Using the EM Algorithm to Train Neural Networks: Misconceptions and a New Algorithm for Multiclass Classification

The expectation-maximization (EM) algorithm has been of considerable interest in recent years as the basis for various algorithms in application areas of neural networks such as pattern recognition. However, there exists some misconceptions concerning its application to neural networks. In this paper, we clarify these misconceptions and consider how the EM algorithm can be adopted to train multilayer perceptron (MLP) and mixture of experts (ME) networks in applications to multiclass classification. We identify some situations where the application of the EM algorithm to train MLP networks may be of limited value and discuss some ways of handling the difficulties. For ME networks, it is reported in the literature that networks trained by the EM algorithm using iteratively reweighted least squares (IRLS) algorithm in the inner loop of the M-step, often performed poorly in multiclass classification. However, we found that the convergence of the IRLS algorithm is stable and that the log likelihood is monotonic increasing when a learning rate smaller than one is adopted. Also, we propose the use of an expectation-conditional maximization (ECM) algorithm to train ME networks. Its performance is demonstrated to be superior to the IRLS algorithm on some simulated and real data sets.

Symptoms and diagnosis of borderline, early and advanced epithelial ovarian cancer

Objective. To document symptoms associated with borderline, early and advanced ovarian cancer and identify personal characteristics associated with early versus late diagnosis. Methods. Information concerning symptoms and diagnosis history was available from 811 women with ovarian cancer who took part in an Australian case–control study in the early 1990s. Women were classified into three groups for comparison based on their diagnosis: borderline, early (stage I–II) and advanced (stage III–IV) invasive cancer. Results. Sixteen percent of women with borderline tumors, 7% with early cancer and 4% with advanced cancer experienced no symptoms before diagnosis (P < 0.0001). Among women with symptoms, abdominal pain (44%) or swelling (39%) were most frequently reported; an abdominal mass (12%) and gynecological symptoms (12%) were less common. Compared to advanced stage cancer, women with early stage cancer were more likely to report an abdominal mass or urinary symptoms but less likely to report gastrointestinal problems or general malaise. General malaise and ‘other’ symptoms were least common in borderline disease. Older women, and those with higher parity or a family history of breast or ovarian cancer, were more likely to be diagnosed at an advanced stage of disease. Conclusions. Women who experience persistent or recurrent abdominal symptoms, particularly swelling and/or pain should be encouraged to seek medical attention and physicians should be alert to the possibility of ovarian cancer even in the absence of an abdominal mass. Further information about the prevalence of these symptoms in the general population is essential to assist physicians in patient management.

Use of the EM algorithm to detect QTL affecting multiple-traits in an across half-sib family analysis

QTL detection experiments in livestock species commonly use the half-sib design. Each male is mated to a number of females, each female producing a limited number of progeny. Analysis consists of attempting to detect associations between phenotype and genotype measured on the progeny. When family sizes are limiting experimenters may wish to incorporate as much information as possible into a single analysis. However, combining information across sires is problematic because of incomplete linkage disequilibrium between the markers and the QTL in the population. This study describes formulae for obtaining MLEs via the expectation maximization (EM) algorithm for use in a multiple-trait, multiple-family analysis. A model specifying a QTL with only two alleles, and a common within sire error variance is assumed. Compared to single-family analyses, power can be improved up to fourfold with multi-family analyses. The accuracy and precision of QTL location estimates are also substantially improved. With small family sizes, the multi-family, multi-trait analyses reduce substantially, but not totally remove, biases in QTL effect estimates. In situations where multiple QTL alleles are segregating the multi-family analysis will average out the effects of the different QTL alleles.

A gradient descent algorithm for minimizing amino acid coupling reactions when synthesizing cyclic-peptide libraries

Combinatorial chemistry has become an invaluable tool in medicinal chemistry for the identification of new drug leads. For example, libraries of predetermined sequences and head-to-tail cyclized peptides are routinely synthesized in our laboratory using the IRORI approach. Such libraries are used as molecular toolkits that enable the development of pharmacophores that define activity and specificity at receptor targets. These libraries can be quite large and difficult to handle, due to physical and chemical constraints imposed by their size. Therefore, smaller sub-libraries are often targeted for synthesis. The number of coupling reactions required can be greatly reduced if the peptides having common amino acids are grouped into the same sub-library (batching). This paper describes a schedule optimizer to minimize the number of coupling reactions by rotating and aligning sequences while simultaneously batching. The gradient descent method thereby reduces the number of coupling reactions required for synthesizing cyclic peptide libraries. We show that the algorithm results in a 75% reduction in the number of coupling reactions for a typical cyclic peptide library.

Enhanced detection-guided NLMS estimation of sparse FIR-modeled signal channels

In various signal-channel-estimation problems, the channel being estimated may be well approximated by a discrete finite impulse response (FIR) model with sparsely separated active or nonzero taps. A common approach to estimating such channels involves a discrete normalized least-mean-square (NLMS) adaptive FIR filter, every tap of which is adapted at each sample interval. Such an approach suffers from slow convergence rates and poor tracking when the required FIR filter is "long." Recently, NLMS-based algorithms have been proposed that employ least-squares-based structural detection techniques to exploit possible sparse channel structure and subsequently provide improved estimation performance. However, these algorithms perform poorly when there is a large dynamic range amongst the active taps. In this paper, we propose two modifications to the previous algorithms, which essentially remove this limitation. The modifications also significantly improve the applicability of the detection technique to structurally time varying channels. Importantly, for sparse channels, the computational cost of the newly proposed detection-guided NLMS estimator is only marginally greater than that of the standard NLMS estimator. Simulations demonstrate the favourable performance of the newly proposed algorithm. © 2006 IEEE.

Efficient spatial clustering algorithm using binary tree

In this paper we present an efficient k-Means clustering algorithm for two dimensional data. The proposed algorithm re-organizes dataset into a form of nested binary tree*. Data items are compared at each node with only two nearest means with respect to each dimension and assigned to the one that has the closer mean. The main intuition of our research is as follows: We build the nested binary tree. Then we scan the data in raster order by in-order traversal of the tree. Lastly we compare data item at each node to the only two nearest means to assign the value to the intendant cluster. In this way we are able to save the computational cost significantly by reducing the number of comparisons with means and also by the least use to Euclidian distance formula. Our results showed that our method can perform clustering operation much faster than the classical ones. © Springer-Verlag Berlin Heidelberg 2005