arXiv subjects

Lihui Chen

Publications and source records attributed to Lihui Chen.

32 records · Page 2Linked to original sources

arXiv2023

SimCSE++: Improving Contrastive Learning for Sentence Embeddings from Two Perspectives

This paper improves contrastive learning for sentence embeddings from two perspectives: handling dropout noise and addressing feature corruption. Specifically, for the first perspective, we identify that the dropout noise from negative pairs affects the model's performance. Therefore, we propose a simple yet effective method to deal with such type of noise. Secondly, we pinpoint the rank bottleneck of current solutions to feature corruption and propose a dimension-wise contrastive learning objective to address this issue. Both proposed methods are generic and can be applied to any contrastive learning based models for sentence embeddings. Experimental results on standard benchmarks demonstrate that combining both proposed methods leads to a gain of 1.8 points compared to the strong baseline SimCSE configured with BERT base. Furthermore, applying the proposed method to DiffCSE, another strong contrastive learning based baseline, results in a gain of 1.4 points.

cs.CL↗

arXiv2023

BiGSeT: Binary Mask-Guided Separation Training for DNN-based Hyperspectral Anomaly Detection

Hyperspectral anomaly detection (HAD) aims to recognize a minority of anomalies that are spectrally different from their surrounding background without prior knowledge. Deep neural networks (DNNs), including autoencoders (AEs), convolutional neural networks (CNNs) and vision transformers (ViTs), have shown remarkable performance in this field due to their powerful ability to model the complicated background. However, for reconstruction tasks, DNNs tend to incorporate both background and anomalies into the estimated background, which is referred to as the identical mapping problem (IMP) and leads to significantly decreased performance. To address this limitation, we propose a model-independent binary mask-guided separation training strategy for DNNs, named BiGSeT. Our method introduces a separation training loss based on a latent binary mask to separately constrain the background and anomalies in the estimated image. The background is preserved, while the potential anomalies are suppressed by using an efficient second-order Laplacian of Gaussian (LoG) operator, generating a pure background estimate. In order to maintain separability during training, we periodically update the mask using a robust proportion threshold estimated before the training. In our experiments, We adopt a vanilla AE as the network to validate our training strategy on several real-world datasets. Our results show superior performance compared to some state-of-the-art methods. Specifically, we achieved a 90.67% AUC score on the HyMap Cooke City dataset. Additionally, we applied our training strategy to other deep network structures, achieving improved detection performance compared to their original versions, demonstrating its effective transferability. The code of our method will be available at https://github.com/enter-i-username/BiGSeT.

eess.IV↗

arXiv2023

Resource Efficient Neural Networks Using Hessian Based Pruning

Neural network pruning is a practical way for reducing the size of trained models and the number of floating-point operations. One way of pruning is to use the relative Hessian trace to calculate sensitivity of each channel, as compared to the more common magnitude pruning approach. However, the stochastic approach used to estimate the Hessian trace needs to iterate over many times before it can converge. This can be time-consuming when used for larger models with many millions of parameters. To address this problem, we modify the existing approach by estimating the Hessian trace using FP16 precision instead of FP32. We test the modified approach (EHAP) on ResNet-32/ResNet-56/WideResNet-28-8 trained on CIFAR10/CIFAR100 image classification tasks and achieve faster computation of the Hessian trace. Specifically, our modified approach can achieve speed ups ranging from 17% to as much as 44% during our experiments on different combinations of model architectures and GPU devices. Our modified approach also takes up around 40% less GPU memory when pruning ResNet-32 and ResNet-56 models, which allows for a larger Hessian batch size to be used for estimating the Hessian trace. Meanwhile, we also present the results of pruning using both FP16 and FP32 Hessian trace calculation and show that there are no noticeable accuracy differences between the two. Overall, it is a simple and effective way to compute the relative Hessian trace faster without sacrificing on pruned model performance. We also present a full pipeline using EHAP and quantization aware training (QAT), using INT8 QAT to compress the network further after pruning. In particular, we use symmetric quantization for the weights and asymmetric quantization for the activations.

cs.CV↗

arXiv2021

LA-HCN: Label-based Attention for Hierarchical Multi-label TextClassification Neural Network

