查看更多>>摘要：Knowledge distillation is a highly effective method for transferring knowledge from a cum-bersome teacher network to a lightweight student network. However, teacher networks are not always available. An alternative method called online knowledge distillation, which applies a group of peer networks to learn collaboratively with each other, has been pro -posed previously. In this study, we revisit online knowledge distillation and find that the existing training strategy limits the diversity among peer networks. Thus, online knowledge distillation cannot achieve its full potential. To address this, a novel online knowledge dis-tillation with elastic peer (KDEP) strategy is introduced here. The entire training process is divided into two phases by KDEP. In each phase, a specific training strategy is applied to adjust the diversity to an appropriate degree. Extensive experiments have been conducted on individual benchmarks, including CIFAR-100, CINIC-10, Tiny ImageNet, and Caltech-UCSD Birds. The results demonstrate the superiority of KDEP. For example, when the peer networks are ShuffleNetV2-1.0 and ShuffleNetV2-0.5, the target peer network ShuffleNetV2-0.5 achieves 57:00% top-1 accuracy on Tiny ImageNet via KDEP. (c) 2021 Elsevier Inc. All rights reserved.
查看更多>>摘要：Morphological perceptrons (MPs) can be characterized as feedforward morphological neural networks (MNNs) with applications in classification and regression. The neuronal aggregation functions of current MP versions are drawn from gray-scale mathematical morphology (MM) that can be described in terms of matrix products in a lattice algebra called minimax algebra. Specifically, MPs have components each of which computes a pair-wise infimum of an erosion and an anti-dilation that can be expressed in terms of products of matrices with entries in a complete l-group extension. In this paper, we use the novel concept of an interval descriptor and an n-ary aggregation function on a bounded poset in order to generalize existing gray-scale and fuzzy morphological components (MCs) of morphological and hybrid morphological/linear perceptrons (HMLPs). In addition, we present several other examples of generalized morphological components (GMCs) that can and will be incorporated as computational units into shallow and deep artificial neural networks. (c) 2021 Elsevier Inc. All rights reserved.
查看更多>>摘要：Few studies of conflict analysis have analyzed conflict situations from the perspective of the consistency degree of cliques. Quantifying the conflict degree is instructive for weakening or resolving conflicts. In this paper, we propose the improved Pawlak model of conflict analysis, a common consistency measurement is introduced to quantify the consistency degree of cliques, which explains the internal causes of conflict situation from a new perspective. In addition, the proposed consistency measurement reflects the fit between feasible strategy and clique, which inspires us to formulate the criteria for selecting feasible strategies in a three-valued situation table. Moreover, considering that many-valued ratings are more powerful and effective to describe vague information, a novel model of fuzzy conflict analysis is established with a fuzzy conflict function and a fuzzy consistency measurement in a fuzzy situation table, which could degenerate into the improved Pawlak model of conflict analysis. The criteria of selecting feasible strategies is illustrated by an NBA labor negotiation. Finally, the performance of proposed models is compared with other advanced models. (c) 2021 Elsevier Inc. All rights reserved.
查看更多>>摘要：Task scheduling is an important research direction in cloud computing. The current research on task scheduling considers mainly the design of scheduling strategies and algorithms and rarely gives attention to the influences of uncertain factors, such as the network bandwidth and millions of instructions per second (MIPS), on the scheduling process. The network bandwidth and MIPS directly affect the performance of a virtual machine (VM), which further influences the scheduling performance. In this paper, uncertain factors are transformed into interval parameters. The make-span, scheduling cost, load balance, and task completion rate are simultaneously considered in the scheduling process. Then, an interval many-objective cloud task scheduling optimization (I-MCTSO) model is designed to simulate real cloud computing task scheduling. To implement this model, an interval many-objective evolutionary algorithm (InMaOEA) is proposed. An interval credibility strategy is employed to improve the convergence performance. The hyper-volume and degree of overlap based on the interval congestion distance strategy are used to increase the population diversity. Simulation results demonstrate the effectiveness and superior performance of InMaOEA in comparision with other algorithms. The proposed approaches can provide decision-makers with an efficient allocation plan for cloud task scheduling. (c) 2021 Elsevier Inc. All rights reserved.
查看更多>>摘要：Existing block-based progressive visual cryptography schemes (BPVCS) that prevent cheating are limited to (2, n) and (k, n) thresholds. They are unsuitable for implementing complex sharing techniques in the actual world. A cheating immune BPVCS for general access structure (GAS) is proposed in this paper. To find the parameters that make up the proposed scheme, a simplified multiple assignment procedure based on integer linear programming is introduced. The secret parts can be decrypted by the members of a qualified set. Participants in a forbidden set, on the other hand, are not allowed to know about the secret. The proposed scheme includes cheat-prevention and progressive recovery capabilities. In addition, to generate meaningful shadows for BPVCS, a modified extended VCS (EVCS) is proposed. Our modified EVCS can solve the cover-interference problem in conventional BPVCS. The effectiveness and benefits of the proposed BPVCS are supported through the experimental findings and comparisons. (c) 2021 Elsevier Inc. All rights reserved.
Zamfirache, Iuliu AlexandruPrecup, Radu-EmilRoman, Raul-CristianPetriu, Emil M....
