Haoran Xie

Bio: Haoran Xie is an academic researcher from Lingnan University. The author has contributed to research in topics: Computer science & Topic model. The author has an hindex of 29, co-authored 246 publications receiving 7117 citations. Previous affiliations of Haoran Xie include Hong Kong Institute of Education & Caritas Institute of Higher Education.

Least Squares Generative Adversarial Networks

Abstract: Unsupervised learning with generative adversarial networks (GANs) has proven hugely successful. Regular GANs hypothesize the discriminator as a classifier with the sigmoid cross entropy loss function. However, we found that this loss function may lead to the vanishing gradients problem during the learning process. To overcome such a problem, we propose in this paper the Least Squares Generative Adversarial Networks (LSGANs) which adopt the least squares loss function for the discriminator. We show that minimizing the objective function of LSGAN yields minimizing the Pearson X2 divergence. There are two benefits of LSGANs over regular GANs. First, LSGANs are able to generate higher quality images than regular GANs. Second, LSGANs perform more stable during the learning process. We evaluate LSGANs on LSUN and CIFAR-10 datasets and the experimental results show that the images generated by LSGANs are of better quality than the ones generated by regular GANs. We also conduct two comparison experiments between LSGANs and regular GANs to illustrate the stability of LSGANs.

Least Squares Generative Adversarial Networks

Abstract: Unsupervised learning with generative adversarial networks (GANs) has proven hugely successful. Regular GANs hypothesize the discriminator as a classifier with the sigmoid cross entropy loss function. However, we found that this loss function may lead to the vanishing gradients problem during the learning process. To overcome such a problem, we propose in this paper the Least Squares Generative Adversarial Networks (LSGANs) which adopt the least squares loss function for the discriminator. We show that minimizing the objective function of LSGAN yields minimizing the Pearson $\chi^2$ divergence. There are two benefits of LSGANs over regular GANs. First, LSGANs are able to generate higher quality images than regular GANs. Second, LSGANs perform more stable during the learning process. We evaluate LSGANs on five scene datasets and the experimental results show that the images generated by LSGANs are of better quality than the ones generated by regular GANs. We also conduct two comparison experiments between LSGANs and regular GANs to illustrate the stability of LSGANs.

News impact on stock price return via sentiment analysis

Abstract: Financial news articles are believed to have impacts on stock price return. Previous works model news pieces in bag-of-words space, which analyzes the latent relationship between word statistical patterns and stock price movements. However, news sentiment, which is an important ring on the chain of mapping from the word patterns to the price movements, is rarely touched. In this paper, we first implement a generic stock price prediction framework, and plug in six different models with different analyzing approaches. To take one step further, we use Harvard psychological dictionary and Loughran–McDonald financial sentiment dictionary to construct a sentiment space. Textual news articles are then quantitatively measured and projected onto the sentiment space. Instance labeling method is rigorously discussed and tested. We evaluate the models' prediction accuracy and empirically compare their performance at different market classification levels. Experiments are conducted on five years historical Hong Kong Stock Exchange prices and news articles. Results show that (1) at individual stock, sector and index levels, the models with sentiment analysis outperform the bag-of-words model in both validation set and independent testing set; (2) the models which use sentiment polarity cannot provide useful predictions; (3) there is a minor difference between the models using two different sentiment dictionaries.

Trends and development in technology-enhanced adaptive/personalized learning: A systematic review of journal publications from 2007 to 2017

Abstract: In this study, the trends and developments of technology-enhanced adaptive/personalized learning have been studied by reviewing the related journal articles in the recent decade (i.e., from 2007 to 2017). To be specific, we investigated many research issues such as the parameters of adaptive/personalized learning, learning supports, learning outcomes, subjects, participants, hardware, and so on. Furthermore, this study reveals that personalized/adaptive learning has always been an attractive topic in this field, and personalized data sources, for example, students’ preferences, learning achievements, profiles, and learning logs have become the main parameters for supporting personalized/adaptive learning. In addition, we found that the majority of the studies on personalized/adaptive learning still only supported traditional computers or devices, while only a few studies have been conducted on wearable devices, smartphones and tablet computers. In other words, personalized/adaptive learning has a significant number of potential applications on the above smart devices with the rapid development of artificial intelligence, virtual reality, cloud computing and wearable computing. Through the in-depth analysis of the trends and developments in the various dimensions of personalized/adaptive learning, the future research directions, issues and challenges are discussed in our paper.

Detecting latent topics and trends in educational technologies over four decades using structural topic modeling: A retrospective of all volumes of Computers & Education

Abstract: Computers & Education has been leading the field of computers in education for over 40 years, during which time it has developed into a well-known journal with significant influences on the educational technology research community. Questions such as “in what research topics were the academic community of Computers & Education interested?” “how did such research topics evolve over time?” and “what were the main research concerns of its major contributors?” are important to both the editorial board and readership of Computers & Education. To address these issues, this paper conducted a structural topic modeling analysis of 3963 articles published in Computers & Education between 1976 and 2018 bibliometrically. A structural topic model was used to profile the research hotspots. By further exploring annual topic proportion trends and topic correlations, potential future research directions and inter-topic research areas were identified. The major research concerns of the publications in Computers & Education by prolific countries/regions were shown and compared. Thus, this work provided useful insights and implications, and it could be used as a guide for contributors to Computers & Education.

A survey on Image Data Augmentation for Deep Learning

Abstract: Deep convolutional neural networks have performed remarkably well on many Computer Vision tasks. However, these networks are heavily reliant on big data to avoid overfitting. Overfitting refers to the phenomenon when a network learns a function with very high variance such as to perfectly model the training data. Unfortunately, many application domains do not have access to big data, such as medical image analysis. This survey focuses on Data Augmentation, a data-space solution to the problem of limited data. Data Augmentation encompasses a suite of techniques that enhance the size and quality of training datasets such that better Deep Learning models can be built using them. The image augmentation algorithms discussed in this survey include geometric transformations, color space augmentations, kernel filters, mixing images, random erasing, feature space augmentation, adversarial training, generative adversarial networks, neural style transfer, and meta-learning. The application of augmentation methods based on GANs are heavily covered in this survey. In addition to augmentation techniques, this paper will briefly discuss other characteristics of Data Augmentation such as test-time augmentation, resolution impact, final dataset size, and curriculum learning. This survey will present existing methods for Data Augmentation, promising developments, and meta-level decisions for implementing Data Augmentation. Readers will understand how Data Augmentation can improve the performance of their models and expand limited datasets to take advantage of the capabilities of big data.

Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks

Abstract: Image-to-image translation is a class of vision and graphics problems where the goal is to learn the mapping between an input image and an output image using a training set of aligned image pairs. However, for many tasks, paired training data will not be available. We present an approach for learning to translate an image from a source domain $X$ to a target domain $Y$ in the absence of paired examples. Our goal is to learn a mapping $G: X \rightarrow Y$ such that the distribution of images from $G(X)$ is indistinguishable from the distribution $Y$ using an adversarial loss. Because this mapping is highly under-constrained, we couple it with an inverse mapping $F: Y \rightarrow X$ and introduce a cycle consistency loss to push $F(G(X)) \approx X$ (and vice versa). Qualitative results are presented on several tasks where paired training data does not exist, including collection style transfer, object transfiguration, season transfer, photo enhancement, etc. Quantitative comparisons against several prior methods demonstrate the superiority of our approach.

Improved Training of Wasserstein GANs

Abstract: Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only low-quality samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to undesired behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.