Seniha Esen Yuksel

Twenty Years of Mixture of Experts

In this paper, we provide a comprehensive survey of the mixture of experts (ME). We discuss the fundamental models for regression and classification and also their training with the expectation-maximization algorithm. We follow the discussion with improvements to the ME model and focus particularly on the mixtures of Gaussian process experts. We provide a review of the literature for other training methods, such as the alternative localized ME training, and cover the variational learning of ME in detail. In addition, we describe the model selection literature which encompasses finding the optimum number of experts, as well as the depth of the tree. We present the advances in ME in the classification area and present some issues concerning the classification model. We list the statistical properties of ME, discuss how the model has been modified over the years, compare ME to some popular algorithms, and list several applications. We conclude our survey with future directions and provide a list of publicly available datasets and a list of publicly available software that implement ME. Finally, we provide examples for regression and classification. We believe that the study described in this paper will provide quick access to the relevant literature for researchers and practitioners who would like to improve or use ME, and that it will stimulate further studies in ME.

Hierarchical Methods for Landmine Detection with Wideband Electro-Magnetic Induction and Ground Penetrating Radar Multi-Sensor Systems

A variety of algorithms are presented and employed in a hierarchical fashion to discriminate both anti-tank (AT) and anti-personnel (AP) landmines using data collected from wideband electromagnetic induction (WEMI) and ground penetrating radar (GPR) sensors mounted on a robotic platform. The two new algorithms for WEMI are based on the In-phase vs. Quadrature plot (the Argand diagram) of the complex measurement obtained at a single spatial location. The angle prototype match method uses the sequence of angles as a feature vector. Prototypes are constructed from these feature vectors and used to assign mine confidence to a test sample. The angle model based KNN method uses a two parameter model; where the parameters are fit to the In-phase and Quadrature data. For the GPR data, the Linear Prediction Processing and Spectral Features are calculated. All four features from WEMI and GPR are used in a Hierarchical Mixture of Experts model to increase the landmine detection rate. The EM algorithm is used to estimate the parameters of the hierarchical mixture. Instead of a two way mine/non-mine decision, the HME structure is trained to make a five way decision which aids in the detection of the low metal anti personnel mines.

A new CAD system for the evaluation of kidney diseases using DCE-MRI

Acute rejection is the most common reason of graft failure after kidney transplantation, and early detection is crucial to survive the transplanted kidney function. In this paper, we introduce a new approach for the automatic classification of normal and acute rejection transplants from Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI). The proposed algorithm consists of three main steps; the first step isolates the kidney from the surrounding anatomical structures by evolving a deformable model based on two density functions; the first function describes the distribution of the gray level inside and outside the kidney region and the second function describes the prior shape of the kidney. In the second step, a new nonrigid registration approach is employed to account for the motion of the kidney due to patient breathing. To validate our registration approach, we use a simulation of deformations based on biomechanical modelling of the kidney tissue using the finite element method (F.E.M.). Finally, the perfusion curves that show the transportation of the contrast agent into the tissue are obtained from the cortex and used in the classification of normal and acute rejection transplants. Applications of the proposed approach yield promising results that would, in the near future, replace the use of current technologies such as nuclear imaging and ultrasonography, which are not specific enough to determine the type of kidney dysfunction.

2D and 3D shape based segmentation using deformable models

A novel shape based segmentation approach is proposed by modifying the external energy component of a deformable model. The proposed external energy component depends not only on the gray level of the images but also on the shape information which is obtained from the signed distance maps of objects in a given data set. The gray level distribution and the signed distance map of the points inside and outside the object of interest are accurately estimated by modelling the empirical density function with a linear combination of discrete Gaussians (LCDG) with positive and negative components. Experimental results on the segmentation of the kidneys from low-contrast DCE-MRI and on the segmentation of the ventricles from brain MRIs show how the approach is accurate in segmenting 2-D and 3-D data sets. The 2D results for the kidney segmentation have been validated by a radiologist and the 3D results of the ventricle segmentation have been validated with a geometrical phantom.

