arXiv subjects

Zhi Chen

Publications and source records attributed to Zhi Chen.

At least 253 records · Page 14Linked to original sources

arXiv2021

Mitigating Generation Shifts for Generalized Zero-Shot Learning

Generalized Zero-Shot Learning (GZSL) is the task of leveraging semantic information (e.g., attributes) to recognize the seen and unseen samples, where unseen classes are not observable during training. It is natural to derive generative models and hallucinate training samples for unseen classes based on the knowledge learned from the seen samples. However, most of these models suffer from the `generation shifts', where the synthesized samples may drift from the real distribution of unseen data. In this paper, we conduct an in-depth analysis on this issue and propose a novel Generation Shifts Mitigating Flow (GSMFlow) framework, which is comprised of multiple conditional affine coupling layers for learning unseen data synthesis efficiently and effectively. In particular, we identify three potential problems that trigger the generation shifts, i.e., semantic inconsistency, variance decay, and structural permutation and address them respectively. First, to reinforce the correlations between the generated samples and the respective attributes, we explicitly embed the semantic information into the transformations in each of the coupling layers. Second, to recover the intrinsic variance of the synthesized unseen features, we introduce a visual perturbation strategy to diversify the intra-class variance of generated data and hereby help adjust the decision boundary of the classifier. Third, to avoid structural permutation in the semantic space, we propose a relative positioning strategy to manipulate the attribute embeddings, guiding which to fully preserve the inter-class geometric structure. Experimental results demonstrate that GSMFlow achieves state-of-the-art recognition performance in both conventional and generalized zero-shot settings. Our code is available at: https://github.com/uqzhichen/GSMFlow

cs.CV↗

arXiv2021

RobustECD: Enhancement of Network Structure for Robust Community Detection

Community detection, which focuses on clustering vertex interactions, plays a significant role in network analysis. However, it also faces numerous challenges like missing data and adversarial attack. How to further improve the performance and robustness of community detection for real-world networks has raised great concerns. In this paper, we explore robust community detection by enhancing network structure, with two generic algorithms presented: one is named robust community detection via genetic algorithm (RobustECD-GA), in which the modularity and the number of clusters are combined in a fitness function to find the optimal structure enhancement scheme; the other is called robust community detection via similarity ensemble (RobustECD-SE), integrating multiple information of community structures captured by various vertex similarities, which scales well on large-scale networks. Comprehensive experiments on real-world networks demonstrate, by comparing with two traditional enhancement strategies, that the new methods help six representative community detection algorithms achieve more significant performance improvement. Moreover, experiments on the corresponding adversarial networks indicate that the new methods could also optimize the network structure to a certain extent, achieving stronger robustness against adversarial attack. The source code of this paper is released on https://github.com/jjzhou012/RobustECD.

cs.SI↗

arXiv2021

LGESQL: Line Graph Enhanced Text-to-SQL Model with Mixed Local and Non-Local Relations

This work aims to tackle the challenging heterogeneous graph encoding problem in the text-to-SQL task. Previous methods are typically node-centric and merely utilize different weight matrices to parameterize edge types, which 1) ignore the rich semantics embedded in the topological structure of edges, and 2) fail to distinguish local and non-local relations for each node. To this end, we propose a Line Graph Enhanced Text-to-SQL (LGESQL) model to mine the underlying relational features without constructing meta-paths. By virtue of the line graph, messages propagate more efficiently through not only connections between nodes, but also the topology of directed edges. Furthermore, both local and non-local relations are integrated distinctively during the graph iteration. We also design an auxiliary task called graph pruning to improve the discriminative capability of the encoder. Our framework achieves state-of-the-art results (62.8% with Glove, 72.0% with Electra) on the cross-domain text-to-SQL benchmark Spider at the time of writing.

