Annals of Data Science最新文献

Effective value measurement and pricing methods can greatly promote the healthy development of data sharing, exchange and reuse. However, the uncertainty of data value and neglect of interactivity lead to information asymmetry in the transaction process. A perfect pricing system and well-designed data trading market (hereafter called data market) can widely promote data transactions. We take the three-agents data market as an example to construct a sound data trading process. The data owner who provides data records, the model buyer who is interested in buying machine learning (ML) model instances, and the data broker who interacts between the data owner and the model buyer. Based on the characteristics of data market, like truthfulness, revenue maximization, version control, fairness and non-arbitrage, we propose a data pricing methods based on different model versions. Firstly, we utilize market research and construct a revenue maximization (RM) problem to price the different versions of ML models and solve it with the RM-ILP process. However, the RM model based on market research has two major problems: one is that the model buyer has no incentive to tell the truth, that is, the model buyer will lie in the market research to obtain a lower model price; the other is that it asks the data broker to release version menu in advance, resulting in an inefficient operation of the data market. In view of the defects of the RM transaction model, we propose a model buyers behavior analysis, establish the revenue maximization function based on different data versions to establish a bi-level linear programming model. We further add the incentive compatibility constraint and the individual rationality constraint, taking the utility of the model buyer and the revenue of the data broker into account. This reflects the consumer driven model in the data transaction mode. Finally, the RM-BLP process is proposed to transform RM problem into an equivalent single-level integer programming problem and we solve it with the “Gurobi” solver. The validity of the model is verified by experiments.

This paper tackle the challenges associated with low recognition accuracy and the detection of occlusions when identifying long-range and diminutive targets (such as UAVs). We introduce a sophisticated detection framework named UAV-YOLOv5, which amalgamates the strengths of Swin Transformer V2 and YOLOv5. Firstly, we introduce Focal-EIOU, a refinement of the K-means algorithm tailored to generate anchor boxes better suited for the current dataset, thereby improving detection performance. Second, the convolutional and pooling layers in the network with step size greater than 1 are replaced to prevent information loss during feature extraction. Then, the Swin Transformer V2 module is introduced in the Neck to improve the accuracy of the model, and the BiFormer module is introduced to improve the ability of the model to acquire global and local feature information at the same time. In addition, BiFPN is introduced to replace the original FPN structure so that the network can acquire richer semantic information and fuse features across scales more effectively. Lastly, a small target detection head is appended to the existing architecture, augmenting the model’s proficiency in detecting smaller targets with heightened precision. Furthermore, various experiments are conducted on the comprehensive dataset to verify the effectiveness of UAV-YOLOv5, achieving an average accuracy of 87%. Compared with YOLOv5, the mAP of UAV-YOLOv5 is improved by 8.5%, which verifies that it has high-precision long-range small-target UAV optoelectronic detection capability.

The measurement of network topology through various spatial topological indices like Alpha, Beta and Gamma are widely used for spatial data analysis. However, explaining the classification of the network topology of a city based on Alpha, Beta and Gamma indices is not conclusive, as the result of individual indices are different. To address an efficient classification of network topology, a Modified Synthetic Indicator (MSI) has been proposed and criticised over existing synthetic indicators based on the Composite Weighted Connectivity Index (CWCI), the linear combination of Alpha, Beta and Gamma indices. Application of the proposed MSI in micro-level (ward level) classification of network topology i.e., road network connectivity, has been verified in Agartala City and calibrates the efficiency of CWCI over Alpha, Beta and Gamma indices. The study reveals that the proposed CWCI is more robust than any individual graph-theoretic measure.

The image harmonization task endeavors to adjust foreground information within an image synthesis process to achieve visual consistency by leveraging background information. In academic research, this task conventionally involves the utilization of simple synthesized images and matching masks as inputs. However, obtaining precise masks for image harmonization in practical applications poses a significant challenge, thereby creating a notable disparity between research findings and real-world applicability. To mitigate this disparity, we propose a redefinition of the image harmonization task as “Unified Image Harmonization,” where the input comprises only a single image, thereby enhancing its applicability in real-world scenarios. To address this challenge, we have developed a novel framework. Within this framework, we initially employ inharmonious region localization to detect the mask, which is subsequently utilized for harmonization tasks. The pivotal aspect of the harmonization process lies in normalization, which is accountable for information transfer. Nonetheless, the current background-to-foreground information transfer and guidance mechanisms are limited by single-layer guidance, thereby constraining their effectiveness. To overcome this limitation, we introduce Region Augmented Attention Normalization (RA2N), which enhances the attention mechanism for foreground feature alignment, consequently leading to improved alignment and transfer capabilities. Through qualitative and quantitative comparisons on the iHarmony4 dataset, our model exhibits exceptional performance not only in unified image harmonization but also in conventional image harmonization tasks.

In the field of bioinformatics, changes in protein functionality are mainly influenced by protein mutations. Accurately predicting these functional changes can enhance our understanding of evolutionary mechanisms, promote developments in protein engineering-related fields, and accelerate progress in medical research. In this study, we introduced two different models: one based on bidirectional long short-term memory (BiLSTM), and the other based on self-attention. These models were integrated using a weighted fusion method to predict protein functional changes associated with mutation sites. The findings indicate that the model's predictive precision matches that of the current model, along with its capacity for generalization. Furthermore, the ensemble model surpasses the performance of the single models, highlighting the value of utilizing their synergistic capabilities. This finding may improve the accuracy of predicting protein functional changes associated with mutations and has potential applications in protein engineering and drug research. We evaluated the efficacy of our models under different scenarios by comparing the predicted results of protein functional changes across various numbers of mutation sites. As the number of mutation sites increases, the prediction accuracy decreases significantly, highlighting the inherent limitations of these models in handling cases involving more mutation sites.

Accurately predicting students’ performance plays a crucial role in achieving the intellectualization of courses. This paper studied intelligent courses in English education based on neural networks and designed a firefly algorithm-back propagation neural network (FA-BPNN) method. The correlation between various features and final grades was calculated using the students’ online learning data. Features with higher correlation were selected as the input for the FA-BPNN algorithm to estimate the final score that students achieved in the “College English” course. It was found that the training time of the FA-BPNN algorithm was 3.42 s, the root-mean-square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) values of the FA-BPNN algorithm were 0.986, 0.622, and 0.205, respectively. They were lower than those of the BPNN, genetic algorithm (GA)-BPNN, and particle swarm optimization (PSO)-BPNN algorithms, as well as the adaptive neuro-fuzzy inference system approach. The results indicated the efficacy of the FA for optimizing the parameters of the BPNN algorithm. The comparison between the predicted results and actual values suggested that the average error of the FA-BPNN algorithm was only 0.5, which was the smallest. The experimental results demonstrate the reliability of the FA-BPNN algorithm for performance prediction and its practical application feasibility.

Recently, ranked set samples schemes have become quite popular in reliability analysis and life-testing problems. Based on ordered ranked set sample, the Bayesian estimators and credible intervals for the entropy of the Rayleigh model are studied and compared with the corresponding estimators based on simple random sampling. These Bayes estimators for entropy are developed and computed with various loss functions, such as square error, linear-exponential, Al-Bayyati, and general entropy loss functions. A comparison study for various estimates of entropy based on mean squared error is done. A real-life data set and simulation are applied to illustrate our procedures.