Li Xiao

Peng Wang, Xiao Li, Can Yaras et al.

Over the past decade, deep learning has proven to be a highly effective tool for learning meaningful features from raw data. However, it remains an open question how deep networks perform hierarchical feature learning across layers. In this work, we attempt to unveil this mystery by investigating the structures of intermediate features. Motivated by our empirical findings that linear layers mimic the roles of deep layers in nonlinear networks for feature learning, we explore how deep linear networks transform input data into output by investigating the output (i.e., features) of each layer after training in the context of multi-class classification problems. Toward this goal, we first define metrics to measure within-class compression and between-class discrimination of intermediate features, respectively. Through theoretical analysis of these two metrics, we show that the evolution of features follows a simple and quantitative pattern from shallow to deep layers when the input data is nearly orthogonal and the network weights are minimum-norm, balanced, and approximate low-rank: Each layer of the linear network progressively compresses within-class features at a geometric rate and discriminates between-class features at a linear rate with respect to the number of layers that data have passed through. To the best of our knowledge, this is the first quantitative characterization of feature evolution in hierarchical representations of deep linear networks. Empirically, our extensive experiments not only validate our theoretical results numerically but also reveal a similar pattern in deep nonlinear networks which aligns well with recent empirical studies. Moreover, we demonstrate the practical implications of our results in transfer learning. Our code is available at https://github.com/Heimine/PNC_DLN.

Zijie Geng, Xijun Li, Jie Wang et al.

In the past few years, there has been an explosive surge in the use of machine learning (ML) techniques to address combinatorial optimization (CO) problems, especially mixed-integer linear programs (MILPs). Despite the achievements, the limited availability of real-world instances often leads to sub-optimal decisions and biased solver assessments, which motivates a suite of synthetic MILP instance generation techniques. However, existing methods either rely heavily on expert-designed formulations or struggle to capture the rich features of real-world instances. To tackle this problem, we propose G2MILP, the first deep generative framework for MILP instances. Specifically, G2MILP represents MILP instances as bipartite graphs, and applies a masked variational autoencoder to iteratively corrupt and replace parts of the original graphs to generate new ones. The appealing feature of G2MILP is that it can learn to generate novel and realistic MILP instances without prior expert-designed formulations, while preserving the structures and computational hardness of real-world datasets, simultaneously. Thus the generated instances can facilitate downstream tasks for enhancing MILP solvers under limited data availability. We design a suite of benchmarks to evaluate the quality of the generated MILP instances. Experiments demonstrate that our method can produce instances that closely resemble real-world datasets in terms of both structures and computational hardness. The deliverables are released at https://miralab-ustc.github.io/L2O-G2MILP.

Hanqing Guo, Guangjing Wang, Yuanda Wang et al.

In this paper, we propose PhantomSound, a query-efficient black-box attack toward voice assistants. Existing black-box adversarial attacks on voice assistants either apply substitution models or leverage the intermediate model output to estimate the gradients for crafting adversarial audio samples. However, these attack approaches require a significant amount of queries with a lengthy training stage. PhantomSound leverages the decision-based attack to produce effective adversarial audios, and reduces the number of queries by optimizing the gradient estimation. In the experiments, we perform our attack against 4 different speech-to-text APIs under 3 real-world scenarios to demonstrate the real-time attack impact. The results show that PhantomSound is practical and robust in attacking 5 popular commercial voice controllable devices over the air, and is able to bypass 3 liveness detection mechanisms with >95% success rate. The benchmark result shows that PhantomSound can generate adversarial examples and launch the attack in a few minutes. We significantly enhance the query efficiency and reduce the cost of a successful untargeted and targeted adversarial attack by 93.1% and 65.5% compared with the state-of-the-art black-box attacks, using merely ~300 queries (~5 minutes) and ~1,500 queries (~25 minutes), respectively.

Ying Chen, Xiao Li, Hongbo Zhang et al.

Chlorophyll concentration can well reflect the nutritional status and algal blooms of water bodies, and is an important indicator for evaluating water quality. The prediction of chlorophyll concentration change trend is of great significance to environmental protection and aquaculture. However, there is a complex and indistinguishable nonlinear relationship between many factors affecting chlorophyll concentration. In order to effectively mine the nonlinear features contained in the data. This paper proposes a time-series decomposition adaptive graph-time convolutional network ( AGTCNSD ) prediction model. Firstly, the original sequence is decomposed into trend component and periodic component by moving average method. Secondly, based on the graph convolutional neural network, the water quality parameter data is modeled, and a parameter embedding matrix is defined. The idea of matrix decomposition is used to assign weight parameters to each node. The adaptive graph convolution learns the relationship between different water quality parameters, updates the state information of each parameter, and improves the learning ability of the update relationship between nodes. Finally, time dependence is captured by time convolution to achieve multi-step prediction of chlorophyll concentration. The validity of the model is verified by the water quality data of the coastal city Beihai. The results show that the prediction effect of this method is better than other methods. It can be used as a scientific resource for environmental management decision-making.

