arXiv subjects

Wanquan Liu

Publications and source records attributed to Wanquan Liu.

47 records · Page 3Linked to original sources

arXiv2020

iffDetector: Inference-aware Feature Filtering for Object Detection

Modern CNN-based object detectors focus on feature configuration during training but often ignore feature optimization during inference. In this paper, we propose a new feature optimization approach to enhance features and suppress background noise in both the training and inference stages. We introduce a generic Inference-aware Feature Filtering (IFF) module that can easily be combined with modern detectors, resulting in our iffDetector. Unlike conventional open-loop feature calculation approaches without feedback, the IFF module performs closed-loop optimization by leveraging high-level semantics to enhance the convolutional features. By applying Fourier transform analysis, we demonstrate that the IFF module acts as a negative feedback that theoretically guarantees the stability of feature learning. IFF can be fused with CNN-based object detectors in a plug-and-play manner with negligible computational cost overhead. Experiments on the PASCAL VOC and MS COCO datasets demonstrate that our iffDetector consistently outperforms state-of-the-art methods by significant margins\footnote{The test code and model are anonymously available in https://github.com/anonymous2020new/iffDetector }.

cs.CV↗

arXiv2020

Regularizing Semi-supervised Graph Convolutional Networks with a Manifold Smoothness Loss

Existing graph convolutional networks focus on the neighborhood aggregation scheme. When applied to semi-supervised learning, they often suffer from the overfitting problem as the networks are trained with the cross-entropy loss on a small potion of labeled data. In this paper, we propose an unsupervised manifold smoothness loss defined with respect to the graph structure, which can be added to the loss function as a regularization. We draw connections between the proposed loss with an iterative diffusion process, and show that minimizing the loss is equivalent to aggregate neighbor predictions with infinite layers. We conduct experiments on multi-layer perceptron and existing graph networks, and demonstrate that adding the proposed loss can improve the performance consistently.

cs.LG↗

arXiv2019

A Novel Euler's Elastica based Segmentation Approach for Noisy Images via using the Progressive Hedging Algorithm

Euler's Elastica based unsupervised segmentation models have strong capability of completing the missing boundaries for existing objects in a clean image, but they are not working well for noisy images. This paper aims to establish a Euler's Elastica based approach that properly deals with random noises to improve the segmentation performance for noisy images. We solve the corresponding optimization problem via using the progressive hedging algorithm (PHA) with a step length suggested by the alternating direction method of multipliers (ADMM). Technically, all the simplified convex versions of the subproblems derived from the major framework of PHA can be obtained by using the curvature weighted approach and the convex relaxation method. Then an alternating optimization strategy is applied with the merits of using some powerful accelerating techniques including the fast Fourier transform (FFT) and generalized soft threshold formulas. Extensive experiments have been conducted on both synthetic and real images, which validated some significant gains of the proposed segmentation models and demonstrated the advantages of the developed algorithm.

cs.CV↗

arXiv2019

Semi-supervised Learning on Graph with an Alternating Diffusion Process

Graph-based semi-supervised learning usually involves two separate stages, constructing an affinity graph and then propagating labels for transductive inference on the graph. It is suboptimal to solve them independently, as the correlation between the affinity graph and labels are not fully exploited. In this paper, we integrate the two stages into one unified framework by formulating the graph construction as a regularized function estimation problem similar to label propagation. We propose an alternating diffusion process to solve the two problems simultaneously, which allows us to learn the graph and unknown labels in an iterative fashion. With the proposed framework, we are able to adequately leverage both the given labels and estimated labels to construct a better graph, and effectively propagate labels on such a dynamic graph updated simultaneously with the newly obtained labels. Extensive experiments on various real-world datasets have demonstrated the superiority of the proposed method compared to other state-of-the-art methods.

cs.CV↗

arXiv2016

Sparse Subspace Clustering via Diffusion Process

Subspace clustering refers to the problem of clustering high-dimensional data that lie in a union of low-dimensional subspaces. State-of-the-art subspace clustering methods are based on the idea of expressing each data point as a linear combination of other data points while regularizing the matrix of coefficients with L1, L2 or nuclear norms for a sparse solution. L1 regularization is guaranteed to give a subspace-preserving affinity (i.e., there are no connections between points from different subspaces) under broad theoretical conditions, but the clusters may not be fully connected. L2 and nuclear norm regularization often improve connectivity, but give a subspace-preserving affinity only for independent subspaces. Mixed L1, L2 and nuclear norm regularization could offer a balance between the subspace-preserving and connectedness properties, but this comes at the cost of increased computational complexity. This paper focuses on using L1 norm and alleviating the corresponding connectivity problem by a simple yet efficient diffusion process on subspace affinity graphs. Without adding any tuning parameter , our method can achieve state-of-the-art clustering performance on Hopkins 155 and Extended Yale B data sets.

cs.CV↗

arXiv2014

On the $k$-error linear complexity for $p^n$-periodic binary sequences via hypercube theory

The linear complexity and the $k$-error linear complexity of a binary sequence are important security measures for key stream strength. By studying binary sequences with the minimum Hamming weight, a new tool named as hypercube theory is developed for $p^n$-periodic binary sequences. In fact, hypercube theory is based on a typical sequence decomposition and it is a very important tool in investigating the critical error linear complexity spectrum proposed by Etzion et al. To demonstrate the importance of hypercube theory, we first give a standard hypercube decomposition based on a well-known algorithm for computing linear complexity and show that the linear complexity of the first hypercube in the decomposition is equal to the linear complexity of the original sequence. Second, based on such decomposition, we give a complete characterization for the first decrease of the linear complexity for a $p^n$-periodic binary sequence $s$. This significantly improves the current existing results in literature. As to the importance of the hypercube, we finally derive a counting formula for the $m$-hypercubes with the same linear complexity.

