Songfeng Zheng

Cooperative Sensing via Sequential Detection

Efficient and reliable spectrum sensing plays a critical role in cognitive radio networks. This paper presents a cooperative sequential detection scheme to reduce the average sensing time that is required to reach a detection decision. In the scheme, each cognitive radio computes the log-likelihood ratio for its every measurement, and the base station sequentially accumulates these log-likelihood statistics and determines whether to stop making measurement. The paper studies how to implement the scheme in a robust manner when the assumed signal models have unknown parameters, such as signal strength and noise variance. These ideas are illustrated through two examples in spectrum sensing. One assumes both the signal and noise are Gaussian distributed, while the other assumes the target signal is deterministic.

Automated Extraction of the Cortical Sulci Based on a Supervised Learning Approach

It is important to detect and extract the major cortical sulci from brain images, but manually annotating these sulci is a time-consuming task and requires the labeler to follow complex protocols , . This paper proposes a learning-based algorithm for automated extraction of the major cortical sulci from magnetic resonance imaging (MRI) volumes and cortical surfaces. Unlike alternative methods for detecting the major cortical sulci, which use a small number of predefined rules based on properties of the cortical surface such as the mean curvature, our approach learns a discriminative model using the probabilistic boosting tree algorithm (PBT) . PBT is a supervised learning approach which selects and combines hundreds of features at different scales, such as curvatures, gradients and shape index. Our method can be applied to either MRI volumes or cortical surfaces. It first outputs a probability map which indicates how likely each voxel lies on a major sulcal curve. Next, it applies dynamic programming to extract the best curve based on the probability map and a shape prior. The algorithm has almost no parameters to tune for extracting different major sulci. It is very fast (it runs in under 1 min per sulcus including the time to compute the discriminative models) due to efficient implementation of the features (e.g., using the integral volume to rapidly compute the responses of 3-D Haar filters). Because the algorithm can be applied to MRI volumes directly, there is no need to perform preprocessing such as tissue segmentation or mapping to a canonical space. The learning aspect of our approach makes the system very flexible and general. For illustration, we use volumes of the right hemisphere with several major cortical sulci manually labeled. The algorithm is tested on two groups of data, including some brains from patients with Williams Syndrome, and the results are very encouraging

Detecting Object Boundaries Using Low-, Mid-, and High-level Information

Object boundary detection and segmentation is a central problem in computer vision. The importance of combining low-level, mid-level, and high-level cues has been realized in recent literature. However, it is unclear how to efficiently and effectively engage and fuse different levels of information. In this paper, we emphasize a learning based approach to explore different levels of information, both implicitly and explicitly. First, we learn low-level cues for object boundaries and interior regions using a probabilistic boosting tree (PBT). Second, we learn short and long range context information based on the results from the first stage. Both stages implicitly contain object-specific information such as texture and local geometry, and it is shown that this implicit knowledge is extremely powerful. Third, we use high-level shape information explicitly to further refine the object segmentation and to parse the object into components. The algorithm is trained and tested on a challenging dataset of horses [2], and the results obtained are very encouraging compared with other approaches. In detailed experiments we show significantly better performance (e.g. F-values of 0.75 compared to 0.66) than the best comparable reported performance on this dataset. Furthermore, the system only needs 1.5 minutes for a typical image. Although our system is illustrated on horse images, the approach can be directly applied to detecting/segmenting other types of objects.

Integrating local distribution information with level set for boundary extraction

This paper presents a general object boundary extraction model for piecewise smooth images, which incorporates local intensity distribution information into an edge-based implicit active contour. Unlike traditional edge-based active contours that use gradient to detect edges, our model derives the neighborhood distribution and edge information with two different region-based operators: a Gaussian mixture model (GMM)-based intensity distribution estimator and the Hueckel operator. We propose the local distribution fitting model for more accurate segmentation, which incorporates the operator outcomes into the recent local binary fitting (LBF) model. The GMM and the Hueckel model parameters are estimated before contour evolution, which enables the use of the proposed model without the need for initial contour selection, i.e., the level set function is initialized with a random constant instead of a distance map. Thus our model essentially alleviates the initialization sensitivity problem of most active contours. Experiments on synthetic and real images show the improved performance of our approach over the LBF model.

Cooperative spectrum sensing via sequential detection for cognitive radio networks

Efficient and reliable spectrum sensing plays a critical role in cognitive radio networks. This paper presents a cooperative sequential detection scheme tominimize the average sensing time that is required to reach a detection decision. In the scheme, each cognitive radio computes the Log-Likelihood ratio for its every measurement, and the base station sequentially accumulates these Log-Likelihood statistics and determines whether to stop making measurement. The average number of required samples depends on the Kullback-Leibler distance between the distributions of the two hypotheses under test. This suggests a criterion for selecting the most efficient radios to facilitate spectrum sensing. The paper also studies how to implement the scheme in a robust manner when the assumed statistical models have uncertainties. These ideas are illustrated through an example that assumes both the signal and noise are Gaussian distributed.

