Lithology recognition is an essential part of reservoir parameter prediction. Compared to conventional algorithms, deep learning that needs a large amount of training data as support can extract features automatically. In the process of real data acquisition, the labeled data account for only a small portion due to high drilling cost, and it is difficult to achieve the data size required for deep learning training, resulting in a significant variance of the recognition model. For this shortage, in this paper, a semi-supervised algorithm based on Generative Adversarial Network(GAN) with Gini-regularization is proposed, called SGAN_G, which takes borehole-side data as labeled data and seismic data as unlabeled data. First, SGAN_G is trained by Adam(a method for stochastic optimization) algorithm and utilizes a discriminator to lithology recognition. And then, we add the entropy regularization to the initial loss function which enhances the convergence speed and accuracy of the model. Eventually, we propose a novel sampling approach which employs multiple sampling points of seismic data as inputs to use the stratum information implicitly. Through the experimental comparison with a variety of supervised approaches, we can see that SGAN_G can achieve higher prediction accuracy by using unlabeled data effectively.

Mobile edge computing (MEC) is viewed as a promising technology to address the challenges of intensive computing demands in hotspots (HSs). In this paper, we consider a unmanned aerial vehicle (UAV)-assisted backscattering MEC system. The UAVs can fly from parking aprons to HSs, providing energy to HSs via RF beamforming and collecting data from wireless users in HSs through backscattering. We aim to maximize the long-term utility of all HSs, subject to the stability of the HSs' energy queues. This problem is a joint optimization of the data offloading decision and contract design that should be adaptive to the users' random task demands and the time-varying wireless channel conditions. A deep reinforcement learning based contract incentive (DRLCI) strategy is proposed to solve this problem in two steps. Firstly, we use deep Q-network (DQN) algorithm to update the HSs' offloading decisions according to the changing network environment. Secondly, to motivate the UAVs to participate in resource sharing, a contract specific to each type of UAVs has been designed, utilizing Lagrangian multiplier method to approach the optimal contract. Simulation results show the feasibility and efficiency of the proposed strategy, demonstrating a better performance than the natural DQN and Double-DQN algorithms.

Feature selection or attribute reduction is an important data preprocessing technique for dimensionality reduction in machine learning and data mining. In this paper, a novel feature selection ensemble learning algorithm is proposed based on Tsallis entropy and Dempster–Shafer evidence theory (TSDS). First, an improved correlation criterion is used to obtain the relevant feature based on Tsallis entropy. A forward sequential approximate Markov blanket is then defined to eliminate the redundant feature. An ensemble learning is employed to achieve approximately optimal global feature selection, which can acquire the feature subsets from different perspectives. Finally, by fusing all the feature subsets, the improved evidence theory approach is utilized to gain the final feature subset. To verify the effectiveness of TSDS, nine datasets from two different domains are used in the experimental analysis. The experimental results demonstrate that the proposed algorithm can select feature subset more effectively and enhance the classification performance significantly.

Few-shot learning (FSL) is a challenging task in classifying new classes from few labelled examples. Many existing models embed class structural knowledge as prior knowledge to enhance FSL against data scarcity. However, they fall short of connecting the class structural knowledge with the limited visual information which plays a decisive role in FSL model performance. In this paper, we propose a unified FSL framework with multi-granularity knowledge fusion and decision-making (MGKFD) to overcome the limitation. We aim to simultaneously explore the visual information and structural knowledge, working in a mutual way to enhance FSL. On the one hand, we strongly connect global and local visual information with multigranularity class knowledge to explore intra-image and inter-class relationships, generating specific multi-granularity class representations with limited images. On the other hand, a weight fusion strategy is introduced to integrate multi-granularity knowledge and visual information to make the classification decision of FSL. It enables models to learn more effectively from limited labelled examples and allows generalization to new classes. Moreover, considering varying erroneous predictions, a hierarchical loss is established by structural knowledge to minimize the classification loss, where greater degree of misclassification is penalized more. Experimental results on three benchmark datasets show the advantages of MGKFD over several advanced models.