Manuel Vargas, VictorAntonio Gutierrez, PedroHervas-Martinez, Cesar
10页
查看更多>>摘要：Currently, the use of deep learning for solving ordinal classification problems, where categories follow a natural order, has not received much attention. In this paper, we propose an unimodal regularisation based on the beta distribution applied to the cross-entropy loss. This regularisation encourages the dis-tribution of the labels to be a soft unimodal distribution, more appropriate for ordinal problems. Given that the beta distribution has two parameters that must be adjusted, a method to automatically deter-mine them is proposed. The regularised loss function is used to train a deep neural network model with an ordinal scheme in the output layer. The results obtained are statistically analysed and show that the combination of these methods increases the performance in ordinal problems. Moreover, the proposed beta distribution performs better than other distributions proposed in previous works, achieving also a reduced computational cost. (c) 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ )
查看更多>>摘要：Deep Neural Networks (DNNs) are demonstrated to be vulnerable to adversarial examples, which are crafted by adding adversarial perturbations to the legitimate examples. To address this issue, some defense methods have been proposed. Among them, the adversarial training (AT) is a popular method to improve the robustness of DNNs. However, theory analysis has shown that in the adversarial training framework, the improvement of the robustness will lead to a decline of standard accuracy. In this paper, we propose a modularized defense framework, namely Adversarial Domain Adaptation to Defense ((AD)(2)). Different from all adversarial training methods, (AD)(2) detects adversarial example using a generative algorithm and applies the adversarial domain adaptation method to remove adversarial perturbation. Experimental results show that (AD)(2) is effective to remove the adversarial perturbation and mitigate the odds between the robustness and standard accuracy for DNNs. (C) 2021 Elsevier Ltd. All rights reserved.
查看更多>>摘要：In this paper, we have implemented a high-frequency quantitative system that can obtain stable returns for the Chinese A-share market, which has been running for more than 3 months (from March 27, 2020 to June 30, 2020) with the expected results. A number of rules and barriers exist in the Chinese A-share market such as trading restrictions and high fees, as well as scarce and expensive hedging tools. It is difficult to achieve stable absolute returns in such a market. Stock correlation analysis and price prediction play an important role to achieve any profitable trading. The portfolio management and subsequent trading decisions highly depend on the results of stock correlation analysis and price prediction. However, it is nontrivial to analyze and predict any stocks, being time-varying and affected by unlimited factors in a given market. Traditional methods only take some certain factors into consideration but ignore others that may be changed dynamically. In this paper, we propose a novel machine learning model named Graph Attention Long Short-Term Memory (GALSTM) to learn the correlations between stocks and predict their future prices automatically. First, a multi-Hawkes Process is used to initial a correlation graph between stocks. This procedure provides a good training start as the multi-Hawkes Processes will be studied on the most saint feature fluctuations with any correlations being statistically significant. Then an attention based LSTM is built to learn the weighting matrix underlying the dynamic graph. In addition, we also build matching data process plus portfolio management modules to form a complete system. The proposed GALSTM enables us to expand the scope of stock selection under the premise of controlling risks with limited hedging tools in the A-share market, thereby effectively increasing high-frequency excess returns. We then construct a long and short positions combination, select long positions in the A shares of the entire market, and use stock index futures to short. With GALSTM model, the products managed by our fully automatic quantitative trading system achieved an absolute annual return rate of 44.71% and the standard deviation of daily returns is only 0.42% in three months of operation. Only 1 week loss in 13 weeks of running time. (c) 2021 Published by Elsevier Ltd.
查看更多>>摘要：Face alignment in-the-wild still faces great challenges due to that i) partial occlusion blurs the inter features spatial relations of faces and ii) traditional CNN makes the network more difficult to capture the spatial positional relations between landmarks. To address the issues above, we propose a face alignment algorithm named Dual-attentional Spatial-aware Capsule Network (DSCN). Firstly, the spatial-aware module builds a more accurate inter-features spatial constrained model with the hourglass capsule network (HGCaps) as the backbone, which can effectively enhance its robustness against occlusions. Then, two sorts of attention mechanisms, namely capsule attention and spatial attention, are added to the attention-guided module to make the network focus more on the advantageous features and suppress other unrelated ones for more effective f eature recalibration. Our method achieves 1.08% failure rate on the COFW dataset, which is much lower than the current state-of-the-art algorithms. The mean error under 300W dataset and WFLW dataset are respectively 3.91% and 5.66%, which shows that DSCN is more robust to occlusion and outperforms state-of-the-art methods in the literature. (c) 2021 Elsevier Ltd. All rights reserved.
查看更多>>摘要：Generative Adversarial Networks (GANs) have become the most used networks towards solving the problem of image generation. Self-supervised GANs are later proposed to avoid the catastrophic forgetting of the discriminator and to improve the image generation quality without needing the class labels. However, the generalizability of the self-supervision tasks on different GAN architectures is not studied before. To that end, we extensively analyze the contribution of a previously proposed self-supervision task, deshuffling of the DeshuffleGANs in the generalizability context. We assign the deshuffling task to two different GAN discriminators and study the effects of the task on both architectures. We extend the evaluations compared to the previously proposed DeshuffleGANs on various datasets. We show that the DeshuffleGAN obtains the best FID results for several datasets compared to the other self-supervised GANs. Furthermore, we compare the deshuffling with the rotation prediction that is firstly deployed to the GAN training and demonstrate that its contribution exceeds the rotation prediction. We design the conditional DeshuffleGAN called cDeshuffleGAN to evaluate the quality of the learnt representations. Lastly, we show the contribution of the self-supervision tasks to the GAN training on the loss landscape and present that the effects of these tasks may not be cooperative to the adversarial training in some settings. Our code can be found at https://github.com/gulcinbaykal/DeshuffleGAN . (c) 2021 Elsevier Ltd. All rights reserved.
