Searcharxiv⌕ Search

arXiv subjects

Haifeng Li

Publications and source records attributed to Haifeng Li.

At least 127 records · Page 7Linked to original sources
arXiv2019

Understanding the Importance of Single Directions via Representative Substitution

Understanding the internal representations of deep neural networks (DNNs) is crucal to explain their behavior. The interpretation of individual units, which are neurons in MLPs or convolution kernels in convolutional networks, has been paid much attention given their fundamental role. However, recent research (Morcos et al. 2018) presented a counterintuitive phenomenon, which suggests that an individual unit with high class selectivity, called interpretable units, has poor contributions to generalization of DNNs. In this work, we provide a new perspective to understand this counterintuitive phenomenon, which makes sense when we introduce Representative Substitution (RS). Instead of individually selective units with classes, the RS refers to the independence of a unit's representations in the same layer without any annotation. Our experiments demonstrate that interpretable units have high RS which are not critical to network's generalization. The RS provides new insights into the interpretation of DNNs and suggests that we need to focus on the independence and relationship of the representations.

cs.LG↗
arXiv2018

T-GCN: A Temporal Graph ConvolutionalNetwork for Traffic Prediction

Accurate and real-time traffic forecasting plays an important role in the Intelligent Traffic System and is of great significance for urban traffic planning, traffic management, and traffic control. However, traffic forecasting has always been considered an open scientific issue, owing to the constraints of urban road network topological structure and the law of dynamic change with time, namely, spatial dependence and temporal dependence. To capture the spatial and temporal dependence simultaneously, we propose a novel neural network-based traffic forecasting method, the temporal graph convolutional network (T-GCN) model, which is in combination with the graph convolutional network (GCN) and gated recurrent unit (GRU). Specifically, the GCN is used to learn complex topological structures to capture spatial dependence and the gated recurrent unit is used to learn dynamic changes of traffic data to capture temporal dependence. Then, the T-GCN model is employed to traffic forecasting based on the urban road network. Experiments demonstrate that our T-GCN model can obtain the spatio-temporal correlation from traffic data and the predictions outperform state-of-art baselines on real-world traffic datasets. Our tensorflow implementation of the T-GCN is available at https://github.com/lehaifeng/T-GCN.

cs.LG↗
arXiv2018

Understanding the Importance of Single Directions via Representative Substitution

Understanding the internal representations of deep neural networks (DNNs) is crucal to explain their behavior. The interpretation of individual units, which are neurons in MLPs or convolution kernels in convolutional networks, has been paid much attention given their fundamental role. However, recent research (Morcos et al. 2018) presented a counterintuitive phenomenon, which suggests that an individual unit with high class selectivity, called interpretable units, has poor contributions to generalization of DNNs. In this work, we provide a new perspective to understand this counterintuitive phenomenon, which makes sense when we introduce Representative Substitution (RS). Instead of individually selective units with classes, the RS refers to the independence of a unit's representations in the same layer without any annotation. Our experiments demonstrate that interpretable units have high RS which are not critical to network's generalization. The RS provides new insights into the interpretation of DNNs and suggests that we need to focus on the independence and relationship of the representations.

cs.CV↗
arXiv2018

Overcoming Catastrophic Forgetting by Soft Parameter Pruning

Catastrophic forgetting is a challenge issue in continual learning when a deep neural network forgets the knowledge acquired from the former task after learning on subsequent tasks. However, existing methods try to find the joint distribution of parameters shared with all tasks. This idea can be questionable because this joint distribution may not present when the number of tasks increase. On the other hand, It also leads to "long-term" memory issue when the network capacity is limited since adding tasks will "eat" the network capacity. In this paper, we proposed a Soft Parameters Pruning (SPP) strategy to reach the trade-off between short-term and long-term profit of a learning model by freeing those parameters less contributing to remember former task domain knowledge to learn future tasks, and preserving memories about previous tasks via those parameters effectively encoding knowledge about tasks at the same time. The SPP also measures the importance of parameters by information entropy in a label free manner. The experiments on several tasks shows SPP model achieved the best performance compared with others state-of-the-art methods. Experiment results also indicate that our method is less sensitive to hyper-parameter and better generalization. Our research suggests that a softer strategy, i.e. approximate optimize or sub-optimal solution, will benefit alleviating the dilemma of memory. The source codes are available at https://github.com/lehaifeng/Learning_by_memory.

