---

GMM is one of the most successful models in the field of automatic language identification. In this paper we have proposed a new model named adapted weight GMM (AW-GMM). This model is similar to GMM but the weights are determined using GMM-VSM LID system based on the power of each component in discriminating one language from the others. Also considering the computational complexity of GMM-VSM, we have proposed a technique for constructing bigram sequences of components which could be used for higher sequence orders and decreases the complexity. Experiments on four languages of OGI corpus including English, Farsi, French and German have shown the effectiveness of proposed techniques.

---

Hyper spectral remote sensing imagery, due to its rich source of spectral information provides an efficient tool for ground classifications in complex geographical areas with similar classes. Referring to robustness of Support Vector Machines (SVMs) in high dimensional space, they are efficient tool for classification of hyper spectral imagery. However, there are two optimization issues which strongly effect on the SVMs performance: Optimum SVMs parameters determination and optimum feature subset selection. Traditional optimization algorithms are appropriate in limited search space but they usually trap in local optimum in high dimensional space, therefore it is inevitable to apply meta-heuristic optimization algorithms such as Genetic Algorithm to obtain global optimum solution. This paper evaluates the potential of different proposed optimization scenarios in determining of SVMs parameters and feature subset selection based on Genetic Algorithm (GA). Obtained results on AVIRIS Hyper spectral imagery demonstrate superior performance of SVMs achieved by simultaneously optimization of SVMs parameters and input feature subset. In Gaussian and Polynomial kernels, the classification accuracy improves by about 5% and15% respectively and more than 90 redundant bands are eliminated. For comparison, the evaluation is also performed by applying it to Simulated Annealing (SA) that shows a better performance of Genetic Algorithm especially in complex search space where parameter determination and feature selection are solve simultaneously.

---

One of the machine learning tasks is supervised learning. In supervised learning we infer a function from labeled training data. The goal of supervised learning algorithms is learning a good hypothesis that minimizes the sum of the errors. A wide range of supervised algorithms is available such as decision tress, SVM, and KNN methods. In this paper we focus on decision tree algorithms. When we use the decision tree algorithms, the data is partitioned by axis- aligned hyper planes. The geometric concept of decision tree algorithms is relative to separability problems in computational geometry. One of the famous problems in separability concept is computing the maximum bichromatic discrepancy problem. There exists an -time algorithm to compute the maximum bichromatic discrepancy in d dimensions. This problem is closely relative to decision trees in machine learning. We implement this problem in 1, 2, 3 and d dimension. Also, we implement the C4.5 algorithm. The experiments showed that results of this algorithm and C4.5 algorithm are comparable.

---

Alzheimer’s disease (AD) is a progressive and irreversible disease which gradually makes patients unable to do their daily routines. Although the present treatments can not cure the disease completely, its early detection can reduce symptoms and enhance the patients’ life quality. In the current literature, using the grey matter (GM) tissue which is known as an appropriate biomarker is highly common in AD diagnosis. However, two other brain tissues known as cerebrospinal fluid (CSF) and white matter (WM) seem to reveal beneficial information about the patients’ brain changes. The aim of the present study is to develop an automatic system for the early diagnosis of Alzheimer’s disease from structural MRI by simultaneously considering suitable features of all GM, CSF and WM tissues. A SVM-RBF classifier is trained and evaluated on the OASIS database to separate AD from healthy control (HC) subjects. The obtained results represent higher accuracy and sensitivity of the proposed algorithm in comparison with similar method.