Descent algorithm for nonsmooth stochastic multiobjective optimization

Poirion, Fabrice; Mercier, Quentin; Désidéri, Jean‐Antoine

doi:10.1007/s10589-017-9921-x

Cited by 23 publications

(18 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These methods use multi-target Karush-Kuhn-Tucker (KKT) conditions [25] and find ways to reduce the descent direction of all targets. This approach extends the stochastic gradient descent proposed in [26,27]. Our work applies gradient-based multi-objective optimization to multi-task scheduling problem.…”

Section: Related Workmentioning

confidence: 94%

Multi-Task Scheduling Based on Classification in Mobile Edge Computing

et al. 2019

View full text Add to dashboard Cite

In this paper, a dynamic multi-task scheduling prototype is proposed to improve the limited resource utilization in the vehicular networks (VNET) assisted by mobile edge computing (MEC). To ensure quality of service (QoS) and meet the growing data demands, multi-task scheduling strategies should be specially constructed by considering vehicle mobility and hardware service constraints. We investigate the rational scheduling of multiple computing tasks to minimize the VNET loss. To avoid conflicts between tasks when the vehicle moves, we regard multi-task scheduling (MTS) as a multi-objective optimization (MOO) problem, and the whole goal is to find the Pareto optimal solution. Therefore, we develop some gradient-based multi-objective optimization algorithms. Those optimization algorithms are unable to deal with large-scale task scheduling because they become unscalable as the task number and gradient dimensions increase. We therefore further investigate an upper bound of the loss of multi-objective and prove that it can be optimized in an effective way. Moreover, we also reach the conclusion that, with practical assumptions, we can produce a Pareto optimal solution by upper bound optimization. Compared with the existing methods, the experimental results show that the accuracy is significantly improved.

show abstract

Section: Related Workmentioning

confidence: 94%

Multi-Task Scheduling Based on Classification in Mobile Edge Computing

et al. 2019

View full text Add to dashboard Cite

show abstract

“…For example, the gradient-based multi-objective optimization methods (Fliege & Svaiter, 2000;Désidéri, 2012) use multi-objective Karush-Kuhn-Tucker (KKT) conditions (Gordon & Tibshirani, 2012) to find a descent direction that decreases all objectives. This approach was then extended to the case of a stochastic gradient descent (Peitz & Dellnitz, 2016;Poirion, Mercier, & Désidéri, 2017). One key disadvantage of these approaches is that they scale poorly with the dimensionality of the gradients.…”

Section: Previous Workmentioning

confidence: 99%

Efficient Multi-objective Reinforcement Learning via Multiple-gradient Descent with Iteratively Discovered Weight-Vector Sets

Cao

Zhan²

2021

jair

View full text Add to dashboard Cite

Solving multi-objective optimization problems is important in various applications where users are interested in obtaining optimal policies subject to multiple (yet often conflicting) objectives. A typical approach to obtain the optimal policies is to first construct a loss function based on the scalarization of individual objectives and then derive optimal policies that minimize the scalarized loss function. Albeit simple and efficient, the typical approach provides no insights/mechanisms on the optimization of multiple objectives due to the lack of ability to quantify the inter-objective relationship. To address the issue, we propose to develop a new efficient gradient-based multi-objective reinforcement learning approach that seeks to iteratively uncover the quantitative inter-objective relationship via finding a minimum-norm point in the convex hull of the set of multiple policy gradients when the impact of one objective on others is unknown a priori. In particular, we first propose a new PAOLS algorithm that integrates pruning and approximate optimistic linear support algorithm to efficiently discover the weight-vector sets of multiple gradients that quantify the inter-objective relationship. Then we construct an actor and a multi-objective critic that can co-learn the policy and the multi-objective vector value function. Finally, the weight discovery process and the policy and vector value function learning process can be iteratively executed to yield stable weight-vector sets and policies. To validate the effectiveness of the proposed approach, we present a quantitative evaluation of the approach based on three case studies.

show abstract

“…The threat region has a range of 25km × 25km × 3km, the geographical location and types of threat sources are presented in Table 8. The coordinate of the starting point is (5, 5, 2)km, and the coordinate of the target point is (25,25,2)km, The numerical results of penetration route planning which is applied by different algorithms in the third scenario are depicted in Fig. 16.…”

Section: The Third Threat Scenariomentioning

confidence: 99%

“…The most common search algorithm can be divided into a random search algorithm and a deterministic search algorithm. On the one hand, deterministic search algorithms are adopted to solve UAV route planning, such as A-Star algorithm [20]- [22],D-Star algorithm [23]- [25], artificial potential field algorithm [26], [27],Dijkstra algorithm [28], [29], dynamic programming algorithm [30]- [32]. However, the A-Star algorithm, Dijkstra algorithm, and dynamic programming algorithm can only be applied for route planning in a small-scale static environment.…”

Section: Introductionmentioning

confidence: 99%

A Novel Real-Time Penetration Path Planning Algorithm for Stealth UAV in 3D Complex Dynamic Environment

Zhang

Dai

et al. 2020

IEEE Access

View full text Add to dashboard Cite

In recent years, stealth aircraft penetration path planning has been a significant research subject in the field of low altitude combat. However, previous works have mainly concentrated on the path planning for stealth unmanned aerial vehicle(UAV) in 2D static environment. In contrast, this paper addresses a novel real-time path planning algorithm for stealth UAV to realize the rapid penetration, which aims to devise a route penetration strategy based on the improved A-Star algorithm to address the problems of replanning for stealth UAV in 3D complex dynamic environment. The proposed method introduces the kinematic model of stealth UAV and detection performance of netted radar in the process of low altitude penetration. Additionally, POP-UP threats are adopted into three different threat scenarios, which is closer to the real combat environment. Moreover, further combined with the prediction technique and route planning algorithm, the multi-step search strategy is developed for stealth UAV to deal with POP-UP threats and complete the replanning of the route in different scenarios. Furthermore, the attitude angle information is integrated into the improved A-Star algorithm, which reflects the characteristics of the dynamic radar cross section(RCS) and conforms to the actual flight requirements for the stealth UAV. Finally, the improved A-Star algorithm, the sparse A-Star Search(SAS), and the dynamic A-Star algorithm(D-Star) are respectively adopted to address the problem of penetration route planning for stealth UAV in three different threat scenarios. Numerical simulations are performed to illustrate that the proposed approach can achieve rapid penetration route planning for stealth UAV in a dynamic threat scenario, and verify the validity of the improved A-Star algorithm which is compared to the other two algorithms. INDEX TERMS Stealth unmanned aerial vehicle, dynamic radar cross-section, netted radar, real-time path planning, A-Star algorithm.

show abstract

Descent algorithm for nonsmooth stochastic multiobjective optimization

Cited by 23 publications

References 13 publications

Multi-Task Scheduling Based on Classification in Mobile Edge Computing

Multi-Task Scheduling Based on Classification in Mobile Edge Computing

Efficient Multi-objective Reinforcement Learning via Multiple-gradient Descent with Iteratively Discovered Weight-Vector Sets

A Novel Real-Time Penetration Path Planning Algorithm for Stealth UAV in 3D Complex Dynamic Environment

Contact Info

Product

Resources

About