cs.LG↗
arXiv2018

Learning to Measure Change: Fully Convolutional Siamese Metric Networks for Scene Change Detection

A critical challenge problem of scene change detection is that noisy changes generated by varying illumination, shadows and camera viewpoint make variances of a scene difficult to define and measure since the noisy changes and semantic ones are entangled. Following the intuitive idea of detecting changes by directly comparing dissimilarities between a pair of features, we propose a novel fully Convolutional siamese metric Network(CosimNet) to measure changes by customizing implicit metrics. To learn more discriminative metrics, we utilize contrastive loss to reduce the distance between the unchanged feature pairs and to enlarge the distance between the changed feature pairs. Specifically, to address the issue of large viewpoint differences, we propose Thresholded Contrastive Loss (TCL) with a more tolerant strategy to punish noisy changes. We demonstrate the effectiveness of the proposed approach with experiments on three challenging datasets: CDnet, PCD2015, and VL-CMU-CD. Our approach is robust to lots of challenging conditions, such as illumination changes, large viewpoint difference caused by camera motion and zooming. In addition, we incorporate the distance metric into the segmentation framework and validate the effectiveness through visualization of change maps and feature distribution. The source code is available at https://github.com/gmayday1997/ChangeDet.

cs.CV↗
arXiv2018

A New Analysis for Support Recovery with Block Orthogonal Matching Pursuit

Compressed Sensing (CS) is a signal processing technique which can accurately recover sparse signals from linear measurements with far fewer number of measurements than those required by the classical Shannon-Nyquist theorem. Block sparse signals, i.e., the sparse signals whose nonzero coefficients occur in few blocks, arise from many fields. Block orthogonal matching pursuit (BOMP) is a popular greedy algorithm for recovering block sparse signals due to its high efficiency and effectiveness. By fully using the block sparsity of block sparse signals, BOMP can achieve very good recovery performance. This paper proposes a sufficient condition to ensure that BOMP can exactly recover the support of block $K$-sparse signals under the noisy case. This condition is better than existing ones.

cs.IT↗
arXiv2017

Inverse Perron values and connectivity of a uniform hypergraph

In this paper, we show that a uniform hypergraph $\mathcal{G}$ is connected if and only if one of its inverse Perron values is larger than $0$. We give some bounds on the bipartition width, isoperimetric number and eccentricities of $\mathcal{G}$ in terms of inverse Perron values. By using the inverse Perron values, we give an estimation of the edge connectivity of a $2$-design, and determine the explicit edge connectivity of a symmetric design. Moreover, relations between the inverse Perron values and resistance distance of a connected graph are presented.

math.CO↗
arXiv2017

An Extended Kalman Filter Enhanced Hilbert-Huang Transform in Oscillation Detection

Hilbert-Huang transform (HHT) has drawn great attention in power system analysis due to its capability to deal with dynamic signal and provide instantaneous characteristics such as frequency, damping, and amplitudes. However, its shortcomings, including mode mixing and end effects, are as significant as its advantages. A preliminary result of an extended Kalman filter (EKF) method to enhance HHT and hopefully to overcome these disadvantages is presented in this paper. The proposal first removes dynamic DC components in signals using empirical mode decomposition. Then an EKF model is applied to extract instant coefficients. Numerical results using simulated and real-world low-frequency oscillation data suggest the proposal can help to overcome the mode mixing and end effects with a properly chosen number of modes.

eess.SP↗
arXiv2017

Nonlinear focal modulation microscopy

Here we report nonlinear focal modulation microscopy (NFOMM) to achieve super-resolution imaging. Abandoning the previous persistence on minimizing the size of Gaussian emission pattern by directly narrowing (e.g. Minimizing the detection pinhole in Airyscan, Zeiss) or by indirectly peeling its outer profiles (e.g., Depleting the outer emission region in STED, stimulated emission microscopy) in pointwise scanning scenarios, we stick to a more general basis------ maximizing the system frequency shifting ability. In NFOMM, we implement a nonlinear focal modulation by applying phase modulations with high-intensity illumination, thereby extending the effective spatial-frequency bandwidth of the imaging system for reconstructing super-resolved images. NFOMM employs a spatial light modulator (SLM) for assisting pattern-modulated pointwise scanning, making the system single beam path while achieving a transverse resolution of 60 nm on imaging fluorescent nanoparticles. While exploring a relatively simple and flexible system, the imagingperformance of NFOMM is comparable with STED as evidenced in imaging nuclear pore complexes, demonstrating NFOMM is a suitable observation tool for fundamental studies in biology. Since NFOMM is implemented as an add-on module to an existing laser scanning microscope and easy to be aligned, we anticipate it will be adopted rapidly by the biological community.