Hierarchical multi-label text classification (HMTC) has been gaining popularity in recent years thanks to its applicability to a plethora of real-world applications. The existing HMTC algorithms largely focus on the design of classifiers, such as the local, global, or a combination of them. However, very few studies have focused on hierarchical feature extraction and explore the association between the hierarchical labels and the text. In this paper, we propose a Label-based Attention for Hierarchical Mutlti-label Text Classification Neural Network (LA-HCN), where the novel label-based attention module is designed to hierarchically extract important information from the text based on the labels from different hierarchy levels. Besides, hierarchical information is shared across levels while preserving the hierarchical label-based information. Separate local and global document embeddings are obtained and used to facilitate the respective local and global classifications. In our experiments, LA-HCN outperforms other state-of-the-art neural network-based HMTC algorithms on four public HMTC datasets. The ablation study also demonstrates the effectiveness of the proposed label-based attention module as well as the novel local and global embeddings and classifications. By visualizing the learned attention (words), we find that LA-HCN is able to extract meaningful information corresponding to the different labels which provides explainability that may be helpful for the human analyst.

cs.CL↗

arXiv2021

mSHINE: A Multiple-meta-paths Simultaneous Learning Framework for Heterogeneous Information Network Embedding

Heterogeneous information networks(HINs) become popular in recent years for its strong capability of modelling objects with abundant information using explicit network structure. Network embedding has been proved as an effective method to convert information networks into lower-dimensional space, whereas the core information can be well preserved. However, traditional network embedding algorithms are sub-optimal in capturing rich while potentially incompatible semantics provided by HINs. To address this issue, a novel meta-path-based HIN representation learning framework named mSHINE is designed to simultaneously learn multiple node representations for different meta-paths. More specifically, one representation learning module inspired by the RNN structure is developed and multiple node representations can be learned simultaneously, where each representation is associated with one respective meta-path. By measuring the relevance between nodes with the designed objective function, the learned module can be applied in downstream link prediction tasks. A set of criteria for selecting initial meta-paths is proposed as the other module in mSHINE which is important to reduce the optimal meta-path selection cost when no prior knowledge of suitable meta-paths is available. To corroborate the effectiveness of mSHINE, extensive experimental studies including node classification and link prediction are conducted on five real-world datasets. The results demonstrate that mSHINE outperforms other state-of-the-art HIN embedding methods.

cs.LG↗

arXiv2018

apk2vec: Semi-supervised multi-view representation learning for profiling Android applications

Building behavior profiles of Android applications (apps) with holistic, rich and multi-view information (e.g., incorporating several semantic views of an app such as API sequences, system calls, etc.) would help catering downstream analytics tasks such as app categorization, recommendation and malware analysis significantly better. Towards this goal, we design a semi-supervised Representation Learning (RL) framework named apk2vec to automatically generate a compact representation (aka profile/embedding) for a given app. More specifically, apk2vec has the three following unique characteristics which make it an excellent choice for largescale app profiling: (1) it encompasses information from multiple semantic views such as API sequences, permissions, etc., (2) being a semi-supervised embedding technique, it can make use of labels associated with apps (e.g., malware family or app category labels) to build high quality app profiles, and (3) it combines RL and feature hashing which allows it to efficiently build profiles of apps that stream over time (i.e., online learning). The resulting semi-supervised multi-view hash embeddings of apps could then be used for a wide variety of downstream tasks such as the ones mentioned above. Our extensive evaluations with more than 42,000 apps demonstrate that apk2vec's app profiles could significantly outperform state-of-the-art techniques in four app analytics tasks namely, malware detection, familial clustering, app clone detection and app recommendation.

cs.LG↗

arXiv2017

graph2vec: Learning Distributed Representations of Graphs

Recent works on representation learning for graph structured data predominantly focus on learning distributed representations of graph substructures such as nodes and subgraphs. However, many graph analytics tasks such as graph classification and clustering require representing entire graphs as fixed length feature vectors. While the aforementioned approaches are naturally unequipped to learn such representations, graph kernels remain as the most effective way of obtaining them. However, these graph kernels use handcrafted features (e.g., shortest paths, graphlets, etc.) and hence are hampered by problems such as poor generalization. To address this limitation, in this work, we propose a neural embedding framework named graph2vec to learn data-driven distributed representations of arbitrary sized graphs. graph2vec's embeddings are learnt in an unsupervised manner and are task agnostic. Hence, they could be used for any downstream task such as graph classification, clustering and even seeding supervised representation learning approaches. Our experiments on several benchmark and large real-world datasets show that graph2vec achieves significant improvements in classification and clustering accuracies over substructure representation learning approaches and are competitive with state-of-the-art graph kernels.

