arXiv subjects

Xudong Ma

Publications and source records attributed to Xudong Ma.

33 records · Page 2Linked to original sources

arXiv2022

BiFSMN: Binary Neural Network for Keyword Spotting

The deep neural networks, such as the Deep-FSMN, have been widely studied for keyword spotting (KWS) applications. However, computational resources for these networks are significantly constrained since they usually run on-call on edge devices. In this paper, we present BiFSMN, an accurate and extreme-efficient binary neural network for KWS. We first construct a High-frequency Enhancement Distillation scheme for the binarization-aware training, which emphasizes the high-frequency information from the full-precision network's representation that is more crucial for the optimization of the binarized network. Then, to allow the instant and adaptive accuracy-efficiency trade-offs at runtime, we also propose a Thinnable Binarization Architecture to further liberate the acceleration potential of the binarized network from the topology perspective. Moreover, we implement a Fast Bitwise Computation Kernel for BiFSMN on ARMv8 devices which fully utilizes registers and increases instruction throughput to push the limit of deployment efficiency. Extensive experiments show that BiFSMN outperforms existing binarization methods by convincing margins on various datasets and is even comparable with the full-precision counterpart (e.g., less than 3% drop on Speech Commands V1-12). We highlight that benefiting from the thinnable architecture and the optimized 1-bit implementation, BiFSMN can achieve an impressive 22.3x speedup and 15.5x storage-saving on real-world edge hardware. Our code is released at https://github.com/htqin/BiFSMN.

cs.CL↗

arXiv2022

ICIP 2022 Challenge on Parasitic Egg Detection and Classification in Microscopic Images: Dataset, Methods and Results

Manual examination of faecal smear samples to identify the existence of parasitic eggs is very time-consuming and can only be done by specialists. Therefore, an automated system is required to tackle this problem since it can relate to serious intestinal parasitic infections. This paper reviews the ICIP 2022 Challenge on parasitic egg detection and classification in microscopic images. We describe a new dataset for this application, which is the largest dataset of its kind. The methods used by participants in the challenge are summarised and discussed along with their results.

cs.CV↗

arXiv2022

Unsupervised Image Fusion Using Deep Image Priors

A significant number of researchers have applied deep learning methods to image fusion. However, most works require a large amount of training data or depend on pre-trained models or frameworks to capture features from source images. This is inevitably hampered by a shortage of training data or a mismatch between the framework and the actual problem. Deep Image Prior (DIP) has been introduced to exploit convolutional neural networks' ability to synthesize the 'prior' in the input image. However, the original design of DIP is hard to be generalized to multi-image processing problems, particularly for image fusion. Therefore, we propose a new image fusion technique that extends DIP to fusion tasks formulated as inverse problems. Additionally, we apply a multi-channel approach to enhance DIP's effect further. The evaluation is conducted with several commonly used image fusion assessment metrics. The results are compared with state-of-the-art image fusion methods. Our method outperforms these techniques for a range of metrics. In particular, it is shown to provide the best objective results for most metrics when applied to medical images.

cs.CV↗

arXiv2016

Multi-Camera Occlusion and Sudden-Appearance-Change Detection Using Hidden Markovian Chains

This paper was originally submitted to Xinova as a response to a Request for Invention (RFI) on new event monitoring methods. In this paper, a new object tracking algorithm using multiple cameras for surveillance applications is proposed. The proposed system can detect sudden-appearance-changes and occlusions using a hidden Markovian statistical model. The experimental results confirm that our system detect the sudden-appearance changes and occlusions reliably.

cs.CV↗

arXiv2014

Joint Write-Once-Memory and Error-Control Codes

Write-Once-Memory (WOM) is a model for many modern non-volatile memories, such as flash memories. Recently, several capacity-achieving WOM coding schemes have been proposed based on polar coding. Due to the fact that practical computer memory systems always contain noises, a nature question to ask next is how may we generalize these coding schemes, such that they may also have the error-control capabilities. In this paper, we discuss a joint WOM and error-control coding scheme, which is a generalization of the capacity-achieving WOM codes based on source polarization. In this paper, we prove a sufficient and necessary condition for the noisy reading channel being less noisy than the test channel in data encoding in the polar WOM coding. Such a sufficient and necessary condition is usually satisfied in reality. As a consequence of the sufficient and necessary condition, the high entropy set related to the noisy channel is usually strictly contained in the high entropy set related to the test channel in data encoding. Therefore the low-complexity polar joint WOM and error-control codes are sufficient for most practical coding scenarios.

