Convolutional neural networks (CNNs) have achieved significant performance on various real-life tasks. However, the large number of parameters in convolutional layers requires huge storage and computation resources, making it challenging to deploy CNNs on memory-constrained embedded devices. In this article, we propose a novel compression method that generates the convolution filters in each layer using a set of learnable low-dimensional quantized filter bases. The proposed method reconstructs the convolution filters by stacking the linear combinations of these filter bases. By using quantized values in weights, the compact filters can be represented using fewer bits so that the network can be highly compressed. Furthermore, we explore the sparsity of coefficients through $L_{1}$ -ball projection when conducting linear combination to further reduce the storage consumption and prevent overfitting. We also provide a detailed analysis of the compression performance of the proposed method. Evaluations of image classification and object detection tasks using various network structures demonstrate that the proposed method achieves a higher compression ratio with comparable accuracy compared with the existing state-of-the-art filter decomposition and network quantization methods.

Recent analyses have disclosed that existing rumor detection techniques, despite playing a pivotal role in countering the dissemination of misinformation on social media, are vulnerable to both white-box and surrogate-based black-box adversarial attacks. However, such attacks depend heavily on unrealistic assumptions, e.g., modifiable user data and white-box access to the rumor detection models, or appropriate selections of surrogate models, which are impractical in the real world. Thus, existing analyses fail to uncover the robustness of rumor detectors in practice. In this work, we take a further step towards the investigation about the robustness of existing rumor detection solutions. Specifically, we focus on the state-of-the-art rumor detectors, which leverage graph neural network based models to predict whether a post is rumor based on the Message Propagation Tree (MPT), a conversation tree with the post as its root and the replies to the post as the descendants of the root. We propose a novel black-box attack method, HMIA-LLM, against these rumor detectors, which uses the Large Language Model to generate malicious messages and inject them into the targeted MPTs. Our extensive evaluation conducted across three rumor detection datasets, four target rumor detectors, and three baselines for comparison demonstrates the effectiveness of our proposed attack method in compromising the performance of the state-of-the-art rumor detectors.

Sentiment analysis is an important and challenging task in natural language processing. It has been studied for a few decades. Recently, Bidirectional Encoder Representations from Transformer (BERT) model has been introduced to tackle this task and gain very promising results. However, most existing studies on fine-tuning BERT models for sentiment analysis focus on high-resource language (e.g., En-glish or Mandarin). This paper studies the sentiment analysis of Cantonese political posts on Hong Kong local forums. We first collected and labeled posts related to Anti-Extradition Law Amendment Bill (Anti-ELAB) movement in Hong Kong discussion forums. We then examined the performance of dictionary-based sentiment analysis, traditional machine learning-based, fine-tuned BERT and fine-tuned multilingual BERT (mBERT) models. Our results show that fine-tuned mBERT model achieves the best performance on our collected and labeled Cantonese dataset.

Preventing the spread of fake news is one of the most challenging issues in the age of social media. Traditional manual fact-checking (i.e., expert-based and crowd-sourced fact-checking) is time-consuming and labor-extensive, which cannot scale up with the unprecedented amount of dis- and mis-information on social media. Automated fact-checking based on machine learning is a promising strategy to address the scalability issues. Nevertheless, an end-to-end full automated fact-checking system without human supervision is still impractical. A more realistic solution will be developing an Artificial Intelligence (AI)-based system to facilitate the human fact-checkers during the fact-checking process. Therefore, this paper proposes a novel annotation system to facilitate human fact-checkers. With our designed procedures and schema, our developed system can help to improve the efficiency and effectiveness of human fact-checkers by automatically identifying worth-to-check news. We conduct a real-case study to demonstrate that our system can effectively identify worth-to-check news and ease the annotation process with the help of several automatic detection functions.