cs.CL↗

arXiv2021

Decoupled Dialogue Modeling and Semantic Parsing for Multi-Turn Text-to-SQL

Recently, Text-to-SQL for multi-turn dialogue has attracted great interest. Here, the user input of the current turn is parsed into the corresponding SQL query of the appropriate database, given all previous dialogue history. Current approaches mostly employ end-to-end models and consequently face two challenges. First, dialogue history modeling and Text-to-SQL parsing are implicitly combined, hence it is hard to carry out interpretable analysis and obtain targeted improvement. Second, SQL annotation of multi-turn dialogue is very expensive, leading to training data sparsity. In this paper, we propose a novel decoupled multi-turn Text-to-SQL framework, where an utterance rewrite model first explicitly solves completion of dialogue context, and then a single-turn Text-to-SQL parser follows. A dual learning approach is also proposed for the utterance rewrite model to address the data sparsity problem. Compared with end-to-end approaches, the proposed decoupled method can achieve excellent performance without any annotated in-domain data. With just a few annotated rewrite cases, the decoupled method outperforms the released state-of-the-art end-to-end models on both SParC and CoSQL datasets.

cs.CL↗

arXiv2021

Improving THz Coverage for 6G URLLC Services via Exploiting Mobile Computing

Terahertz (THz) communication (0.1-10 THz) is regarded as a promising technology, which provides rich available bandwidth and high data rate of terahertz bit per second (Tbps). However, THz signals suffer from high path loss, which profoundly decreases the transmission distance. To improve THz coverage, we consider the aid of mobile computing. Specifically, job offloading decision in mobile computing and frequency allocation in communication are co-designed to maximize distance and concurrently support ultra-reliable low-latency communications (URLLC) services for the sixth-generation (6G) mobile communication. Further, the above optimization problem is non-convex, then an effective and low-complexity method is proposed via exploiting the special structure of this problem. Finally, numerical results verify the effectiveness of our work.

cs.IT↗

arXiv2021

Bring Your Own Codegen to Deep Learning Compiler

Deep neural networks (DNNs) have been ubiquitously applied in many applications, and accelerators are emerged as an enabler to support the fast and efficient inference tasks of these applications. However, to achieve high model coverage with high performance, each accelerator vendor has to develop a full compiler stack to ingest, optimize, and execute the DNNs. This poses significant challenges in the development and maintenance of the software stack. In addition, the vendors have to contiguously update their hardware and/or software to cope with the rapid evolution of the DNN model architectures and operators. To address these issues, this paper proposes an open source framework that enables users to only concentrate on the development of their proprietary code generation tools by reusing as many as possible components in the existing deep learning compilers. Our framework provides users flexible and easy-to-use interfaces to partition their models into segments that can be executed on "the best" processors to take advantage of the powerful computation capability of accelerators. Our case study shows that our framework has been deployed in multiple commercial vendors' compiler stacks with only a few thousand lines of code.

cs.LG↗

arXiv2021

Supervised Anomaly Detection via Conditional Generative Adversarial Network and Ensemble Active Learning

Anomaly detection has wide applications in machine intelligence but is still a difficult unsolved problem. Major challenges include the rarity of labeled anomalies and it is a class highly imbalanced problem. Traditional unsupervised anomaly detectors are suboptimal while supervised models can easily make biased predictions towards normal data. In this paper, we present a new supervised anomaly detector through introducing the novel Ensemble Active Learning Generative Adversarial Network (EAL-GAN). EAL-GAN is a conditional GAN having a unique one generator vs. multiple discriminators architecture where anomaly detection is implemented by an auxiliary classifier of the discriminator. In addition to using the conditional GAN to generate class balanced supplementary training data, an innovative ensemble learning loss function ensuring each discriminator makes up for the deficiencies of the others is designed to overcome the class imbalanced problem, and an active learning algorithm is introduced to significantly reduce the cost of labeling real-world data. We present extensive experimental results to demonstrate that the new anomaly detector consistently outperforms a variety of SOTA methods by significant margins. The codes are available on Github.

