Liang Xu

Dr. Liang Xu

Institute of Mechanical Engineering, École polytechnique fédérale de Lausanne, CH-1015 Lausanne, Switzerland

Research Focus & Metrics

Academic Career Length

12Years (2015-Present)

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

Representative Publications

View All Publications
12 June 2018 in IEEE Transactions on Automatic Control
Verified
Journal Article
Closed access
JZJianying Zheng
Jianying Zheng
LXLiang Xu
Liang Xu
LXLihua Xie
Lihua Xie
KYKeyou You
Keyou You
In this paper, we study the mean-square consensusability problem of discrete-time linear multi-agent systems over directed networks with constant communication delay and packet dropouts. When the packet dropouts in different communication channels are identical, the consensusability problem is transformed into a simultaneous mean-square stabilization problem of systems with constant time delay and packet dropout. When the packet dropouts are non-identical, we build the dynamics over edges based on the notion of compressed edge Laplacian for directed graphs to separate the packet dropouts from the network topology. Sufficient consensusability conditions are obtained for both cases in terms of the communication delay, the packet dropout rates, the communication network topology and the agent dynamics.

24 January 2023 in IEEE Transactions on Automatic Control
Verified
Journal Article
Closed access
CGClara Lucía Galimberti
Clara Lucía Galimberti
LFLuca Furieri
Luca Furieri
LXLiang Xu
Liang Xu
GFGiancarlo Ferrari-Trecate
Giancarlo Ferrari-Trecate
Deep neural networks (DNNs) training can be difficult due to vanishing and exploding gradients during weight optimization through backpropagation. To address this problem, we propose a general class of Hamiltonian DNNs (H-DNNs) that stem from the discretization of continuous-time Hamiltonian systems and include several existing DNN architectures based on ordinary differential equations. Our main result is that a broad set of H-DNNs ensures nonvanishing gradients by design for an arbitrary network depth. This is obtained by proving that, using a semi-implicit Euler discretization scheme, the backward sensitivity matrices involved in gradient computations are symplectic. We also provide an upper bound to the magnitude of sensitivity matrices and show that exploding gradients can be controlled through regularization. The good performance of H-DNNs is demonstrated on benchmark classification problems, including image classification with the MNIST dataset.

9 May 2019 in IEEE Transactions on Automatic Control
Verified
Journal Article
Closed access
LXLiang Xu
Liang Xu
YMYilin Mo
Yilin Mo
LXLihua Xie
Lihua Xie
This paper studies the consensusability problem of multi-agent systems (MASs), where agents communicate with each other through Markovian packet loss channels. We try to determine conditions under which there exists a linear distributed consensus controller such that the MAS can achieve mean square consensus. We first provide a necessary and sufficient consensus condition for MASs with single input and i.i.d. channel losses, which complements existing results. Then we proceed to study the case with identical Markovian packet losses. A necessary and sufficient consensus condition is firstly derived based on the stability of Markov jump linear systems. Then a numerically verifiable consensus criterion in terms of the feasibility of linear matrix inequalities (LMIs) is proposed. Furthermore, analytic sufficient conditions and necessary conditions for mean square consensusability are provided for general MASs. The case with nonidentical packet loss is studied subsequently. The necessary and sufficient consensus condition and a sufficient consensus condition in terms of LMIs are proposed. In the end, numerical simulations are conducted to verify the derived results.

23 June 2020 in IEEE Transactions on Automatic Control
Verified
Journal Article
Closed access
LXLiang Xu
Liang Xu
YMYilin Mo
Yilin Mo
LXLihua Xie
Lihua Xie
This article studies the remote state estimation problem of linear time-invariant systems with stochastic event-triggered sensor schedules in the presence of packet drops between the sensor and the estimator. Due to the existence of packet drops, the Gaussianity at the estimator side no longer holds. It is proved that the system state conditioned on the available information at the estimator side is Gaussian mixture distributed. The minimum-mean-square-error (MMSE) estimator can be obtained from the bank of Kalman filters. Since the optimal estimators require exponentially increasing computation and memory with time, suboptimal estimators to reduce computational complexities by limiting the length and numbers of hypotheses are further provided. In the end, simulations are conducted to illustrate the performance of the optimal and suboptimal estimators.

28 June 2022 in IEEE Control Systems Letters
Verified
Journal Article
Closed access
MZMuhammad Zakwan
Muhammad Zakwan
LXLiang Xu
Liang Xu
GFGiancarlo Ferrari-Trecate
Giancarlo Ferrari-Trecate
Deep neural networks can be fragile and sensitive to small input perturbations that might cause a significant change in the output. In this letter, we employ contraction theory to improve the robustness of neural ODEs (NODEs). A dynamical system is contractive if all solutions with different initial conditions converge to each other exponentially fast. As a consequence, perturbations in initial conditions become less and less relevant over time. Since in NODEs the input data corresponds to the initial condition of dynamical systems, we show contractivity can mitigate the effect of input perturbations. More precisely, inspired by NODEs with Hamiltonian dynamics, we propose a class of contractive Hamiltonian NODEs (CH-NODEs). By properly tuning a scalar parameter, CH-NODEs ensure contractivity by design and can be trained using standard backpropagation. Moreover, CH-NODEs enjoy built-in guarantees of non-exploding gradients, which ensure a well-posed training process. Finally, we demonstrate the robustness of CH-NODEs on the MNIST image classification problem with noisy test data.

24 April 2024 in IEEE Transactions on Circuits and Systems I: Regular Papers
Verified
Journal Article
Closed access
DLDonghao Liu
Donghao Liu
Zehui MaoZehui Mao
Zehui Mao
Bin JiangBin Jiang
Bin Jiang
LXLiang Xu
Liang Xu
This paper considers the distributed fault-tolerant consensus for heterogeneous nonlinear multiagent systems (HNMASs) with actuator faults and full-state constraints via adaptive dynamic programming (ADP). In order to handle the state constraint problem with multiple constraint types, a unified universal barrier function is introduced to convert the original constrained system into a non-constrained system. For the purpose of improving control efficiency and guaranteeing system reliability, a novel value function including fault estimations and control inputs is established. Considering the limitation of the computation and communication resources, a simplified ADP method incorporating the dynamic event-triggered strategy is developed to learn a distributed event-triggered fault-tolerant control policy. It is strictly proven that the HNMASs’ stability and the neural network weights’ convergence are guaranteed by the Lyapunov theory in the sense of uniform ultimate boundedness. Simulations are presented to verify the proposed control policy.