cs.CR↗

arXiv2014

Structure Analysis on the $k$-error Linear Complexity for $2^n$-periodic Binary Sequences

In this paper, in order to characterize the critical error linear complexity spectrum (CELCS) for $2^n$-periodic binary sequences, we first propose a decomposition based on the cube theory. Based on the proposed $k$-error cube decomposition, and the famous inclusion-exclusion principle, we obtain the complete characterization of $i$th descent point (critical point) of the k-error linear complexity for $i=2,3$. Second, by using the sieve method and Games-Chan algorithm, we characterize the second descent point (critical point) distribution of the $k$-error linear complexity for $2^n$-periodic binary sequences. As a consequence, we obtain the complete counting functions on the $k$-error linear complexity of $2^n$-periodic binary sequences as the second descent point for $k=3,4$. This is the first time for the second and the third descent points to be completely characterized. In fact, the proposed constructive approach has the potential to be used for constructing $2^n$-periodic binary sequences with the given linear complexity and $k$-error linear complexity (or CELCS), which is a challenging problem to be deserved for further investigation in future.

cs.CR↗

arXiv2013

The 4-error linear complexity distribution for $2^n$-periodic binary sequences

By using the sieve method of combinatorics, we study $k$-error linear complexity distribution of $2^n$-periodic binary sequences based on Games-Chan algorithm. For $k=4,5$, the complete counting functions on the $k$-error linear complexity of $2^n$-periodic balanced binary sequences (with linear complexity less than $2^n$) are presented. As a consequence of the result, the complete counting functions on the 4-error linear complexity of $2^n$-periodic binary sequences (with linear complexity $2^n$ or less than $2^n$) are obvious. Generally, the complete counting functions on the $k$-error linear complexity of $2^n$-periodic binary sequences can be obtained with a similar approach.

cs.CR↗

arXiv2013

On the $k$-error linear complexity for $2^n$-periodic binary sequences via Cube Theory

The linear complexity and k-error linear complexity of a sequence have been used as important measures of keystream strength, hence designing a sequence with high linear complexity and $k$-error linear complexity is a popular research topic in cryptography. In this paper, the concept of stable $k$-error linear complexity is proposed to study sequences with stable and large $k$-error linear complexity. In order to study k-error linear complexity of binary sequences with period $2^n$, a new tool called cube theory is developed. By using the cube theory, one can easily construct sequences with the maximum stable $k$-error linear complexity. For such purpose, we first prove that a binary sequence with period $2^n$ can be decomposed into some disjoint cubes and further give a general decomposition approach. Second, it is proved that the maximum $k$-error linear complexity is $2^n-(2^l-1)$ over all $2^n$-periodic binary sequences, where $2^{l-1}\le k<2^{l}$. Thirdly, a characterization is presented about the $t$th ($t>1$) decrease in the $k$-error linear complexity for a $2^n$-periodic binary sequence $s$ and this is a continuation of Kurosawa et al. recent work for the first decrease of k-error linear complexity. Finally, A counting formula for $m$-cubes with the same linear complexity is derived, which is equivalent to the counting formula for $k$-error vectors. The counting formula of $2^n$-periodic binary sequences which can be decomposed into more than one cube is also investigated, which extends an important result by Etzion et al..

cs.CR↗

arXiv2011

Characterization of $2^n$-periodic binary sequences with fixed 3-error or 4-error linear complexity

The linear complexity and the $k$-error linear complexity of a sequence have been used as important security measures for key stream sequence strength in linear feedback shift register design. By using the sieve method of combinatorics, the $k$-error linear complexity distribution of $2^n$-periodic binary sequences is investigated based on Games-Chan algorithm. First, for $k=2,3$, the complete counting functions on the $k$-error linear complexity of $2^n$-periodic binary sequences with linear complexity less than $2^n$ are characterized. Second, for $k=3,4$, the complete counting functions on the $k$-error linear complexity of $2^n$-periodic binary sequences with linear complexity $2^n$ are presented. Third, for $k=4,5$, the complete counting functions on the $k$-error linear complexity of $2^n$-periodic binary sequences with linear complexity less than $2^n$ are derived. As a consequence of these results, the counting functions for the number of $2^n$-periodic binary sequences with the 3-error linear complexity are obtained, and the complete counting functions on the 4-error linear complexity of $2^n$-periodic binary sequences are obvious.

cs.CR↗

arXiv2011

The $k$-error linear complexity distribution for $2^n$-periodic binary sequences

The linear complexity and the $k$-error linear complexity of a sequence have been used as important security measures for key stream sequence strength in linear feedback shift register design. By studying the linear complexity of binary sequences with period $2^n$, one could convert the computation of $k$-error linear complexity into finding error sequences with minimal Hamming weight. Based on Games-Chan algorithm, the $k$-error linear complexity distribution of $2^n$-periodic binary sequences is investigated in this paper. First, for $k=2,3$, the complete counting functions on the $k$-error linear complexity of $2^n$-periodic balanced binary sequences (with linear complexity less than $2^n$) are characterized. Second, for $k=3,4$, the complete counting functions on the $k$-error linear complexity of $2^n$-periodic binary sequences with linear complexity $2^n$ are presented. Third, as a consequence of these results, the counting functions for the number of $2^n$-periodic binary sequences with the $k$-error linear complexity for $k = 2$ and 3 are obtained. Further more, an important result in a recent paper is proved to be not completely correct.

cs.CR↗