查看更多>>摘要：Point correspondence is a fundamental problem in pattern recognition and computer vision, which can be tackled by graph matching. Since graph matching is basically an NP-complete problem, some approximate methods are proposed to solve it. Continuous relaxation offers an effective approximate method for graph matching problem. However, the discrete constraint is not taken into consideration in the optimization step. In this paper, a fast normalized cut based graph matching method is proposed, where the discrete constraint is introduced into the optimization step. Specifically, first a semidefinite positive affinity matrix based form objective function is constructed by introducing a regularization term which is related to the discrete constraint. Then the fast normalized cut algorithm is utilized to find the continuous solution. Last, the discrete solution of graph matching is obtained by a multiplicative update algorithm. Experiments on both synthetic points and real-world images validate the effectiveness of the proposed method by comparing it with the state-of-the-art methods. 0 2021 Elsevier Ltd. All rights reserved.
查看更多>>摘要：In this paper, we propose a multiscale hierarchical attention approach for supervised video summarization. Different from most existing supervised methods which employ bidirectional long short-term memory networks, our method exploits the underlying hierarchical structure of video sequences and learns both the short-range and long-range temporal representations via a intra-block and a inter-block attention. Specifically, we first separate each video sequence into blocks of equal length and employ the intrablock and inter-block attention to learn local and global information, respectively. Then, we integrate the frame-level, block-level, and video-level representations for the frame-level importance score prediction. Next, we conduct shot segmentation and compute shot-level importance scores. Finally, we perform key shot selection to produce video summaries. Moreover, we extend our method into a two-stream framework, where appearance and motion information is leveraged. Experimental results on the SumMe and TVSum datasets validate the effectiveness of our method against state-of-the-art methods. (c) 2021 Elsevier Ltd. All rights reserved.
查看更多>>摘要：Embedding learning (EL) and feature synthesizing (FS) are two of the popular categories of fine-grained GZSL methods. EL or FS using global features cannot discriminate fine details in the absence of local features. On the other hand, EL or FS methods exploiting local features either neglect direct attribute guidance or global information. Consequently, neither method performs well. In this paper, we propose to explore global and direct attribute-supervised local visual features for both EL and FS categories in an integrated manner for fine-grained GZSL. The proposed integrated network has an EL sub-network and a FS sub-network. Consequently, the proposed integrated network can be tested in two ways. We propose a novel two-step dense attention mechanism to discover attribute-guided local visual features. We introduce new mutual learning between the sub-networks to exploit mutually beneficial information for optimization. Moreover, we propose to compute source-target class similarity based on mutual information and transfer-learn the target classes to reduce bias towards the source domain during testing. We demonstrate that our proposed method outperforms contemporary methods on benchmark datasets. (c) 2021 Elsevier Ltd. All rights reserved.
查看更多>>摘要：In this paper, we investigate the task of Fine-grained Sketch-based Image Retrieval (FG-SBIR), which uses hand-drawn sketches as input queries to retrieve the relevant images at the fine-grained instance level. The sketches and images come from different modalities, thus the similarity computation needs to consider both fine-grained and cross-modal characteristics. Existing solutions only focus on fine-grained details or spatial contexts, while ignoring the channel context and spatial sequence information. To mitigate such challenging problems, we propose a novel deep FG-SBIR model, which aims at inferring attention maps along channel dimension and spatial dimension, improving modules of channel attention and spatial attention, and exploring Transformer to enhance the model's ability for constructing and understanding spatial sequence information. We focus not only on the correlation information between two modalities of sketch and image, but also on the discrimination information inside the single modality. Mutual Loss is especially proposed to enhance the traditional triplet loss, and promote the internal discrimination ability of the model on a single modality. Extensive experiments show that our AE-Net obtains promising results on Sketchy , which is the largest public dataset available for FG-SBIR at present. (c) 2021 Elsevier Ltd. All rights reserved.
查看更多>>摘要：In this study, we present an incremental machine learning framework called Adaptive Decision Forest (ADF), which produces a decision forest to classify new records. Based on our two novel theorems, we introduce a new splitting strategy called iSAT, which allows ADF to classify new records even if they are associated with previously unseen classes. ADF is capable of identifying and handling concept drift; it, however, does not forget previously gained knowledge. Moreover, ADF is capable of handling big data if the data can be divided into batches. We evaluate ADF on nine publicly available natural datasets and one synthetic dataset, and compare the performance of ADF against the performance of eight state-of-the-art techniques. We also examine the effectiveness of ADF in some challenging situations. Our experimen-tal results, including statistical sign test and Nemenyi test analyses, indicate a clear superiority of the proposed framework over the state-of-the-art techniques. (c) 2021 Elsevier Ltd. All rights reserved.
NSTL
Elsevier