Search

Adaptive algorithms are widely used in deep learning because of their fast convergence. Among them, Adam is the most widely used algorithm. However, studies have shown that Adam's generalization ability is weak. AdaX is a variant of Adam, which introduces a novel second-order momentum, modifies the second-order moment of Adam, and has good generalization ability. However, these algorithms may fail to converge due to instability and extreme learning rates during training. In this paper, we propose a new adaptive and momental bound algorithm, called AdaXod, which characterizes of exponentially averaging the learning rate and is particularly useful for training deep neural networks. By setting an adaptively limited learning rate in the AdaX algorithm, the resultant AdaXod can effectively eliminate the problem of excessive learning rate in the later stage of neural networks training and thus results in stable training. We conduct extensive experiments on different datasets and verify the advantages of the AdaXod algorithm by comparing with other advanced adaptive optimization algorithms. AdaXod eliminates large learning rates during neural networks training and outperforms other optimizers, especially for some neural networks with complex structures, such as DenseNet. [ABSTRACT FROM AUTHOR]

Principles of typical motion planning algorithms are investigated and analyzed in this paper. These algorithms include traditional planning algorithms, classical machine learning algorithms, optimal value reinforcement learning, and policy gradient reinforcement learning. Traditional planning algorithms investigated include graph search algorithms, sampling-based algorithms, interpolating curve algorithms, and reaction-based algorithms. Classical machine learning algorithms include multiclass support vector machine, long short-term memory, Monte-Carlo tree search and convolutional neural network. Optimal value reinforcement learning algorithms include Q learning, deep Q-learning network, double deep Q-learning network, dueling deep Q-learning network. Policy gradient algorithms include policy gradient method, actor-critic algorithm, asynchronous advantage actor-critic, advantage actor-critic, deterministic policy gradient, deep deterministic policy gradient, trust region policy optimization and proximal policy optimization. New general criteria are also introduced to evaluate the performance and application of motion planning algorithms by analytical comparisons. The convergence speed and stability of optimal value and policy gradient algorithms are specially analyzed. Future directions are presented analytically according to principles and analytical comparisons of motion planning algorithms. This paper provides researchers with a clear and comprehensive understanding about advantages, disadvantages, relationships, and future of motion planning algorithms in robots, and paves ways for better motion planning algorithms in academia, engineering, and manufacturing. [ABSTRACT FROM AUTHOR]

To meet the demands of massive connections in the Internet-of-vehicle communications, non-orthogonal multiple access (NOMA) is utilized in the local wireless networks. In NOMA technique, various optimization methods have been proposed to provide optimal resource allocation, but they are limited by computational complexity. Recently, the deep reinforcement learning network is utilized for resource optimization in NOMA system, where a uniform sampled experience replay algorithm is used to reduce the correlation between samples. However, the uniform sampling ignores the importance of sample. To this point, this paper proposes a joint prioritized DQN user grouping and DDPG power allocation algorithm to maximize the system sum rate. At the user grouping stage, a prioritized sampling method based on TD-error (temporal-difference error) is proposed. At the power allocation stage, to deal with the problem that DQN cannot process continuous tasks and needs to quantify power into discrete form, a DDPG network is utilized. Simulation results show that the proposed algorithm with prioritized sampling can increase the learning rate and perform a more stable training process. Compared with the previous DQN algorithm, the proposed method improves the sum rate of the system by 2% and reaches 94% and 93% of the exhaustive search algorithm and optimal iterative power optimization algorithm, respectively. Although the sum rate is improved by only 2%, the computational complexity is reduced by 43% and 64% compared to the exhaustive search algorithm and the optimal iterative power optimization algorithm, respectively. [ABSTRACT FROM AUTHOR]