cs.LG↗

arXiv2021

ShadowGNN: Graph Projection Neural Network for Text-to-SQL Parser

Given a database schema, Text-to-SQL aims to translate a natural language question into the corresponding SQL query. Under the setup of cross-domain, traditional semantic parsing models struggle to adapt to unseen database schemas. To improve the model generalization capability for rare and unseen schemas, we propose a new architecture, ShadowGNN, which processes schemas at abstract and semantic levels. By ignoring names of semantic items in databases, abstract schemas are exploited in a well-designed graph projection neural network to obtain delexicalized representation of question and schema. Based on the domain-independent representations, a relation-aware transformer is utilized to further extract logical linking between question and schema. Finally, a SQL decoder with context-free grammar is applied. On the challenging Text-to-SQL benchmark Spider, empirical results show that ShadowGNN outperforms state-of-the-art models. When the annotated data is extremely limited (only 10\% training set), ShadowGNN gets over absolute 5\% performance gain, which shows its powerful generalization ability. Our implementation will be open-sourced at \url{https://github.com/WowCZ/shadowgnn}.

cs.CL↗

arXiv2021

3D Beam Training in Terahertz Communication: A Quadruple-UPA Architecture

In this paper, we propose a quadruple-UPA architecture for realizing the beamforming with wide coverage in terahertz (THz) communication. Considering the severe path loss suffered in the THz wave propagation, the pilot signals of conventional channel estimation strategies may not be effectively detected. In sight of this, we propose a fast 3D beam training strategy to jointly realize the angle estimation and the beamforming. Specifically, we first develop a novel hierarchical codebook that predefines some codewords (including narrow beams and wide beams) stage by stage. The narrow beams are designed as the beamforming solution candidates, whereas the wide beams are designed for reducing the training complexity. Then, we propose a 3D beam training procedure to find the optimal beamforming solution, i.e., the optimal narrow-beam pair, in a fast manner. Numerical results are presented to validate the effectiveness of the proposed scheme.

cs.IT↗

arXiv2021

Nimble: Efficiently Compiling Dynamic Neural Networks for Model Inference

Modern deep neural networks increasingly make use of features such as dynamic control flow, data structures and dynamic tensor shapes. Existing deep learning systems focus on optimizing and executing static neural networks which assume a pre-determined model architecture and input data shapes--assumptions which are violated by dynamic neural networks. Therefore, executing dynamic models with deep learning systems is currently both inflexible and sub-optimal, if not impossible. Optimizing dynamic neural networks is more challenging than static neural networks; optimizations must consider all possible execution paths and tensor shapes. This paper proposes Nimble, a high-performance and flexible system to optimize, compile, and execute dynamic neural networks on multiple platforms. Nimble handles model dynamism by introducing a dynamic type system, a set of dynamism-oriented optimizations, and a light-weight virtual machine runtime. Our evaluation demonstrates that Nimble outperforms state-of-the-art deep learning frameworks and runtime systems for dynamic neural networks by up to 20x on hardware platforms including Intel CPUs, ARM CPUs, and Nvidia GPUs.

cs.PL↗

arXiv2021

Cascade Network with Guided Loss and Hybrid Attention for Finding Good Correspondences

Finding good correspondences is a critical prerequisite in many feature based tasks. Given a putative correspondence set of an image pair, we propose a neural network which finds correct correspondences by a binary-class classifier and estimates relative pose through classified correspondences. First, we analyze that due to the imbalance in the number of correct and wrong correspondences, the loss function has a great impact on the classification results. Thus, we propose a new Guided Loss that can directly use evaluation criterion (Fn-measure) as guidance to dynamically adjust the objective function during training. We theoretically prove that the perfect negative correlation between the Guided Loss and Fn-measure, so that the network is always trained towards the direction of increasing Fn-measure to maximize it. We then propose a hybrid attention block to extract feature, which integrates the Bayesian attentive context normalization (BACN) and channel-wise attention (CA). BACN can mine the prior information to better exploit global context and CA can capture complex channel context to enhance the channel awareness of the network. Finally, based on our Guided Loss and hybrid attention block, a cascade network is designed to gradually optimize the result for more superior performance. Experiments have shown that our network achieves the state-of-the-art performance on benchmark datasets. Our code will be available in https://github.com/wenbingtao/GLHA.

cs.CV↗

arXiv2021

Terahertz Multi-User Massive MIMO with Intelligent Reflecting Surface: Beam Training and Hybrid Beamforming