physics.optics↗
arXiv2017

On the Selective and Invariant Representation of DCNN for High-Resolution Remote Sensing Image Recognition

Human vision possesses strong invariance in image recognition. The cognitive capability of deep convolutional neural network (DCNN) is close to the human visual level because of hierarchical coding directly from raw image. Owing to its superiority in feature representation, DCNN has exhibited remarkable performance in scene recognition of high-resolution remote sensing (HRRS) images and classification of hyper-spectral remote sensing images. In-depth investigation is still essential for understanding why DCNN can accurately identify diverse ground objects via its effective feature representation. Thus, we train the deep neural network called AlexNet on our large scale remote sensing image recognition benchmark. At the neuron level in each convolution layer, we analyze the general properties of DCNN in HRRS image recognition by use of a framework of visual stimulation-characteristic response combined with feature coding-classification decoding. Specifically, we use histogram statistics, representational dissimilarity matrix, and class activation mapping to observe the selective and invariance representations of DCNN in HRRS image recognition. We argue that selective and invariance representations play important roles in remote sensing images tasks, such as classification, detection, and segment. Also selective and invariance representations are significant to design new DCNN liked models for analyzing and understanding remote sensing images.

cs.CV↗
arXiv2017

A Probabilistic Embedding Clustering Method for Urban Structure Detection

Urban structure detection is a basic task in urban geography. Clustering is a core technology to detect the patterns of urban spatial structure, urban functional region, and so on. In big data era, diverse urban sensing datasets recording information like human behaviour and human social activity, suffer from complexity in high dimension and high noise. And unfortunately, the state-of-the-art clustering methods does not handle the problem with high dimension and high noise issues concurrently. In this paper, a probabilistic embedding clustering method is proposed. Firstly, we come up with a Probabilistic Embedding Model (PEM) to find latent features from high dimensional urban sensing data by learning via probabilistic model. By latent features, we could catch essential features hidden in high dimensional data known as patterns; with the probabilistic model, we can also reduce uncertainty caused by high noise. Secondly, through tuning the parameters, our model could discover two kinds of urban structure, the homophily and structural equivalence, which means communities with intensive interaction or in the same roles in urban structure. We evaluated the performance of our model by conducting experiments on real-world data and experiments with real data in Shanghai (China) proved that our method could discover two kinds of urban structure, the homophily and structural equivalence, which means clustering community with intensive interaction or under the same roles in urban space.

cs.SI↗
arXiv2017

What do We Learn by Semantic Scene Understanding for Remote Sensing imagery in CNN framework?

Recently, deep convolutional neural network (DCNN) achieved increasingly remarkable success and rapidly developed in the field of natural image recognition. Compared with the natural image, the scale of remote sensing image is larger and the scene and the object it represents are more macroscopic. This study inquires whether remote sensing scene and natural scene recognitions differ and raises the following questions: What are the key factors in remote sensing scene recognition? Is the DCNN recognition mechanism centered on object recognition still applicable to the scenarios of remote sensing scene understanding? We performed several experiments to explore the influence of the DCNN structure and the scale of remote sensing scene understanding from the perspective of scene complexity. Our experiment shows that understanding a complex scene depends on an in-depth network and multiple-scale perception. Using a visualization method, we qualitatively and quantitatively analyze the recognition mechanism in a complex remote sensing scene and demonstrate the importance of multi-objective joint semantic support.