Yuzhi Wu, Li Xiao, Jun Liu et al.

The emerging 4D millimeter-wave radar, measuring the range, azimuth, elevation, and Doppler velocity of objects, is recognized for its cost-effectiveness and robustness in autonomous driving. Nevertheless, its point clouds exhibit significant sparsity and noise, restricting its standalone application in 3D object detection. Recent 4D radar-camera fusion methods have provided effective perception. Most existing approaches, however, adopt explicit Bird's-Eye-View fusion paradigms originally designed for LiDAR-camera fusion, neglecting radar's inherent drawbacks. Specifically, they overlook the sparse and incomplete geometry of radar point clouds and restrict fusion to coarse scene-level integration. To address these problems, we propose MLF-4DRCNet, a novel two-stage framework for 3D object detection via multi-level fusion of 4D radar and camera images. Our model incorporates the point-, scene-, and proposal-level multi-modal information, enabling comprehensive feature representation. It comprises three crucial components: the Enhanced Radar Point Encoder (ERPE) module, the Hierarchical Scene Fusion Pooling (HSFP) module, and the Proposal-Level Fusion Enhancement (PLFE) module. Operating at the point-level, ERPE densities radar point clouds with 2D image instances and encodes them into voxels via the proposed Triple-Attention Voxel Feature Encoder. HSFP dynamically integrates multi-scale voxel features with 2D image features using deformable attention to capture scene context and adopts pooling to the fused features. PLFE refines region proposals by fusing image features, and further integrates with the pooled features from HSFP. Experimental results on the View-of-Delft (VoD) and TJ4DRadSet datasets demonstrate that MLF-4DRCNet achieves the state-of-the-art performance. Notably, it attains performance comparable to LiDAR-based models on the VoD dataset.

Gang Qu, Li Xiao, Wenxing Hu et al.

Objective: Multi-modal functional magnetic resonance imaging (fMRI) can be used to make predictions about individual behavioral and cognitive traits based on brain connectivity networks. Methods: To take advantage of complementary information from multi-modal fMRI, we propose an interpretable multi-modal graph convolutional network (MGCN) model, incorporating the fMRI time series and the functional connectivity (FC) between each pair of brain regions. Specifically, our model learns a graph embedding from individual brain networks derived from multi-modal data. A manifold-based regularization term is then enforced to consider the relationships of subjects both within and between modalities. Furthermore, we propose the gradient-weighted regression activation mapping (Grad-RAM) and the edge mask learning to interpret the model, which is used to identify significant cognition-related biomarkers. Results: We validate our MGCN model on the Philadelphia Neurodevelopmental Cohort to predict individual wide range achievement test (WRAT) score. Our model obtains superior predictive performance over GCN with a single modality and other competing approaches. The identified biomarkers are cross-validated from different approaches. Conclusion and Significance: This paper develops a new interpretable graph deep learning framework for cognitive ability prediction, with the potential to overcome the limitations of several current data-fusion models. The results demonstrate the power of MGCN in analyzing multi-modal fMRI and discovering significant biomarkers for human brain studies.

Li Xiao, Xiang-Gen Xia, Yu-Ping Wang

The robust Chinese remainder theorem (CRT) has been recently proposed for robustly reconstructing a large nonnegative integer from erroneous remainders. It has found many applications in signal processing, including phase unwrapping and frequency estimation under sub-Nyquist sampling. Motivated by the applications in multidimensional (MD) signal processing, in this paper we propose the MD-CRT and robust MD-CRT for integer vectors. Specifically, by rephrasing the abstract CRT for rings in number-theoretic terms, we first derive the MD-CRT for integer vectors with respect to a general set of integer matrix moduli, which provides an algorithm to uniquely reconstruct an integer vector from its remainders, if it is in the fundamental parallelepiped of the lattice generated by a least common right multiple of all the moduli. For some special forms of moduli, we present explicit reconstruction formulae. Moreover, we derive the robust MD-CRT for integer vectors when the remaining integer matrices of all the moduli left divided by their greatest common left divisor (gcld) are pairwise commutative and coprime. Two different reconstruction algorithms are proposed, and accordingly, two different conditions on the remainder error bound for the reconstruction robustness are obtained, which are related to a quarter of the minimum distance of the lattice generated by the gcld of all the moduli or the Smith normal form of the gcld.

Peyman Hosseinzadeh Kassani, Li Xiao, Gemeng Zhang et al.