Deep Convolutional Neural Networks (CNN) have been successfully applied to many real-life problems. However, the huge memory cost of deep CNN models poses a great challenge of deploying them on memory-constrained devices (e.g., mobile phones). One popular way to reduce the memory cost of deep CNN model is to train binary CNN where the weights in convolution filters are either 1 or -1 and therefore each weight can be efficiently stored using a single bit. However, the compression ratio of existing binary CNN models is upper bounded by ∼ 32. To address this limitation, we propose a novel method to compress deep CNN model by stacking low-dimensional binary convolution filters. Our proposed method approximates a standard convolution filter by selecting and stacking filters from a set of low-dimensional binary convolution filters. This set of low-dimensional binary convolution filters is shared across all filters for a given convolution layer. Therefore, our method will achieve much larger compression ratio than binary CNN models. In order to train our proposed model, we have theoretically shown that our proposed model is equivalent to select and stack intermediate feature maps generated by low-dimensional binary filters. Therefore, our proposed model can be efficiently trained using the split-transform-merge strategy. We also provide detailed analysis of the memory and computation cost of our model in model inference. We compared the proposed method with other five popular model compression techniques on two benchmark datasets. Our experimental results have demonstrated that our proposed method achieves much higher compression ratio than existing methods while maintains comparable accuracy.

Kernel approximation is widely used to scale up kernel SVM training and prediction. However, the memory and computation costs of kernel approximation models are still too large if we want to deploy them on memory-limited devices such as mobile phones, smart watches and IoT devices. To address this challenge, we propose a novel memory and computation-efficient kernel SVM model by using both binary embedding and binary model coefficients. First, we propose an efficient way to generate compact binary embedding of the data which can preserve the kernel similarity. Second, we propose a simple but effective algorithm to learn a linear classification model with binary coefficients which can support different types of loss function and regularizer. Our algorithm can achieve better generalization accuracy than existing works on learning binary coefficients since we allow coefficient to be -1, 0 or 1 during the training stage and coefficient 0 can be removed during model inference. Moreover, we provide detailed analysis on the convergence of our algorithm and the inference complexity of our model. The analysis shows that the convergence to a local optimum is guaranteed and the inference complexity of our model is much lower than other competing methods. Our experimental results on five large real-world datasets have demonstrated that our proposed method can build accurate nonlinear SVM model with memory cost less than 30KB.

With the prevalence of social media, fake news has become one of the greatest challenges in journalism, which has weakened public trust in news outlets and authorities. During the COVID-19 epidemic, the widely circulated pandemic-related fake news on social media misleads or threatens the public. Recent works have investigated fake news detection on social platforms in English and Mandarin, though Cantonese fake news has been understudied. To pave the way for Cantonese COVID-19 fake news detection, we first presented an annotated COVID-19 related Cantonese fake news dataset collected from a popular local discussion forum in Hong Kong. Then, we explored the dataset by applying topic modeling to identify the topics that contain the most significant amount of fake news. Moreover, we evaluated both traditional machine learning algorithms and deep learning algorithms for Cantonese fake news detection. Our empirical results show that deep learning based methods perform slightly better than traditional machine learning methods on TF-IDF features.

Single-cell RNA sequencing has enabled to capture the gene activities at single-cell resolution, thus allowing reconstruction of cell-type-specific gene regulatory networks (GRNs). The available algorithms for reconstructing GRNs are commonly designed for bulk RNA-seq data, and few of them are applicable to analyze scRNA-seq data by dealing with the dropout events and cellular heterogeneity. In this paper, we represent the joint gene expression distribution of a gene pair as an image and propose a novel supervised deep neural network called DeepDRIM which utilizes the image of the target TF-gene pair and the ones of the potential neighbors to reconstruct GRN from scRNA-seq data. Due to the consideration of TF-gene pair's neighborhood context, DeepDRIM can effectively eliminate the false positives caused by transitive gene-gene interactions. We compared DeepDRIM with nine GRN reconstruction algorithms designed for either bulk or single-cell RNA-seq data. It achieves evidently better performance for the scRNA-seq data collected from eight cell lines. The simulated data show that DeepDRIM is robust to the dropout rate, the cell number and the size of the training data. We further applied DeepDRIM to the scRNA-seq gene expression of B cells from the bronchoalveolar lavage fluid of the patients with mild and severe coronavirus disease 2019. We focused on the cell-type-specific GRN alteration and observed targets of TFs that were differentially expressed between the two statuses to be enriched in lysosome, apoptosis, response to decreased oxygen level and microtubule, which had been proved to be associated with coronavirus infection.