cs.AI↗

arXiv2017

Context-aware, Adaptive and Scalable Android Malware Detection through Online Learning (extended version)

It is well-known that Android malware constantly evolves so as to evade detection. This causes the entire malware population to be non-stationary. Contrary to this fact, most of the prior works on Machine Learning based Android malware detection have assumed that the distribution of the observed malware characteristics (i.e., features) does not change over time. In this work, we address the problem of malware population drift and propose a novel online learning based framework to detect malware, named CASANDRA (Contextaware, Adaptive and Scalable ANDRoid mAlware detector). In order to perform accurate detection, a novel graph kernel that facilitates capturing apps' security-sensitive behaviors along with their context information from dependency graphs is proposed. Besides being accurate and scalable, CASANDRA has specific advantages: i) being adaptive to the evolution in malware features over time ii) explaining the significant features that led to an app's classification as being malicious or benign. In a large-scale comparative analysis, CASANDRA outperforms two state-of-the-art techniques on a benchmark dataset achieving 99.23% F-measure. When evaluated with more than 87,000 apps collected in-the-wild, CASANDRA achieves 89.92% accuracy, outperforming existing techniques by more than 25% in their typical batch learning setting and more than 7% when they are continuously retained, while maintaining comparable efficiency.

cs.CR↗

arXiv2017

A Multi-view Context-aware Approach to Android Malware Detection and Malicious Code Localization

Existing Android malware detection approaches use a variety of features such as security sensitive APIs, system calls, control-flow structures and information flows in conjunction with Machine Learning classifiers to achieve accurate detection. Each of these feature sets provides a unique semantic perspective (or view) of apps' behaviours with inherent strengths and limitations. Meaning, some views are more amenable to detect certain attacks but may not be suitable to characterise several other attacks. Most of the existing malware detection approaches use only one (or a selected few) of the aforementioned feature sets which prevent them from detecting a vast majority of attacks. Addressing this limitation, we propose MKLDroid, a unified framework that systematically integrates multiple views of apps for performing comprehensive malware detection and malicious code localisation. The rationale is that, while a malware app can disguise itself in some views, disguising in every view while maintaining malicious intent will be much harder. MKLDroid uses a graph kernel to capture structural and contextual information from apps' dependency graphs and identify malice code patterns in each view. Subsequently, it employs Multiple Kernel Learning (MKL) to find a weighted combination of the views which yields the best detection accuracy. Besides multi-view learning, MKLDroid's unique and salient trait is its ability to locate fine-grained malice code portions in dependency graphs (e.g., methods/classes). Through our large-scale experiments on several datasets (incl. wild apps), we demonstrate that MKLDroid outperforms three state-of-the-art techniques consistently, in terms of accuracy while maintaining comparable efficiency. In our malicious code localisation experiments on a dataset of repackaged malware, MKLDroid was able to identify all the malice classes with 94% average recall.

cs.CR↗

arXiv2016

Adaptive and Scalable Android Malware Detection through Online Learning

It is well-known that malware constantly evolves so as to evade detection and this causes the entire malware population to be non-stationary. Contrary to this fact, prior works on machine learning based Android malware detection have assumed that the distribution of the observed malware characteristics (i.e., features) do not change over time. In this work, we address the problem of malware population drift and propose a novel online machine learning based framework, named DroidOL to handle it and effectively detect malware. In order to perform accurate detection, security-sensitive behaviors are captured from apps in the form of inter-procedural control-flow sub-graph features using a state-of-the-art graph kernel. In order to perform scalable detection and to adapt to the drift and evolution in malware population, an online passive-aggressive classifier is used. In a large-scale comparative analysis with more than 87,000 apps, DroidOL achieves 84.29% accuracy outperforming two state-of-the-art malware techniques by more than 20% in their typical batch learning setting and more than 3% when they are continuously re-trained. Our experimental findings strongly indicate that online learning based approaches are highly suitable for real-world malware detection.

cs.CR↗

arXiv2016

Incremental Minimax Optimization based Fuzzy Clustering for Large Multi-view Data

