Narrow Results

Designing an efficient fingerprint recognition technique is an ill-posed problem. Recently, many researchers have utilized machine learning techniques to improve the fingerprint recognition rate. The random forest (RF) is found to be one of the extensively utilized machine learning techniques for fingerprint recognition. Although it provides good recognition results at significant computational speed, still there is room for improvement. RF is not so-effective for high-dimensional features and also when features contain both discrete and continuous values at the same time. Therefore, in this paper, a novel similarity measure-based random forest (NRF) is proposed. The proposed technique, initially, computes both mutual information and conditional entropy. Thereafter, it uses three designed if-then rules to obtain final information measure. Additionally, to obtain feature set for fingerprint dataset, dual-tree complex wavelet transform is used to evaluate complex detail coefficients. Thereafter, ring project is considered to compute significant moments from these complex detail coefficients. Finally, information gain-based feature selection technique is used to select potential features. To prevent over-fitting, 20-fold cross validation is also used. Extensive experiments are considered to evaluate the effectiveness of the proposed technique. The comparative analyses reveal that the proposed technique outperforms the existing techniques in terms of accuracy, f-measure, sensitivity, specificity, kappa statistics and computational speed.

Recently, deep convolutional neural networks have resulted in noticeable improvements in image classification and have been used to transfer artistic style of images. Gatys et al. proposed the use of a learned Convolutional Neural Network (CNN) architecture VGG to transfer image style, but problems occur during the back propagation process because there is a heavy computational load. This paper solves these problems, including the simplification of the computation of chains of derivatives, accelerating the computation of adjustments, and efficiently choosing weights for different energy functions. The experimental results show that the proposed solutions improve the computational efficiency and render the adjustment of weights for energy functions easier.

We improve the least squares (LS) method for building models of a nonlinear dynamical system given finite time series which are contaminated by observational noise. When the noise level is low, the LS method gives good estimates for the parameters, however, the models selected as the best by information criteria often tend to be over-parameterized or even degenerate. We observe that the correct model is not selected as the best model despite belonging to the chosen model class. To overcome this, we propose a simple but very effective idea to use the LS method more appropriately. We apply the method for model selection. Numerical studies indicate that the method can be used to apply information criteria more effectively, and generally avoid over-fitting and model degeneracy.

Benford’s Law is an interesting and unexpected empirical phenomenon — that if we take a large list of number from real data, the first digits of these numbers follow a certain non-uniform distribution. This law is actively used in economics and finance to check that the data in financial reports are real — and not improperly modified by the reporting company. The first challenge is that the cheaters know about it, and make sure that their modified data satisfies Benford’s law. The second challenge related to this law is that lately, another application of this law has been discovered — namely, an application to deep learning, one of the most effective and most promising machine learning techniques. It turned out that the neurons’ weights obey this law only at the difficult-to-detect stage when the fitting is optimal – and when further attempts attempt to fit will lead to the undesirable over-fitting. In this paper, we provide a possible solution to both challenges: we show how to use this law to make financial cheating practically impossible, and we provide qualitative explanation for the effectiveness of Benford’s Law in machine learning.

In the research of deep learning, an Unsupervised deep convolution-generated Generated Adversarial Network (UGAN) usually needs a large number of data samples to train. However, when faced with some small samples, the performance of the algorithm is often degraded due to over-fitting. Combined with specially designed data enhancement methods, a generated adversarial network optimization algorithm based on adaptive data augmentation (AdauGAN) is proposed. The adaptive data augmentation module is added before the discriminant network, and a spatial transformation is carried out simultaneously at the probability distribution level of generated data and real data. To alleviate the over-fitting phenomenon in the training process, the current enhancement intensity is adjusted adaptively after the over-fitting occurs. The proposed algorithm is verified on SVHN, CelebA and CIFAR-10 data sets. The Frechet Inception Distance (FID) values of AdauGAN achieve 22.10, 23.94, 34.87, respectively, which is close to or even higher than the training results of Deep Convolution Generated Adversarial Network (DCGAN) under all data. Extensive experiment results show that the proposed Adaugan has an excellent performance in small samples. Besides, in some cases, it can catch up with the large sample results of existing algorithms.