cs.IT↗

arXiv2014

Write-Once-Memory Codes by Source Polarization

We propose a new Write-Once-Memory (WOM) coding scheme based on source polarization. By applying a source polarization transformation on the to-be-determined codeword, the proposed WOM coding scheme encodes information into the bits in the high-entropy set. We prove in this paper that the proposed WOM codes are capacity-achieving. WOM codes have found many applications in modern data storage systems, such as flash memories.

cs.IT↗

arXiv2012

Symbol-Index-Feedback Polar Coding Schemes for Low-Complexity Devices

Recently, a new class of error-control codes, the polar codes, have attracted much attention. The polar codes are the first known class of capacity-achieving codes for many important communication channels. In addition, polar codes have low-complexity encoding algorithms. Therefore, these codes are favorable choices for low-complexity devices, for example, in ubiquitous computing and sensor networks. However, the polar codes fall short in terms of finite-length error probabilities, compared with the state-of-the-art codes, such as the low-density parity-check codes. In this paper, in order to improve the error probabilities of the polar codes, we propose novel interactive coding schemes using receiver feedbacks based on polar codes. The proposed coding schemes have very low computational complexities at the transmitter side. By experimental results, we show that the proposed coding schemes achieve significantly lower error probabilities.

cs.IT↗

arXiv2012

On Coding Efficiency for Flash Memories

Recently, flash memories have become a competitive solution for mass storage. The flash memories have rather different properties compared with the rotary hard drives. That is, the writing of flash memories is constrained, and flash memories can endure only limited numbers of erases. Therefore, the design goals for the flash memory systems are quite different from these for other memory systems. In this paper, we consider the problem of coding efficiency. We define the "coding-efficiency" as the amount of information that one flash memory cell can be used to record per cost. Because each flash memory cell can endure a roughly fixed number of erases, the cost of data recording can be well-defined. We define "payload" as the amount of information that one flash memory cell can represent at a particular moment. By using information-theoretic arguments, we prove a coding theorem for achievable coding rates. We prove an upper and lower bound for coding efficiency. We show in this paper that there exists a fundamental trade-off between "payload" and "coding efficiency". The results in this paper may provide useful insights on the design of future flash memory systems.

cs.IT↗

arXiv2010

Large-Scale Clustering Based on Data Compression

This paper considers the clustering problem for large data sets. We propose an approach based on distributed optimization. The clustering problem is formulated as an optimization problem of maximizing the classification gain. We show that the optimization problem can be reformulated and decomposed into small-scale sub optimization problems by using the Dantzig-Wolfe decomposition method. Generally speaking, the Dantzig-Wolfe method can only be used for convex optimization problems, where the duality gaps are zero. Even though, the considered optimization problem in this paper is non-convex, we prove that the duality gap goes to zero, as the problem size goes to infinity. Therefore, the Dantzig-Wolfe method can be applied here. In the proposed approach, the clustering problem is iteratively solved by a group of computers coordinated by one center processor, where each computer solves one independent small-scale sub optimization problem during each iteration, and only a small amount of data communication is needed between the computers and center processor. Numerical results show that the proposed approach is effective and efficient.

cs.LG↗

arXiv2010

Performance Analysis for Data Compression Based Signal Classification Methods

In this paper, we present an information theoretic analysis of the blind signal classification algorithm. We show that the algorithm is equivalent to a Maximum A Posteriori (MAP) estimator based on estimated parametric probability models. We prove a lower bound on the error exponents of the parametric model estimation. It is shown that the estimated model parameters converge in probability to the true model parameters except some small bias terms.

cs.IT↗

arXiv2009

Equalization for Non-Coherent UWB Systems with Approximate Semi-Definite Programming