A framework for the detection of acute renal rejection with dynamic contrast enhanced magnetic resonance imaging

Acute rejection is the most common reason of graft failure after kidney transplantation, and early detection is crucial to survive the transplanted kidney function. In this paper we introduce a new approach for the automatic classification of normal and acute rejection transplants from dynamic contrast enhanced magnetic resonance imaging (DCE-MRI). The proposed algorithm consists of three main steps; the first step isolates the kidney from the surrounding anatomical structures by evolving a deformable model based on two density functions; the first function describes the distribution of the gray level inside and outside the kidney region and the second function describes the prior shape of the kidney. In the second step, nonrigid-registration algorithms are employed to account for the motion of the kidney due to patient breathing, and finally, the perfusion curves that show the transportation of the contrast agent into the tissue are obtained from the cortex and used in the classification of normal and acute rejection transplants. Applications of the proposed approach yield promising results that would, in the near future, replace the use of current technologies such as nuclear imaging and ultrasonography, which are not specific enough to determine the type of kidney dysfunction

Application Of Deformable Models For The Detection Of Acute Renal Rejection

Acute rejection is the most common reason for graft failure after kidney transplantation, and early detection is crucial to survival of function in the transplanted kidney. In this study we introduce a new framework for automatic classification of normal and acute rejection transplants from Dynamic Contrast Enhanced Magnetic Resonance Images (DCE-MRI). The proposed framework consists of three main steps. The first isolates the kidney from the surrounding anatomical structures by evolving a deformable model based on two density functions; the first function describes the distribution of the gray level inside and outside the kidney region and the second describes the prior shape of the kidney. In the second step, nonrigid registration algorithms are employed to account for the motion of the kidney due to the patient’s breathing. In the third step, the perfusion curves that show transportation of the contrast agent into the tissue are obtained from the segmented cortex of the whole image sequence of the patient. In the final step, we collect four features from these curves and use Bayesian classifiers to distinguish between acute rejection and normal transplants. Applications of the proposed approach yield promising results that would, in the near future, replace the use of current technologies such as nuclear imaging and ultrasonography, which are not specific enough to determine the type of kidney dysfunction.

Multiple-Instance Hidden Markov Models With Applications to Landmine Detection

A novel multiple-instance hidden Markov model (MI-HMM) is introduced for classification of time-series data, and its training is developed using stochastic expectation maximization. The MI-HMM provides a single statistical form to learn the parameters of an HMM in a multiple-instance learning framework without introducing any additional parameters. The efficacy of the model is shown both on synthetic data and on a real landmine data set. Experiments on both the synthetic data and the landmine data set show that an MI-HMM can achieve statistically significant performance gains when compared with the best existing HMM for the landmine detection problem, eliminate the ad hoc approaches in training set selection, and introduce a principled way to work with ambiguous time-series data.

Landmine detection with Multiple Instance Hidden Markov Models

A novel Multiple Instance Hidden Markov Model (MI-HMM) is introduced for classification of ambiguous time-series data, and its training is accomplished via Metropolis-Hastings sampling. Without introducing any additional parameters, the MI-HMM provides an elegant and simple way to learn the parameters of an HMM in a Multiple Instance Learning (MIL) framework. The efficacy of the model is shown on a real landmine dataset. Experiments on the landmine dataset show that MI-HMM learning is very effective, and outperforms the state-of-the-art models that are currently being used in the field for landmine detection.

Variational Mixture of Experts for Classification with Applications to Landmine Detection

In this paper, we (1) provide a complete framework for classification using Variational Mixture of Experts (VME); (2) derive the variational lower bound; and (3) apply the method to landmine, or simply mine, detection and compare the results to the Mixtures of Experts trained with Expectation Maximization (EMME). VME has previously been used for regression and Waterhouse explained how to apply VME to classification (which we will call as VMEC). However, the steps to train the model were not made clear since the equations were applicable to vector valued parameters as opposed to matrices for each expert. Also, a variational lower bound was not provided. The variational lower bound provides an excellent stopping criterion that resists over-training. We demonstrate the efficacy of the method on real-world mine classification; in which, training robust mine classification algorithms is difficult because of the small number of samples per class. In our experiments VMEC consistently improved performance over EMME.

Sub-pixel target spectra estimation and detection using functions of multiple instances

The Functions of Multiple Instances (FUMI) method for learning target pattern and non-target patterns is introduced and extended. The FUMI method differs significantly from traditional supervised learning algorithms because only functions of target patterns are available. Moreover, these functions are likely to involve other non-target patterns. In this paper, data points which are convex combinations of a target prototype and several non-target prototypes are considered. The Convex-FUMI (C-FUMI) method learns the target and non-target patterns, the number of non-target patterns, and the weights (or proportions) of all the prototypes for each data point. For hyperspectral image analysis, the target and non-target prototypes estimated using C-FUMI are the end-members for the target material and non-target (background) materials. For this method, training data need only binary labels indicating whether a data point contains or does not contain some proportion of the target endmember; the specific target proportions for the training data are not needed. In this paper, the C-FUMI algorithm is extended to incorporate weights for training data such that target and non-target training data sets are balanced (resulting in the Weighted C-FUMI algorithm). After learning the target prototype using the binary-labeled training data, target detection is performed on test data. Results showing sub-pixel explosives detection and sub-pixel target detection on simulated data are presented.