22页
查看更多>>摘要：This paper presents a novel Reinforcement Learning (RL)-based control approach that uses a combination of a Deep Q-Learning (DQL) algorithm and a metaheuristic Gravitational Search Algorithm (GSA). The GSA is employed to initialize the weights and the biases of the Neural Network (NN) involved in DQL in order to avoid the instability, which is the main drawback of the traditional randomly initialized NNs. The quality of a particular set of weights and biases is measured at each iteration of the GSA-based initialization using a fitness function aiming to achieve the predefined optimal control or learning objective. The data generated during the RL process is used in training a NN-based controller that will be able to autonomously achieve the optimal reference tracking control objective. The proposed approach is compared with other similar techniques which use different algorithms in the initialization step, namely the traditional random algorithm, the Grey Wolf Optimizer algorithm, and the Particle Swarm Optimization algorithm. The NN-based controllers based on each of these techniques are compared using performance indices specific to optimal control as settling time, rise time, peak time, overshoot, and minimum cost function value. Real-time experiments are conducted in order to validate and test the proposed new approach in the framework of the optimal reference tracking control of a nonlinear position servo system. The experimental results show the superiority of this approach versus the other three competing approaches. (c) 2021 Elsevier Inc. All rights reserved.
查看更多>>摘要：This paper proposes an adaptive memetic differential evolution with multi-niche sampling and neighborhood crossover strategies for global optimization. In the proposed algorithm, a multi-niche sampling strategy is designed to sample a subpopulation for evolution at each generation. In this strategy, the entire population is firstly divided into multiple niches by employing a certain niching strategy at each generation. A subpopulation is then dynamically sampled from the resulting niches such that supporting a diverse search at the early stage of evolution while an intensive search towards the end of evolution. The above strategy will be further coupled with a neighborhood crossover, which is devised to encourage high potential solutions for exploitation while low potential solutions for exploration, thus appropriately searching the solution space. Additionally, an adaptive local search (ALS) scheme along with an adaptive elimination operation (AEO) have been designed. The ALS aims to appropriately fine-tune promising solutions in the sampled sub population while the AEO tends to adaptively eliminate unpromising individuals in the population during evolution. The performance of the proposed algorithm has been evaluated on CEC'2015 benchmark functions and compared with related methods. Experimental results show that our algorithm can achieve a superior performance and outperform related methods. The results also confirm the significance of devised strategies in the proposed algorithm. (c) 2021 Elsevier Inc. All rights reserved.
Saeed, R. A.Recupero, Diego ReforgiatoRemagnino, Paolo
22页
查看更多>>摘要：We propose a new method for simulating pedestrian crowd movement in a virtual environment. A crowd consists of groups of different number of people with different attributes such as gender, age, position, velocity, and energy. Each group has its own intention used to generate a trajectory for each pedestrian navigating in the virtual environment. Additionally, an agent -based model is introduced to simulate pedestrian behaviours in the groups, where various steering behaviours are introduced and combined into a single steering force to allow pedestrians in each group to walk toward their destination point. Based on the proposed method, every single pedestrian in each group can continuously adjust their attributes. Moreover, pedestrians optimize their path independently toward the desired goals, while avoiding obstacles and other pedestrians in the scene. This method takes into account the safety-space around each pedestrian to avoid collisions among pedestrians. The proposed method was implemented for several simulation scenarios under various conditions for a wide range of different parameters. Statistical analysis is carried out to evaluate the performance of the proposed method for simulating the crowd movement in the virtual environment. Results indicate that our method can generate each pedestrian's trajectories in each group independently to reach several goal points within a reasonable computational time. Moreover, the obtained results reveal that the mean value of the computational time is not increased significantly with the increasing of the number of pedestrians in the crowd. (c) 2021 Elsevier Inc. All rights reserved.
查看更多>>摘要：Classification is one of the fundamental supervised learning tasks which learns classifiers from the given training data and related labels. The quality of labels is important in classification tasks. However, in many real-world scenarios, data annotation is often corrupted, especially when the annotation process is done by humans. Vaguely labeled data is one of the common problems caused by limited domain knowledge or partial data observation. In this paper, a novel method is proposed to classify vaguely labeled data based on evidential fusion. Vaguely labeled data are divided into several small data groups by the proposed valid label-set cover assignment algorithm. Evidence theory is applied to vaguely labeled data classification by regarding each base classifier on a small data group as one piece of evidence. This gives the chance of classifying unseen precise labeled data from related vague labels. Note that our approach is not restricted to any specific classifiers. It can be generalized to any off-the-shelf classification methods with probabilistic outputs. Finally, experiments are conducted on both synthetic data and real-world data with different base classifiers. Experimental results show that the proposed method achieves superior performance against compared methods. (c) 2021 Elsevier Inc. All rights reserved.
查看更多>>摘要：Crowdsourcing makes it much faster and cheaper to obtain labels for a large amount of data used in supervised learning. In the crowdsourcing scenario, an integrated label is inferred from a multiple noisy label set for each instance using ground truth inference algorithms, which is called label integration. However, a certain level of label noise remains in the integrated dataset, which degrades the performance of the models trained on it. To the best of our knowledge, existing label noise correction algorithms only use the original attribute space and do not use the information contained in the multiple noisy label sets for building models. To solve these problems, we propose a novel integrated label noise correction algorithm called co-training-based noise correction (CTNC). In CTNC, the weight is first calculated from the information provided by the multiple noisy label set for each instance. Subsequently, a label noise filter is used to identify noisy instances; a clean set and a noisy set are thus obtained. Another attribute view of each instance in both the clean and noisy sets is then generated by the classifiers trained on the original attribute view of the clean set. Finally, a co-training framework is used to train two classifiers to relabel the integrated instances. The performance on 34 simulated datasets and 2 real-world datasets demonstrates that our proposed CTNC outperforms all state-of-the-art label noise correction algorithms used for comparison. (c) 2021 Elsevier Inc. All rights reserved.
NSTL
Elsevier