Gradient descent algorithms for quantile regression with smooth approximation

Gradient based optimization methods often converge quickly to a local optimum. However, the check loss function used by quantile regression model is not everywhere differentiable, which prevents the gradient based optimization methods from being applicable. As such, this paper introduces a smooth function to approximate the check loss function so that the gradient based optimization methods could be employed for fitting quantile regression model. The properties of the smooth approximation are discussed. Two algorithms are proposed for minimizing the smoothed objective function. The first method directly applies gradient descent, resulting the gradient descent smooth quantile regression model; the second approach minimizes the smoothed objective function in the framework of functional gradient descent by changing the fitted model along the negative gradient direction in each iteration, which yields boosted smooth quantile regression algorithm. Extensive experiments on simulated data and real-world data show that, compared to alternative quantile regression models, the proposed smooth quantile regression algorithms can achieve higher prediction accuracy and are more efficient in removing noninformative predictors.

QBoost: Predicting quantiles with boosting for regression and binary classification

In the framework of functional gradient descent/ascent, this paper proposes Quantile Boost (QBoost) algorithms which predict quantiles of the interested response for regression and binary classification. Quantile Boost Regression performs gradient descent in functional space to minimize the objective function used by quantile regression (QReg). In the classification scenario, the class label is defined via a hidden variable, and the quantiles of the class label are estimated by fitting the corresponding quantiles of the hidden variable. An equivalent form of the definition of quantile is introduced, whose smoothed version is employed as the objective function, and then maximized by functional gradient ascent to obtain the Quantile Boost Classification algorithm. Extensive experimentation and detailed analysis show that QBoost performs better than the original QReg and other alternatives for regression and binary classification. Furthermore, QBoost is capable of solving problems in high dimensional space and is more robust to noisy predictors.

Learning based coarse-to-fine image registration

This paper describes a coarse-to-fine learning based image registration algorithm which has particular advantages in dealing with multi-modality images. Many existing image registration algorithms use a few designed terms or mutual information to measure the similarity between image pairs. Instead, we push the learning aspect by selecting and fusing a large number of features for measuring the similarity. Moreover, the similarity measure is carried in a coarse-to-fine strategy: global similarity measure is first performed to roughly locate the component, we then learn/compute similarity on the local image patches to capture the fine level information. When estimating the transformation parameters, we also engage a coarse-to-fine strategy. Off-the-shelf interest point detectors such as SIFT have degraded results on medical images. We further push the learning idea to extract the main structures/landmarks. Our algorithm is illustrated on three applications: (1) registration of mouse brain images of different modalities, (2) registering human brain image of MRI T1 and T2 images, (3) faces of different expressions. We show greatly improved results over the existing algorithms based on either mutual information or geometric structures.

A fast algorithm for training support vector regression via smoothed primal function minimization

The support vector regression (SVR) model is usually fitted by solving a quadratic programming problem, which is computationally expensive. To improve the computational efficiency, we propose to directly minimize the objective function in the primal form. However, the loss function used by SVR is not differentiable, which prevents the well-developed gradient based optimization methods from being applicable. As such, we introduce a smooth function to approximate the original loss function in the primal form of SVR, which transforms the original quadratic programming into a convex unconstrained minimization problem. The properties of the proposed smoothed objective function are discussed and we prove that the solution of the smoothly approximated model converges to the original SVR solution. A conjugate gradient algorithm is designed for minimizing the proposed smoothly approximated objective function in a sequential minimization manner. Extensive experiments on real-world datasets show that, compared to the quadratic programming based SVR, the proposed approach can achieve similar prediction accuracy with significantly improved computational efficiency, specifically, it is hundreds of times faster for linear SVR model and multiple times faster for nonlinear SVR model.

An experimental comparison of gene selection by Lasso and Dantzig selector for cancer classification

Selecting a subset of genes with strong discriminative power is a very important step in classification problems based on gene expression data. Lasso and Dantzig selector are known to have automatic variable selection ability in linear regression analysis. This paper applies Lasso and Dantzig selector to select the most informative genes for representing the probability of an example being positive as a linear function of the gene expression data. The selected genes are further used to fit different classifiers for cancer classification. Comparative experiments were conducted on six publicly available cancer datasets, and the detailed comparison results show that in general, Lasso is more capable than Dantzig selector at selecting informative genes for cancer classification.