Robotic motion planning in dense and dynamic indoor scenarios constantly challenges the researchers because of the motion unpredictability of obstacles. Recent progress in reinforcement learning enables robots to better cope with the dense and unpredictable obstacles by encoding complex features of the robot and obstacles into the encoders like the long-short term memory (LSTM). Then these features are learned by the robot using reinforcement learning algorithms, such as the deep Q network and asynchronous advantage actor critic algorithm. However, existing methods depend heavily on expert experiences to enhance the convergence speed of the networks by initializing them via imitation learning. Moreover, those approaches based on LSTM to encode the obstacle features are not always efficient and robust enough, therefore sometimes causing the network overfitting in training. This paper focuses on the advantage actor critic algorithm and introduces an attention-based actor critic algorithm with experience replay algorithm to improve the performance of existing algorithm from two perspectives. First, LSTM encoder is replaced by a robust encoder attention weight to better interpret the complex features of the robot and obstacles. Second, the robot learns from its past prioritized experiences to initialize the networks of the advantage actor-critic algorithm. This is achieved by applying the prioritized experience replay method, which makes the best of past useful experiences to improve the convergence speed. As results, the network based on our algorithm takes only around 15% and 30% experiences to get rid of the early-stage training without the expert experiences in cases with five and ten obstacles, respectively. Then it converges faster to a better reward with less experiences (near 45% and 65% of experiences in cases with ten and five obstacles respectively) when comparing with the baseline LSTM-based advantage actor critic algorithm. Our source code is freely available at the GitHub (https://github.com/CHUENGMINCHOU/AW-PER-A2C). [ABSTRACT FROM AUTHOR]

This paper investigates a computing offloading policy and the allocation of computational resource for multiple user equipments (UEs) in device-to-device (D2D)-aided fog radio access networks (F-RANs). Concerning the dynamically changing wireless environment where the channel state information (CSI) is difficult to predict and know exactly, we formulate the problem of task offloading and resource optimization as a mixed-integer nonlinear programming problem to maximize the total utility of all UEs. Concerning the non-convex property of the formulated problem, we decouple the original problem into two phases to solve. Firstly, a centralized deep reinforcement learning (DRL) algorithm called dueling deep Q-network (DDQN) is utilized to obtain the most suitable offloading mode for each UE. Particularly, to reduce the complexity of the proposed offloading scheme-based DDQN algorithm, a pre-processing procedure is adopted. Then, a distributed deep Q-network (DQN) algorithm based on the training result of the DDQN algorithm is further proposed to allocate the appropriate computational resource for each UE. Combining these two phases, the optimal offloading policy and resource allocation for each UE are finally achieved. Simulation results demonstrate the performance gains of the proposed scheme compared with other existing baseline schemes. [ABSTRACT FROM AUTHOR]

Reinforcement learning based on the deep neural network has attracted much attention and has been widely used in real-world applications. However, the black-box property limits its usage from applying in high-stake areas, such as manufacture and healthcare. To deal with this problem, some researchers resort to the interpretable control policy generation algorithm. The basic idea is to use an interpretable model, such as tree-based genetic programming, to extract policy from other black box modes, such as neural networks. Following this idea, in this paper, we try yet another form of the genetic programming technique, evolutionary feature synthesis, to extract control policy from the neural network. We also propose an evolutionary method to optimize the operator set of the control policy for each specific problem automatically. Moreover, a policy simplification strategy is also introduced. We conduct experiments on four reinforcement learning environments. The experiment results reveal that evolutionary feature synthesis can achieve better performance than tree-based genetic programming to extract policy from the neural network with comparable interpretability. [ABSTRACT FROM AUTHOR]

Virtual network embedding (VNE), as the key challenge of network resource management technology, lies in the contradiction between online embedding decision and pursuing long-term average revenue goals. Most of the previous work ignored the dynamics in Virtual Network (VN) modeling, or could not automatically detect the complex and time-varying network state to provide a reasonable network embedding scheme. In view of this, we model a network embedding framework where the topology and resource allocation change dynamically with the number of network users and workload, and then introduce a deep reinforcement learning method to solve the VNE problem. Further, a dynamic virtual network embedding algorithm based on Deep Reinforcement Learning (DRL), named DVNE-DRL, is proposed. In DVNE-DRL, VNE is modeled as a Markov Decision Process (MDP), and then deep learning is introduced to perceive the current network state through historical data and embedded knowledge, while utilizing reinforcement learning decision-making capabilities to implement the network embedding process. In addition, we improve the method of feature extraction and matrix optimization, and consider the characteristics of virtual network and physical network together to alleviate the problem of redundancy and slow convergence. The simulation results show that compared with the existing advanced algorithms, the acceptance rate and average revenue of DVNE-DRL are increased by about 25% and 35%, respectively. [ABSTRACT FROM AUTHOR]