Incremental clustering approaches have been proposed for handling large data when given data set is too large to be stored. The key idea of these approaches is to find representatives to represent each cluster in each data chunk and final data analysis is carried out based on those identified representatives from all the chunks. However, most of the incremental approaches are used for single view data. As large multi-view data generated from multiple sources becomes prevalent nowadays, there is a need for incremental clustering approaches to handle both large and multi-view data. In this paper we propose a new incremental clustering approach called incremental minimax optimization based fuzzy clustering (IminimaxFCM) to handle large multi-view data. In IminimaxFCM, representatives with multiple views are identified to represent each cluster by integrating multiple complementary views using minimax optimization. The detailed problem formulation, updating rules derivation, and the in-depth analysis of the proposed IminimaxFCM are provided. Experimental studies on several real world multi-view data sets have been conducted. We observed that IminimaxFCM outperforms related incremental fuzzy clustering in terms of clustering accuracy, demonstrating the great potential of IminimaxFCM for large multi-view data analysis.

cs.AI↗

arXiv2016

Multi-View Fuzzy Clustering with Minimax Optimization for Effective Clustering of Data from Multiple Sources

Multi-view data clustering refers to categorizing a data set by making good use of related information from multiple representations of the data. It becomes important nowadays because more and more data can be collected in a variety of ways, in different settings and from different sources, so each data set can be represented by different sets of features to form different views of it. Many approaches have been proposed to improve clustering performance by exploring and integrating heterogeneous information underlying different views. In this paper, we propose a new multi-view fuzzy clustering approach called MinimaxFCM by using minimax optimization based on well-known Fuzzy c means. In MinimaxFCM the consensus clustering results are generated based on minimax optimization in which the maximum disagreements of different weighted views are minimized. Moreover, the weight of each view can be learned automatically in the clustering process. In addition, there is only one parameter to be set besides the fuzzifier. The detailed problem formulation, updating rules derivation, and the in-depth analysis of the proposed MinimaxFCM are provided here. Experimental studies on nine multi-view data sets including real world image and document data sets have been conducted. We observed that MinimaxFCM outperforms related multi-view clustering approaches in terms of clustering accuracy, demonstrating the great potential of MinimaxFCM for multi-view data analysis.

cs.AI↗

arXiv2016

subgraph2vec: Learning Distributed Representations of Rooted Sub-graphs from Large Graphs

In this paper, we present subgraph2vec, a novel approach for learning latent representations of rooted subgraphs from large graphs inspired by recent advancements in Deep Learning and Graph Kernels. These latent representations encode semantic substructure dependencies in a continuous vector space, which is easily exploited by statistical models for tasks such as graph classification, clustering, link prediction and community detection. subgraph2vec leverages on local information obtained from neighbourhoods of nodes to learn their latent representations in an unsupervised fashion. We demonstrate that subgraph vectors learnt by our approach could be used in conjunction with classifiers such as CNNs, SVMs and relational data clustering algorithms to achieve significantly superior accuracies. Also, we show that the subgraph vectors could be used for building a deep learning variant of Weisfeiler-Lehman graph kernel. Our experiments on several benchmark and large-scale real-world datasets reveal that subgraph2vec achieves significant improvements in accuracies over existing graph kernels on both supervised and unsupervised learning tasks. Specifically, on two realworld program analysis tasks, namely, code clone and malware detection, subgraph2vec outperforms state-of-the-art kernels by more than 17% and 4%, respectively.

cs.LG↗

arXiv2016

Contextual Weisfeiler-Lehman Graph Kernel For Malware Detection

In this paper, we propose a novel graph kernel specifically to address a challenging problem in the field of cyber-security, namely, malware detection. Previous research has revealed the following: (1) Graph representations of programs are ideally suited for malware detection as they are robust against several attacks, (2) Besides capturing topological neighbourhoods (i.e., structural information) from these graphs it is important to capture the context under which the neighbourhoods are reachable to accurately detect malicious neighbourhoods. We observe that state-of-the-art graph kernels, such as Weisfeiler-Lehman kernel (WLK) capture the structural information well but fail to capture contextual information. To address this, we develop the Contextual Weisfeiler-Lehman kernel (CWLK) which is capable of capturing both these types of information. We show that for the malware detection problem, CWLK is more expressive and hence more accurate than WLK while maintaining comparable efficiency. Through our large-scale experiments with more than 50,000 real-world Android apps, we demonstrate that CWLK outperforms two state-of-the-art graph kernels (including WLK) and three malware detection techniques by more than 5.27% and 4.87% F-measure, respectively, while maintaining high efficiency. This high accuracy and efficiency make CWLK suitable for large-scale real-world malware detection.

cs.CR↗