Sharif Digital Repository

Spam filtering is one of the large scale applications of machine learning. Much research has been carried out in the machine learning field with regards to spam filtering. Most of this work falls in the areas of batch learning or offline incremental learning. In batch learning, the learning process is carried out once on all the learning data. In applications such as spam filtering, in which the learning data is large in comparison to memory resources and data is generated in a stream, using incremental learning is required, in which the learning phase is repeated periodically. In each learning iteration of an offline incremental learning algorithm, a new set of data is learnt by the... 

Most of current state-of-the-art methods for object tracking use adaptive tracking-by-detection. The performance of state-of-the-art methods is almost real-time with acceptable accuracy. These methods use tracking-by-detection because of its robustness. Tracking-bydetection methods use a detector as a tracker and sweep input for object of interest. They use their predictions to adapt their parameters and therefore be adaptive to appearance change in target. While suitable for cases when the object does not disappear from the scene, these methods tend to fail on occlusions. In this work, we build on a novel approach called Tracking-Learning-Detection (TLD) that overcomes this problem. In... 

As part of the definition of concept drift as an online learning task, concepts change or drift as time goes by. Consequently, these changes have to be monitored and their implication for learning should be recognized. An example of concept drift detection is needed for spam filtering problem. An effective spam filter must be able to handle various changes, including changes in the user’s criteria for filtering spam, changes in message topics, and changes caused by the people sending spam messages. In this thesis, spam detection system has been considered in which emails are given sequentially and learns them one by one. As we mentioned, the purpose of this thesis is detecting spam emails.... 

Concept drift is a challenging problem in the context of data stream processing. As a result of increasing applications of data streams, including network intrusion detection, weather forecasting, and detection of unconventional behavior in financial transactions; numerous studies have been conducted in the field of concept drift detection. In order to solve the problem of concept drift detection, an ideal method should be able to quickly and correctly identify a variety of changes, adapt quickly to new concepts, in the presence of limitations of memory and processing power. In this thesis, a new explicit concept drift detection method based on ensemble classifiers has been proposed for data... 

The problem of learning using high volume of data as stream, has attracted much attention recently. In this thesis, the problem is modeled and analized using Online Convex Optimization tools [1], [2]. General performance bounds are stated and clarified in this framework [8]. Using the practical experience in Online Decision Making (e.g., predicting price in Stock Market), the need for a more flexible model, which adapts to changes in problem, is presented. In this thesis, after reviewing the literature and online convex optimization framework, we will define ”Concept Drift”, which describes changes in the dynamics of the problem and the statistical tools to detect it [13], [5]. And finally,... 

Image classification, i.e. the task of assigning an image to a class chosen from a predefined set of classes, has addressed in this thesis. At first the classifier is divided into two major sub partitions, feature extraction and classifier. Then we show that by using local feature extraction techniques such as BOW the classification accuracy will improve. In addition, using unlabeled data is argued as the fact to deal with high nonlinear structure of features. Recently, many SSL methods have been developed based on the manifold assumption in a batch mode. However, when data arrive sequentially and in large quantities, both computation and storage limitations become a bottleneck. So in large... 

Until now many domain adaptation methods have been proposed. A major limitation of almost all of these methods is their assumption that all test data belong to a single stationary target distribution and a large amount of unlabeled data is available for modeling this target distribution. In fact, in many real world applications, such as classifying scene image with gradually changing lighting and spam email identification, data arrives sequentially and the data distribution is continuously evolving. In this thesis, we tackle the problem of adaptation to a continuously evolving target domain that has been recently introduced and propose the Evolving Domain Adaptation (EDA) method to classify... 

One of the most important problems in online learning is a prediction with expert advice. In each step we make our prediction not only based on previous observation but also use expert information. In this thesis, we study the different well-known algorithms of expert advice and generalize problems when data arrival is in the two-dimensional grid. regret is a well-studied concept to evaluate online learning algorithm. online algorithm when data arrive consecutively in T time step has regret O (√(T)) . regret in two-dimensional grid with T row and P column is O(T√(P)).
2010 MSC: 68Q32 ; 68T05 ; 90C27  

Distance Metric Learning algorithms have been widely used in Machine Learning methods recently. In these algorithms a distance function between objecs (data points) is learned based on their labels or similarity and dissimilarity constraints. Recent works have shown that a good precision is obtained in classification or clustering methods which use these functions. Since in the current systems many of data points do not exist at the beginning and are added to the training set as the algorithm is run, online methods are needed to update learned metric due to new data.
In this thesis, we proposed a new online distance metric learning method that has higher performance than existing... 

As application demands for online convex optimization accelerate, the need for design-ing new methods that simultaneously cover a large class of convex functions and im-pose the lowest possible regret is highly rising. Known online optimization methods usually perform well only in specific settings, e.g., specific parameters such as the diam-eter of decision space, Lipschitz constant, and strong convexity coefficient, where their performance depends highly on the geometry of the decision space and cost functions. However, in practice, the lack of such geometric information leads to confusion in using the appropriate algorithm. To address these issues, some adaptive methods have been proposed... 

One of the most important targets in bio-informatics is the analysis of biological networks. These networks are modeled by graphs. Comparing networks with mapping is a useful tool for analyzing. The mapping between the nodes of a network that preserves some topological and functional structures, is called network alignment. Network alignment has various applications in different fields; such as pattern recognition, social networks, biological networks, and etc. The alignment of the protein-protein interaction network is one of the substantial problems. There are many static algorithms for the alignment of PPI networks. Because of the developments of computer science in recent years,... 

Nowadays, the online markets can easily change and adjust price of the product to an optimal price to increase the profit from the sale of their products. Because of this pricing flexibility, there are many applications of online pricing in online markets and so on.We study the problem of online pricing and specifically feature-based online pricing as an online learning problem in which a seller receives highly differentiated products online and prices them with the goal of obtaining the highest possible profit.The seller does not initially know the values of the different features, but can learn the values of the features based on whether products were sold at the posted prices in the...