Subsampled Randomized Hadamard Transform (SRHT), a popular random projection method that can efficiently project a d-dimensional data into r-dimensional space (r ≪ d) in O(dlog(d)) time, has been widely used to address the challenge of high-dimensionality in machine learning. SRHT works by rotating the input data matrix X ∈ ℝn × d by Randomized Walsh-Hadamard Transform followed with a subsequent uniform column sampling on the rotated matrix. Despite the advantages of SRHT, one limitation of SRHT is that it generates the new low-dimensional embedding without considering any specific properties of a given dataset. Therefore, this data-independent random projection method may result in inferior and unstable performance when used for a particular machine learning task, e.g., classification. To overcome this limitation, we analyze the effect of using SRHT for random projection in the context of linear SVM classification. Based on our analysis, we propose importance sampling and deterministic top-r sampling to produce effective low-dimensional embedding instead of uniform sampling SRHT. In addition, we also proposed a new supervised non-uniform sampling method. Our experimental results have demonstrated that our proposed methods can achieve higher classification accuracies than SRHT and other random projection methods on six real-life datasets.

Factorization Machines (FMs), a general predictor that can efficiently model high-order feature interactions, have been widely used for regression, classification and ranking problems. However, despite many successful applications of FMs, there are two main limitations of FMs: (1) FMs consider feature interactions among input features by using only polynomial expansion which fail to capture complex nonlinear patterns in data. (2) Existing FMs do not provide interpretable prediction to users. In this paper, we present a novel method named Subspace Encoding Factorization Machines (SEFM) to overcome these two limitations by using non-parametric subspace feature mapping. Due to the high sparsity of new feature representation, our proposed method achieves the same time complexity as the standard FMs but can capture more complex nonlinear patterns. Moreover, since the prediction score of our proposed model for a sample is a sum of contribution scores of the bins and grid cells that this sample lies in low-dimensional subspaces, it works similar like a scoring system which only involves data binning and score addition. Therefore, our proposed method naturally provides interpretable prediction. Our experimental results demonstrate that our proposed method efficiently provides accurate and interpretable prediction.

Kernel support vector machines (SVMs) deliver state-of-the-art results in many real-world nonlinear classification problems, but the computational cost can be quite demanding in order to maintain a large number of support vectors. Linear SVM, on the other hand, is highly scalable to large data but only suited for linearly separable problems. In this paper, we propose a novel approach called low-rank linearized SVM to scale up kernel SVM on limited resources. Our approach transforms a nonlinear SVM to a linear one via an approximate empirical kernel map computed from efficient kernel low-rank decompositions. We theoretically analyze the gap between the solutions of the approximate and optimal rank- k kernel map, which in turn provides guidance on the sampling scheme of the Nyström approximation. Furthermore, we extend it to a semisupervised metric learning scenario in which partially labeled samples can be exploited to further improve the quality of the low-rank embedding. Our approach inherits rich representability of kernel SVM and high efficiency of linear SVM. Experimental results demonstrate that our approach is more robust and achieves a better tradeoff between model representability and scalability against state-of-the-art algorithms for large-scale SVMs.

When the amount of labeled data are limited, semisupervised learning can improve the learner's performance by also using the often easily available unlabeled data. In particular, a popular approach requires the learned function to be smooth on the underlying data manifold. By approximating this manifold as a weighted graph, such graph-based techniques can often achieve state-of-the-art performance. However, their high time and space complexities make them less attractive on large data sets. In this paper, we propose to scale up graph-based semisupervised learning using a set of sparse prototypes derived from the data. These prototypes serve as a small set of data representatives, which can be used to approximate the graph-based regularizer and to control model complexity. Consequently, both training and testing become much more efficient. Moreover, when the Gaussian kernel is used to define the graph affinity, a simple and principled method to select the prototypes can be obtained. Experiments on a number of real-world data sets demonstrate encouraging performance and scaling properties of the proposed approach. It also compares favorably with models learned via l1 -regularization at the same level of model sparsity. These results demonstrate the efficacy of the proposed approach in producing highly parsimonious and accurate models for semisupervised learning.

In disease mapping, the spatial scan statistic is used to detect spatial regions where population is exposed to a significantly higher disease risk than expected. In this important application, the current residence is typically used to define the location of individuals from the population. Considering the mobility of humans at various temporal and spatial scales, using only information about the current residence may be an insufficiently informative proxy because it ignores a multitude of exposures that may occur away from home, or which had occurred at previous residences. In this paper, we propose a spatial scan statistic that is appropriate for disease mapping on mobile populations. We formulate a computationally efficient algorithm that uses the proposed statistic to find significant high-risk regions from mobile population's disease status data. The algorithm is applicable on large populations and over dense spatial grids. The experimental results demonstrate that the proposed algorithm is computationally efficient and outperforms the traditional disease clustering approaches at discovering high-risk regions in mobile populations.