The task of intercepting a target moving along a rectilinear or circular trajectory by a Dubins' car is formulated as a problem of time-optimal control with an arbitrary direction of the car's velocity at the time of interception. To solve this problem and to synthesize interception trajectories, neural network methods of unsupervised learning based on the Deep Deterministic Policy Gradient algorithm are used. The analysis of the obtained control laws and interception trajectories is carried out in comparison with the analytical solutions of the interception problem. Mathematical modeling of the target motion parameters, which the neural network had not previously seen during training, is carried out. Model experiments are conducted to test the stability of the neural solution. The effectiveness of using neural network methods for the synthesis of interception trajectories for given classes of target movements is shown. [ABSTRACT FROM AUTHOR]

Aiming at the problem of vehicle model tracking error and overdependence in traditional path planning of intelligent driving vehicles, a path planning method of intelligent driving vehicles based on deep reinforcement learning is proposed. Firstly, the abstract model of real environment is extracted. The model uses deep reinforcement learning end-to-end strategy (DRL-ETE) and vehicle dynamics model to train the reinforcement learning model which approaches the optimal intelligent driving. Secondly, the real scene problem is transferred to the virtual abstract model through the model transfer strategy, and the control and trajectory sequences are calculated according to the trained deep reinforcement learning model in the environment. Finally, the optimal trajectory sequence is selected according to the evaluation function in real environment. Because the storage mode of experience playback mechanism of Deep Q-Network algorithm is FIFO, and the sampling mode of later playback training is average sampling, the efficiency of experience playback is low. These two problems lead to the slow process of intelligent driving vehicle to target and route finding. And because of greedy strategy, the information of exploration map is incomplete, and IDQNPER algorithm model is proposed. When storing samples, the samples are given weight, and sent to the network in priority order for sample training. Meanwhile, the importance data sequence is retained in the experience playback cache area, and the sequence with high similarity is removed. The total reward value is about 10% higher than the reward value of original Deep Q-network, which proves that the accuracy of intelligent driving vehicles tends to target points is higher. In order to further realize the autonomous decision-making of intelligent driving vehicles and solve the problem of relying too much on map information in the traditional human planning framework, an end-to-end path planning method is proposed based on the depth reinforcement learning theory, which maps the action instructions directly from the sensor information and then issues them to the intelligent driving vehicles. Firstly, CNN and LSTM are used to process radar and camera information. By comparing the advantages of DQ, Double DQN, Dueling DQN and PER algorithm, IDQNPER algorithm is used to train the automatic path planning of intelligent driving vehicles. Finally, the simulation and verification experiments are carried out in the static obstacle environment. The test results show that IDQNPER algorithm is adaptable to intelligent vehicles in different environments. The method can deal with the continuous input state and generate the continuous control sequence of the corner control, which can reduce the lateral tracking error. At the same time, the generalization performance of the model can be improved and the overdependence problem can be reduced by experience playback. [ABSTRACT FROM AUTHOR]

Multi-objective reinforcement learning (MORL) algorithms aim to approximate the Pareto frontier uniformly in multi-objective decision making problems. In the scenario of deep reinforcement learning (RL), gradient-based methods are often adopted to learn deep policies/value functions due to the fast convergence speed, while pure gradient-based methods can not guarantee a uniformly approximated Pareto frontier. On the other side, evolution strategies straightly manipulate in the solution space to achieve a well-distributed Pareto frontier, but applying evolution strategies to optimize deep networks is still a challenging topic. To leverage the advantages of both kinds of methods, we propose a two-stage MORL framework combining a gradient-based method and an evolution strategy. First, an efficient multi-policy soft actor-critic algorithm is proposed to learn multiple policies collaboratively. The lower layers of all policy networks are shared. The first-stage learning can be regarded as representation learning. Secondly, the multi-objective covariance matrix adaptation evolution strategy (MO-CMA-ES) is applied to fine-tune policy-independent parameters to approach a dense and uniform estimation of the Pareto frontier. Experimental results on three benchmarks (Deep Sea Treasure, Adaptive Streaming, and Super Mario Bros) show the superiority of the proposed method. [ABSTRACT FROM AUTHOR]

The article, by Mengshan Yu on "Location-based and time-aware service recommendation in mobile edge computing" addresses multidimensional inverse user similarity recommendation algorithm to solve the cold start problem for service recommendation system in edge computing. The article, by Tianba Chen on "M-DRL: Deep Reinforcement Learning based Coflow Traffic Scheduler with MLFQ threshold adaption" proposes M-DRL, a deep reinforcement learning approach to optimize coflow scheduling for data-parallel clusters. The aim of this special issue is to provide an introduction to the burgeoning topic of computer system research for high-performance computers and big-data systems. [Extracted from the article]