modeling the traffic data. The contributions of our method include the following: 1) the nonlocal STR can be synchronously modeled by our integrated attention mechanism and graph convolution in the proposed S²TAT block; 2) the timewise graph convolution and multihead mechanism designed can handle the heterogeneity of data; and 3) we introduce a novel attention-based strategy in the output module, being able to capture more valuable historical information to overcome the shortcoming of conventional average aggregation. Extensive experiments are conducted on PeMS datasets that demonstrate the efficacy of the S²TAT by achieving a top-one accuracy but less computational cost by comparing with the state of the art.

21 June 2023 in IEEE Transactions on Industrial Informatics
Verified
Journal Article
Closed access
YQYiqun Qiao
Yiqun Qiao
JLJinhu Lü
Jinhu Lü
TWTian Wang
Tian Wang
KLKexin Liu
Kexin Liu
Baochang ZhangBaochang Zhang
Baochang Zhang
HSHichem Snoussi
Hichem Snoussi
Industrial sensors capture critical information for intelligent manufacturing maintenance. To promote equipment upgrading and manufacturing processes, intelligent decisions, and information learning play an important role. Although deep learning methods historically obtain excellent results, there is always a tradeoff between fine-tuning existing networks or designing models from scratch for sensor data processing. In this article, we propose the multihead attention self-supervised (MAS) representation model, which is a self-supervised learning-based sensor feature extraction network. To the best of our knowledge, this is the first time a self-supervised contrastive learning method using positive samples that represent multidimensional industry sensor data is being used for anomaly detection. We review alternative data augmentation methods proposed for better-representing sensor sequence data. We use this insight to design a new structure that adapts to the temporal characteristics of the application. We apply our method to a real-world water circulation system that uses a variety of industrial sensors. The effectiveness of the proposed MAS methods is demonstrated.

30 October 2017 in IEEE Transactions on Circuits and Systems I: Regular Papers
Verified
Journal Article
Closed access
KLKexin Liu
Kexin Liu
PDPeihu Duan
Peihu Duan
ZDZhisheng Duan
Zhisheng Duan
HCHaibin Cai
Haibin Cai
JLJinhu Lu
Jinhu Lu
This paper addresses the event-based leader-following consensus of a class of multi-agent systems with switching networks. Fully distributed event-triggered functions are proposed to select the sampling instants automatically. Detailed algorithms are given to design the feedback gains and coupling strength. The positive lower bound of the sampling intervals is captured to exclude Zeno behaviours. Finally, simulations are also carried out to verify the effectiveness of the event-triggered control scheme.

22 June 2023 in IEEE Transactions on Intelligent Vehicles
Verified
Journal Article
Closed access
XZXinzhou Zhao
Xinzhou Zhao
TWTian Wang
Tian Wang
YLYanjing Li
Yanjing Li
Baochang ZhangBaochang Zhang
Baochang Zhang
KLKexin Liu
Kexin Liu
HSHichem Snoussi
Hichem Snoussi
Target-driven visual navigation presents great po- tentials in scientific and industrial fields. It takes the target and environment observations as input. However, during training, we found that the agent sometimes got stuck in specific locations. Based on the analysis on visual information from a novel causal perspective, one of the most critical hurdles is the neglect of confounders in environments, which often leads to spurious correlations. Mitigating the confounding effect helps to discover the real causality and therefore are taken into consideration in other fields such as object detection. In this paper, we propose Causal Intervention Visual Navigation (CIVN), based on deep reinforcement learning (DRL) and causal intervention. We realize causal intervention using front-door adjustment as most con- founders are hard to model explicitly. Specifically, CIVN is imple- mented by Causal Attention, which is a reasonable approximation of causal intervention for visual navigation. Causal attention provides high-quality representation, which is leveraged by DRL and reduces the number of “stuck”. It is worth mentioned that causal intervention is for the first time applied by us in solving the confounding effect in target-driven visual navigation. Extensive experiments on AI2-THOR demonstrate that CIVN achieves better performance than prior arts. Specifically, the generalization for unknown targets and scenes is improved by a large margin, which is a basic research topic in visual navigation. Moreover, to obtain better generalization, we propose a novel experiment utilizing pre-trained models firstly.

28 February 2022 in IEEE Transactions on Systems, Man, and Cybernetics: Systems
Verified
Journal Article
Closed access
JLJiang Long
Jiang Long
WWWei Wang
Wei Wang
Jiangshuai HuangJiangshuai Huang
Jiangshuai Huang
JLJinhu Lu
Jinhu Lu
KLKexin Liu
Kexin Liu
This article investigates the leaderless consensus problem for uncertain high-order nonlinear multiagent systems with event-triggered communication. Under a directed graph condition, a fully distributed adaptive control strategy is presented. The main contributions are summarized as follows: 1) globally Lipschitz condition, as required in many existing literatures, is not needed in this article; 2) for each agent, only one filtered output signal is required to be broadcast to its neighbors, which can effectively relieve network transmission burden; 3) the distributed adaptive controllers and event-triggered conditions are co-designed based on a single Lyapunov function such that continuous monitoring of neighbors’ states is not needed; and 4) by introducing an exponential convergence term to the designed triggering condition, Zeno behavior is avoided while all the agents’ outputs can reach asymptotically leaderless consensus. Moreover, to easily implement the designed triggering condition and further reduce the triggering frequency when the consensus errors converge to the neighborhood of origin, a switching type of triggering condition is presented and bounded consensus can be guaranteed. Simulation results are given to validate the presented distributed consensus control scheme.

10 June 2022 in IEEE Transactions on Aerospace and Electronic Systems
Verified
Journal Article
Closed access
QQQingyu Qu
Qingyu Qu
KLKexin Liu
Kexin Liu
WWWei Wang
Wei Wang
JLJinhu Lu
Jinhu Lu
The rapid development of the aerospace industry puts forward the urgent need for the evolution of autonomous spacecraft rendezvous technology, which has gained significant attention recently due to increased applications in various space missions. This article studies the relative position tracking problem of the autonomous spacecraft rendezvous under the requirement of collision avoidance. An exploration-adaptive deep deterministic policy gradient (DDPG) algorithm is proposed to train a definite control strategy for this mission. Similar to the DDPG algorithm, four neural networks are used in this method, where two of them are used to generate the deterministic policy, whereas the other two are used to score the obtained policy. Differently, adaptive noise is introduced to reduce the possibility of oscillations and divergences and to cut down the unnecessary computation by weakening the exploration of stabilization problems. In addition, in order to effectively and quickly adapt to some other similar scenarios, a metalearning-based idea is introduced by fine-tuning the prior strategy. Finally, two numerical simulations show that the trained control strategy can effectively avoid the oscillation phenomenon caused by the artificial potential function. Benefiting from this, the trained control strategy based on deep reinforcement learning technology can decrease the energy consumption by 16.44% during the close proximity phase, compared with the traditional artificial potential function method. Besides, after introducing the metalearning-based idea, a strategy available for some other perturbed scenarios can be trained in a relatively short period of time, which illustrates its adaptability.