We demonstrate OceanRT, a novel cloud-based infrastructure that performs online analytics in real time, over large-scale temporal data such as call logs from a telecommunication company. Apart from proprietary systems for which few details have been revealed, most existing big-data analytics systems are built on top of an offline, MapReduce-style infrastructure, which inherently limits their efficiency. In contrast, OceanRT employs a novel computing architecture consisting of interconnected Access Query Engines (AQEs), as well as a new storage scheme that ensures data locality and fast access for temporal data. Our preliminary evaluation shows that OceanRT can be up to 10x faster than Impala [10], 12x faster than Shark [5], and 200x faster than Hive [13]. The demo will show how OceanRT manages a real call log dataset (around 5TB per day) from a large mobile network operator in China. Besides presenting the processing of a few preset queries, we also allow the audience to issue ad hoc HiveQL [13] queries, watch how OceanRT answers them, and compare the speed of OceanRT with its competitors.

BACKGROUND: Protein function determination is a key challenge in the post-genomic era. Experimental determination of protein functions is accurate, but time-consuming and resource-intensive. A cost-effective alternative is to use the known information about sequence, structure, and functional properties of genes and proteins to predict functions using statistical methods. In this paper, we describe the Multi-Source k-Nearest Neighbor (MS-kNN) algorithm for function prediction, which finds k-nearest neighbors of a query protein based on different types of similarity measures and predicts its function by weighted averaging of its neighbors' functions. Specifically, we used 3 data sources to calculate the similarity scores: sequence similarity, protein-protein interactions, and gene expressions. RESULTS: We report the results in the context of 2011 Critical Assessment of Function Annotation (CAFA). Prior to CAFA submission deadline, we evaluated our algorithm on 1,302 human test proteins that were represented in all 3 data sources. Using only the sequence similarity information, MS-kNN had term-based Area Under the Curve (AUC) accuracy of Gene Ontology (GO) molecular function predictions of 0.728 when 7,412 human training proteins were used, and 0.819 when 35,622 training proteins from multiple eukaryotic and prokaryotic organisms were used. By aggregating predictions from all three sources, the AUC was further improved to 0.848. Similar result was observed on prediction of GO biological processes. Testing on 595 proteins that were annotated after the CAFA submission deadline showed that overall MS-kNN accuracy was higher than that of baseline algorithms Gotcha and BLAST, which were based solely on sequence similarity information. Since only 10 of the 595 proteins were represented by all 3 data sources, and 66 by two data sources, the difference between 3-source and one-source MS-kNN was rather small. CONCLUSIONS: Based on our results, we have several useful insights: (1) the k-nearest neighbor algorithm is an efficient and effective model for protein function prediction; (2) it is beneficial to transfer functions across a wide range of organisms; (3) it is helpful to integrate multiple sources of protein information.

The multiple instance regression (MIR) problem arises when a data set is a collection of bags, where each bag contains multiple instances sharing the identical real-valued label. The goal is to train a regression model that can accurately predict label of an unlabeled bag. Many remote sensing applications can be studied within this setting. We propose a novel probabilistic framework for MIR that represents bag labels with a mixture model. It is based on an assumption that each bag contains a prime instance which is responsible for the bag label. An expectation-maximization algorithm is proposed to maximize the likelihood of the mixture model. The mixture model MIR framework is quite flexible, and several existing MIR algorithms can be described as its special cases. The proposed algorithms were evaluated on synthetic data and remote sensing data for aerosol retrieval and crop yield prediction. The results show that the proposed MIR algorithms achieve higher accuracy than the previous state of the art.

A major challenge in microarray classification and biomarker discovery is dealing with small-sample high-dimensional data where the number of genes used as features is typically orders of magnitude larger than the number of labeled microarrays. One way to address this challenge is by leveraging information from the publicly accessible repositories of microarray data. Following this idea, a multi-task feature selection filter is proposed that borrows strength from the auxiliary microarray classification data sets. The filter uses Kruskal-Wallis test on auxiliary data sets and ranks genes based on their aggregated p-values. Expressions of the top-ranked genes are used as features to build a classifier on the target data set. The proposed approach was evaluated on 9 microarray data sets related to 9 different types of cancers. Comparison of the classification accuracies reveals that the multi-task feature selection is superior to single-task feature selection. Furthermore, the results strongly suggest that multi-task algorithms could improve microarray classification by exploiting auxiliary data during feature selection and learning.