cs.CV↗
arXiv2016

Queue Theory based Response Time Analyses for Geo-Information Processing Chain

Typical characteristics of remote sensing applications are concurrent tasks, such as those found in disaster rapid response. The existing composition approach to geographical information processing service chain, searches for an optimisation solution and is what can be deemed a "selfish" way. This way leads to problems of conflict amongst concurrent tasks and decreases the performance of all service chains. In this study, a non-cooperative game-based mathematical model to analyse the competitive relationships between tasks, is proposed. A best response function is used, to assure each task maintains utility optimisation by considering composition strategies of other tasks and quantifying conflicts between tasks. Based on this, an iterative algorithm that converges to Nash equilibrium is presented, the aim being to provide good convergence and maximise the utilisation of all tasks under concurrent task conditions. Theoretical analyses and experiments showed that the newly proposed method, when compared to existing service composition methods, has better practical utility in all tasks.

cs.DC↗
arXiv2015

Urban spatial-temporal activity structures: a New Approach to Inferring the Intra-urban Functional Regions via Social Media Check-In Data

Most existing literature focuses on the exterior temporal rhythm of human movement to infer the functional regions in a city, but they neglects the underlying interdependence between the functional regions and human activities which uncovers more detailed characteristics of regions. In this research, we proposed a novel model based on the low rank approximation (LRA) to detect the functional regions using the data from about 15 million check-in records during a yearlong period in Shanghai, China. We find a series of latent structures, called urban spatial-temporal activity structure (USTAS). While interpreting these structures, a series of outstanding underlying associations between the spatial and temporal activity patterns can be found. Moreover, we can not only reproduce the observed data with a lower dimensional representative but also simultaneously project both the spatial and temporal activity patterns in the same coordinate system. By utilizing the K-means clustering algorithm, five significant types of clusters which are directly annotated with a corresponding combination of temporal activities can be obtained. This provides a clear picture of how the groups of regions are associated with different activities at different time of day. Besides the commercial and transportation dominant area, we also detect two kinds of residential areas, the developed residential areas and the developing residential areas. We further verify the spatial distribution of these clusters in the view of urban form analysis. The results shows a high consistency with the government planning from the same periods, indicating our model is applicable for inferring the functional regions via social media check-in data, and can benefit a wide range of fields, such as urban planning, public services and location-based recommender systems and other purposes.

cs.SI↗
arXiv2014

A second-order spin-flop transition in collinear two-sublattice antiferromagnets

Identifying the nature of a spin-flop (SFO) transition, first- or second-order (FO or SO), remains a major challenge in condensed-matter physics due to the technically undistinguishable effect of misalignment between applied-field direction and the relevant antiferromagnetic (AFM) easy axis. A classical SFO transition is believed to be of FO in character. Here a mean-field theoretical calculation endowed with AFM exchange interaction (\emph{J}), easy axis anisotropy ($γ$), uniaxial single-ion exchange anisotropy (\emph{D}), and Zeeman coupling to a magnetic field parallel to the easy axis unambiguously reveals that a SO SFO transition indeed exists by virtue of its relatively lower free energy. Their equilibrium phase conditions are found to be: $D \geq 0$ (FO); $-\frac{1}{2} γ< D < 0$ (SO). Compared numerically to the associated AFM and spin-flip phases, the deduced SO SFO transition results from a negative single-ion anisotropy which is restricted to a certain range by the anisotropic exchange interaction

cond-mat.str-el↗
arXiv2014

Nonmagnetic ordering state of single-crystal SrTm$_2$O$_4$: A polarized and unpolarized neutron-scattering study

Our single-crystal polarized neutron scattering at 65 mK and powder unpolarized neutron diffraction at 0.5 K show no evidence for a long-range magnetic order and even detect no sign of diffuse magnetic neutron scattering in single-crystal SrTm2O4. The data refinements reveal that the two TmO6 octahedral distortion modes are the same as those of the TbO6 octahedra in SrTb2O4, i.e., one distortion is stronger than the other one especially at low temperatures, which is attributed probably to different crystal electric fields for the two inequivalent octahedra. Consequently, we conclude that SrTm2O4 has no magnetic order, neither long-ranged nor short-ranged, even down to 65 mK. Therefore, SrTm2O4 is a different compound from its brethren in the new family of frustrated SrRE2O4 (RE = Gd, Tb, Dy, Ho, Er, and Yb) magnets. We propose that crystal field anisotropy may dominate over weak dipolar spin interactions in SrTm2O4, leading to a virtually nonmagnetic ordering state.

cond-mat.str-el↗