Human brain development is a complex and dynamic process that is affected by several factors such as genetics, sex hormones, and environmental changes. A number of recent studies on brain development have examined functional connectivity (FC) defined by the temporal correlation between time series of different brain regions. We propose to add the directional flow of information during brain maturation. To do so, we extract effective connectivity (EC) through Granger causality (GC) for two different groups of subjects, i.e., children and young adults. The motivation is that the inclusion of causal interaction may further discriminate brain connections between two age groups and help to discover new connections between brain regions. The contributions of this study are threefold. First, there has been a lack of attention to EC-based feature extraction in the context of brain development. To this end, we propose a new kernel-based GC (KGC) method to learn nonlinearity of complex brain network, where a reduced Sine hyperbolic polynomial (RSP) neural network was used as our proposed learner. Second, we used causality values as the weight for the directional connectivity between brain regions. Our findings indicated that the strength of connections was significantly higher in young adults relative to children. In addition, our new EC-based feature outperformed FC-based analysis from Philadelphia neurocohort (PNC) study with better discrimination of the different age groups. Moreover, the fusion of these two sets of features (FC + EC) improved brain age prediction accuracy by more than 4%, indicating that they should be used together for brain development studies.

Xiao Li, Yue Dong, Pieter Peers et al.

We present a new weakly supervised learning-based method for generating novel category-specific 3D shapes from unoccluded image collections. Our method is weakly supervised and only requires silhouette annotations from unoccluded, category-specific objects. Our method does not require access to the object's 3D shape, multiple observations per object from different views, intra-image pixel-correspondences, or any view annotations. Key to our method is a novel multi-projection generative adversarial network (MP-GAN) that trains a 3D shape generator to be consistent with multiple 2D projections of the 3D shapes, and without direct access to these 3D shapes. This is achieved through multiple discriminators that encode the distribution of 2D projections of the 3D shapes seen from a different views. Additionally, to determine the view information for each silhouette image, we also train a view prediction network on visualizations of 3D shapes synthesized by the generator. We iteratively alternate between training the generator and training the view prediction network. We validate our multi-projection GAN on both synthetic and real image datasets. Furthermore, we also show that multi-projection GANs can aid in learning other high-dimensional distributions from lower dimensional training datasets, such as material-class specific spatially varying reflectance properties from images.

Jia Xu, Zu-Zhen Huang, Zhi-Rui Wang et al.

In this paper, with respect to multichannel synthetic aperture radars (SAR), we first formulate the problems of Doppler ambiguities on the radial velocity (RV) estimation of a ground moving target in range-compressed domain, range-Doppler domain and image domain, respectively. It is revealed that in these problems, a cascaded time-space Doppler ambiguity (CTSDA) may encounter, i.e., time domain Doppler ambiguity (TDDA) in each channel arises first and then spatial domain Doppler ambiguity (SDDA) among multi-channels arises second. Accordingly, the multichannel SAR systems with different parameters are investigated in three different cases with diverse Doppler ambiguity properties, and a multi-frequency SAR is then proposed to obtain the RV estimation by solving the ambiguity problem based on Chinese remainder theorem (CRT). In the first two cases, the ambiguity problem can be solved by the existing closed-form robust CRT. In the third case, it is found that the problem is different from the conventional CRT problems and we call it a double remaindering problem in this paper. We then propose a sufficient condition under which the double remaindering problem, i.e., the CTSDA, can also be solved by the closed-form robust CRT. When the sufficient condition is not satisfied for a multi-channel SAR, a searching based method is proposed. Finally, some results of numerical experiments are provided to demonstrate the effectiveness of the proposed methods.

Li Xiao, Xiang-Gen Xia, Wenjie Wang

It is well-known that the traditional Chinese remainder theorem (CRT) is not robust in the sense that a small error in a remainder may cause a large error in the reconstruction solution. A robust CRT was recently proposed for a special case when the greatest common divisor (gcd) of all the moduli is more than 1 and the remaining integers factorized by the gcd of all the moduli are co-prime. In this special case, a closed-form reconstruction from erroneous remainders was proposed and a necessary and sufficient condition on the remainder errors was obtained. It basically says that the reconstruction error is upper bounded by the remainder error level $τ$ if $τ$ is smaller than a quarter of the gcd of all the moduli. In this paper, we consider the robust reconstruction problem for a general set of moduli. We first present a necessary and sufficient condition for the remainder errors for a robust reconstruction from erroneous remainders with a general set of muduli and also a corresponding robust reconstruction method. This can be thought of as a single stage robust CRT. We then propose a two-stage robust CRT by grouping the moduli into several groups as follows. First, the single stage robust CRT is applied to each group. Then, with these robust reconstructions from all the groups, the single stage robust CRT is applied again across the groups. This is then easily generalized to multi-stage robust CRT. Interestingly, with this two-stage robust CRT, the robust reconstruction holds even when the remainder error level $τ$ is above the quarter of the gcd of all the moduli. In this paper, we also propose an algorithm on how to group a set of moduli for a better reconstruction robustness of the two-stage robust CRT in some special cases.