Terahertz (THz) communications open a new frontier for the wireless network thanks to their dramatically wider available bandwidth compared to the current micro-wave and forthcoming millimeter-wave communications. However, due to the short length of THz waves, they also suffer from severe path attenuation and poor diffraction. To compensate the THz-induced propagation loss, this paper proposes to combine two promising techniques, viz., massive multiple input multiple output (MIMO) and intelligent reflecting surface (IRS), in THz multi-user communications, considering their significant beamforming and aperture gains. Nonetheless, channel estimation and low-cost beamforming turn out to be two main obstacles to realizing this combination, due to the passivity of IRS for sending/receiving pilot signals and the large-scale use of expensive RF chains in massive MIMO. In view of these limitations, this paper first develops a cooperative beam training scheme to facilitate the channel estimation with IRS. In particular, we design two different hierarchical codebooks for the proposed training procedure, which are able to balance between the robustness against noise and searching complexity. Based on the training results, we further propose two cost-efficient hybrid beamforming (HB) designs for both single-user and multi-user scenarios, respectively. Simulation results demonstrate that the proposed joint beam training and HB scheme is able to achieve close performance to the optimal fully digital beamforming (FDB) which is implemented even under perfect channel state information (CSI).

cs.IT↗

arXiv2021

Entropy-Based Uncertainty Calibration for Generalized Zero-Shot Learning

Compared to conventional zero-shot learning (ZSL) where recognising unseen classes is the primary or only aim, the goal of generalized zero-shot learning (GZSL) is to recognise both seen and unseen classes. Most GZSL methods typically learn to synthesise visual representations from semantic information on the unseen classes. However, these types of models are prone to overfitting the seen classes, resulting in distribution overlap between the generated features of the seen and unseen classes. The overlapping region is filled with uncertainty as the model struggles to determine whether a test case from within the overlap is seen or unseen. Further, these generative methods suffer in scenarios with sparse training samples. The models struggle to learn the distribution of high dimensional visual features and, therefore, fail to capture the most discriminative inter-class features. To address these issues, in this paper, we propose a novel framework that leverages dual variational autoencoders with a triplet loss to learn discriminative latent features and applies the entropy-based calibration to minimize the uncertainty in the overlapped area between the seen and unseen classes. Specifically, the dual generative model with the triplet loss synthesises inter-class discriminative latent features that can be mapped from either visual or semantic space. To calibrate the uncertainty for seen classes, we calculate the entropy over the softmax probability distribution from a general classifier. With this approach, recognising the seen samples within the seen classes is relatively straightforward, and there is less risk that a seen sample will be misclassified into an unseen class in the overlapped region. Extensive experiments on six benchmark datasets demonstrate that the proposed method outperforms state-of-the-art approaches.

cs.CV↗

arXiv2020

Fully Automated 3D Segmentation of MR-Imaged Calf Muscle Compartments: Neighborhood Relationship Enhanced Fully Convolutional Network

Automated segmentation of individual calf muscle compartments from 3D magnetic resonance (MR) images is essential for developing quantitative biomarkers for muscular disease progression and its prediction. Achieving clinically acceptable results is a challenging task due to large variations in muscle shape and MR appearance. Although deep convolutional neural networks (DCNNs) achieved improved accuracy in various image segmentation tasks, certain problems such as utilizing long-range information and incorporating high-level constraints remain unsolved. We present a novel fully convolutional network (FCN), called FilterNet, that utilizes contextual information in a large neighborhood and embeds edge-aware constraints for individual calf muscle compartment segmentations. An encoder-decoder architecture with flexible backbone blocks is used to systematically enlarge convolution receptive field and preserve information at all resolutions. Edge positions derived from the FCN output muscle probability maps are explicitly regularized using kernel-based edge detection in an end-to-end optimization framework. Our FilterNet was evaluated on 40 T1-weighted MR images of 10 healthy and 30 diseased subjects by 4-fold cross-validation. Mean DICE coefficients of 88.00%--91.29% and mean absolute surface positioning errors of 1.04--1.66 mm were achieved for the five 3D muscle compartments.

eess.IV↗

arXiv2020

Frequency Separation based Adaptive Feedforward Control for Rejecting Wideband Vibration with Application to Hard Disk Drives