In this paper, we propose an approximate semi-definite programming framework for demodulation and equalization of non-coherent ultra-wide-band communication systems with inter-symbol-interference. It is assumed that the communication systems follow non-linear second-order Volterra models. We formulate the demodulation and equalization problems as semi-definite programming problems. We propose an approximate algorithm for solving the formulated semi-definite programming problems. Compared with the existing non-linear equalization approaches, the proposed semi-definite programming formulation and approximate solving algorithm have low computational complexity and storage requirements. We show that the proposed algorithm has satisfactory error probability performance by simulation results. The proposed non-linear equalization approach can be adopted for a wide spectrum of non-coherent ultra-wide-band systems, due to the fact that most non-coherent ultra-wide-band systems with inter-symbol-interference follow non-linear second-order Volterra signal models.

cs.IT↗

arXiv2009

Reduced Complexity Demodulation and Equalization Scheme for Differential Impulse Radio UWB Systems with ISI

In this paper, we consider the demodulation and equalization problem of differential Impulse Radio (IR) Ultra-WideBand (UWB) Systems with Inter-Symbol-Interference (ISI). The differential IR UWB systems have been extensively discussed recently. The advantage of differential IR UWB systems include simple receiver frontend structure. One challenge in the demodulation and equalization of such systems with ISI is that the systems have a rather complex model. The input and output signals of the systems follow a second-order Volterra model. Furthermore, the noise at the output is data dependent. In this paper, we propose a reduced-complexity joint demodulation and equalization algorithm. The algorithm is based on reformulating the nearest neighborhood decoding problem into a mixed quadratic programming and utilizing a semi-definite relaxation. The numerical results show that the proposed demodulation and equalization algorithm has low computational complexity, and at the same time, has almost the same error probability performance compared with the maximal likelihood decoding algorithm.

cs.IT↗

arXiv2009

Novel Blind Signal Classification Method Based on Data Compression

This paper proposes a novel algorithm for signal classification problems. We consider a non-stationary random signal, where samples can be classified into several different classes, and samples in each class are identically independently distributed with an unknown probability distribution. The problem to be solved is to estimate the probability distributions of the classes and the correct membership of the samples to the classes. We propose a signal classification method based on the data compression principle that the accurate estimation in the classification problems induces the optimal signal models for data compression. The method formulates the classification problem as an optimization problem, where a so called {"classification gain"} is maximized. In order to circumvent the difficulties in integer optimization, we propose a continuous relaxation based algorithm. It is proven in this paper that asymptotically vanishing optimality loss is incurred by the continuous relaxation. We show by simulation results that the proposed algorithm is effective, robust and has low computational complexity. The proposed algorithm can be applied to solve various multimedia signal segmentation, analysis, and pattern recognition problems.

cs.IT↗

arXiv2006

Constructing LDPC Codes by 2-Lifts

We propose a new low-density parity-check code construction scheme based on 2-lifts. The proposed codes have an advantage of admitting efficient hardware implementations. With the motivation of designing codes with low error floors, we present an analysis of the low-weight stopping set distributions of the proposed codes. Based on this analysis, we propose design criteria for designing codes with low error floors. Numerical results show that the resulting codes have low error probabilities over binary erasure channels.

cs.IT↗

arXiv2006

Low-Density Parity-Check Code with Fast Decoding Speed

Low-Density Parity-Check (LDPC) codes received much attention recently due to their capacity-approaching performance. The iterative message-passing algorithm is a widely adopted decoding algorithm for LDPC codes \cite{Kschischang01}. An important design issue for LDPC codes is designing codes with fast decoding speed while maintaining capacity-approaching performance. In another words, it is desirable that the code can be successfully decoded in few number of decoding iterations, at the same time, achieves a significant portion of the channel capacity. Despite of its importance, this design issue received little attention so far. In this paper, we address this design issue for the case of binary erasure channel. We prove that density-efficient capacity-approaching LDPC codes satisfy a so called "flatness condition". We show an asymptotic approximation to the number of decoding iterations. Based on these facts, we propose an approximated optimization approach to finding the codes with good decoding speed. We further show that the optimal codes in the sense of decoding speed are "right-concentrated". That is, the degrees of check nodes concentrate around the average right degree.

cs.IT↗