In this paper, a frequency separation based adaptive feedforward control algorithm is developed with the ability to identify the plant and do compensation region by region. In this algorithm, the accelerometer signal is filtered by a series of uniformly distributed bandpass filters to generate a bunch of subband signals which are mutually exclusive in spectrum. In each subband, the corresponding subband signal acts as the feedforward signal and only the frequency response of system in that region needs to be identified, thus a pretty low order model can be expected to have efficient compensation. Starting from the first region, the feedforward control parameters are learned simultaneously with the low order plant model in the same region and then moves to the next region until all the regions are performed.

eess.SY↗

arXiv2020

Concept Whitening for Interpretable Image Recognition

What does a neural network encode about a concept as we traverse through the layers? Interpretability in machine learning is undoubtedly important, but the calculations of neural networks are very challenging to understand. Attempts to see inside their hidden layers can either be misleading, unusable, or rely on the latent space to possess properties that it may not have. In this work, rather than attempting to analyze a neural network posthoc, we introduce a mechanism, called concept whitening (CW), to alter a given layer of the network to allow us to better understand the computation leading up to that layer. When a concept whitening module is added to a CNN, the axes of the latent space are aligned with known concepts of interest. By experiment, we show that CW can provide us a much clearer understanding for how the network gradually learns concepts over layers. CW is an alternative to a batch normalization layer in that it normalizes, and also decorrelates (whitens) the latent space. CW can be used in any layer of the network without hurting predictive performance.

cs.LG↗

arXiv2020

Joint Power Allocation and Passive Beamforming Design for IRS-Assisted Physical-Layer Service Integration

Intelligent reflecting surface (IRS) has emerged as an appealing solution to enhance the wireless communication performance by reconfiguring the wireless propagation environment. In this paper, we propose to apply IRS to the physical-layer service integration (PHY-SI) system, where a single-antenna access point (AP) integrates two sorts of service messages, i.e., multicast message and confidential message, via superposition coding to serve multiple single-antenna users. Our goal is to optimize the power allocation (for transmitting different messages) at the AP and the passive beamforming at the IRS to maximize the achievable secrecy rate region. To this end, we formulate this problem as a bi-objective optimization problem, which is shown equivalent to a secrecy rate maximization problem subject to the constraints on the quality of multicast service. Due to the non-convexity of this problem, we propose two customized algorithms to obtain its high-quality suboptimal solutions, thereby approximately characterizing the secrecy rate region. The resulting performance gap with the globally optimal solution is analyzed. Furthermore, we provide theoretical analysis to unveil the impact of IRS beamforming on the performance of PHY-SI. Numerical results demonstrate the advantages of leveraging IRS in improving the performance of PHY-SI and also validate our theoretical analysis.

cs.IT↗

arXiv2020

Intelligent Reflecting Surface-Assisted Millimeter Wave Communications: Joint Active and Passive Precoding Design

Millimeter wave (MmWave) communications is capable of supporting multi-gigabit wireless access thanks to its abundant spectrum resource. However, the severe path loss and high directivity make it vulnerable to blockage events, which can be frequent in indoor and dense urban environments. To address this issue, in this paper, we introduce intelligent reflecting surface (IRS) as a new technology to provide effective reflected paths to enhance coverage of mmWave signals. In this framework, we study joint active and passive precoding design for IRS-assisted mmWave systems, where multiple IRSs are deployed to assist the data transmission from a base station (BS) to a single-antenna receiver. Our objective is to maximize the received signal power by jointly optimizing the transmit precoding vector at the BS and the phase shift parameters used by IRSs for passive beamforming. Although such an optimization problem is generally non-convex, we show that, by exploiting some important characteristics of mmWave channels, an optimal closed-form solution can be derived for the single IRS case and a near-optimal analytical solution can be obtained for the multi-IRS case. Our analysis reveals that the received signal power increases quadratically with the number of reflecting elements for both the single IRS and multi-IRS cases. Simulation results are included to verify the optimality and near-optimality of our proposed solutions. Results also show that IRSs can help create effective virtual LOS paths and thus substantially improve robustness against blockages in mmWave communications.

eess.SP↗