Sharif Digital Repository

Transfer Learning is one of the most important fields in the Machine Learning area. Respect to the advances that we have seen in the Computer Science, especially in the Machine Learning area, we need a tool that can transfer learnings from different domains to each other. As data distribution varies, many statistical models require restructuring using new training data. In many applications, re-assembling training data and re-structuring models is inefficient and costly, so reducing the need for this practice seems appropriate. In these cases, knowledge transfer or learning transfer between domains may be desirable. For example, in the area of the B rain-Computer Interface, when it... 

Reinforcement learning is a successful model of learning that has received a lot of attention in recent years and has had significant achievements. However, methods based on reinforcement require a lot of data. Therefore, it is important to find ideas to keep learning at a high level despite the lack of data. Many of these ideas are known as statistical regularity. In this thesis, we study methods to enhance the learning rate, including methods for sharing neural network weights between value function and policy networks. In this thesis we will try to gain a more general understanding of the regularization in reinforcement learning and increase the learning rate by implementing these methods... 

In recent years, techniques in articial intelligence have become an important tool in the analysis of physiological signals. While the application of machine learning techniques has proved useful in other elds, researchers have had difficulty proving its utility for the analysis of physiological signals. A major challenge in applying such techniques to the analysis of physiological signals is dealing effectively with inter-patient differences. The morphology and interpretation of physiological signals can vary dep ending on the patient. This poses a problem, since statistical learning techniques aim to estimate the underlying system that produced the data. If the system (or patient) changes... 

The importance of transfer learning in image-related problems comes from its many advantages that are sometimes undeniable. Previous researches have well shown the success of transfer learning in this area using deep neural networks. However, transfer learning for time series data has not yet been done in a conventional and automated manner. The main reason for avoiding transfer learning in this domain relates to the dynamic and stochastic nature of the time series, where they show a time-varying behavior. Previous experiments have shown that transfer learning between two heterogeneous time series could harm the forecasting accuracy of a model. Therefore, in this thesis, we aim to explore... 

This paper develops a novel insight and procedure that includes a variety of algorithms for finding the best solution in a structured multi-agent system with internal communications and a global purpose. In other words, it finds the optimal communication structure among agents and the optimal policy in this structure. First, a unique reinforcement learning algorithm is proposed to find the optimal policy of each agent in a fixed structure with non-linear function approximation like artificial neural networks (ANN) and eligibility traces. Secondly, a mechanism is presented to perform self-organization based on the information of the learned policy. Finally, an algorithm that can discover an... 

Population genetics studies the distribution and changes in allele frequencies under the influence of five main evolutionary processes: natural selection, genetic drift, mutation, gene flow, and recombination. Among these, the recombination process can influence a wide range of biological processes by rearranging genes, repairing DNA structure, and participating actively in cell division mechanisms. Recombination has the ability to create genetic diversity through gene rearrangement, which is the main reason for creating diversity and evolution in organisms. Models such as Hill-Robertson have proven the influential role of recombination in accelerating evolutionary mechanisms. Also,... 

Named entity recognition (NER) is one of important tasks in natural language processing (NLP). Named entities consist of specific nouns such as personal names, organizations, locations, etc., which refer to important entities in text. NER contributes towards other NLP tasks such as machine translation, text summarization ,and text classification. In the recent decade, with respect to development of deep learning (DL) methods, considerable progress has been made in this field. The objective here is to propose an efficient method for NER in Farsi (Persian) text through DL methods. Since deep neural networks require a great deal of training data, and due to the fact that Farsi lacks such data,... 

In some of object recognition problems, labeled data may not be available for all categories. Zero-shot learning utilizes auxiliary information (also called signatures) describing each category in order to find a classifier that can recognize samples from categories with no labeled instance. On the other hand, with recent advances made by deep neural networks in computer vision, a rich representation can be obtained from images that discriminates different categorizes and therefore obtaining a unsupervised information from images is made possible. However, in the previous works, little attention has been paid to using such unsupervised information for the task of zero-shot learning. In this... 

In this research, we study one of the challenges in the field of software engineering, namely the detection of diplicate questions in Stackoverflow, the Q&A community of programmers. The works done in this area has problems such as complexity and reduced performance over time. The proposed solution is based on machine learning and modern representation learning methods. Representation is done with two approaches, domain specific learning and transfer learning. Fasttext and GloVe, the two word embeddings used in domain specific learning, and in transfer learning, the embedding of the universal sentence encoder has been used. Support vector machine and multilayer perceptron used as... 

The use of named entity recognition systems as preprocessing is used in many natural language analysis issues. With the advent of deep learning, the methods of this area were also affected. Today, there is considerable progress in this area due to the development of data resources for English, Chinese, German, and Spanish. They are also good trained models in formal Persian. However, for informal Persian, which contains a large portion of the web content under the Web, the current models do not produce a suitable solution. In this study, we use the same approach to train our models due to achieving state-of-the-art results in pre-trained models. On the other hand, there is a lack of standard... 

Named Entity Recognition (NER) is a key component and the first step of many natural language processing tasks such as question answering systems, information retrieval, machine translation, text summarization, and so on. First NER system initially used rule-based and machine learning methods, which grew significantly with the advent of deep learning architectures as well as the development of hardware and data resources. Traditional deep learning methods used convolutional and recursive neural networks that had disadvantages such as gradient vanishing and non-parallel computing, respectively. In addition, the need for huge corpus and powerful hardware resources was one of the problems of... 

In recent years, Additive Manufacturing has been extensively used by various industries. These manufacturing processes produce components in a layer-by-layer manner; therefore, they do not impose any geometric constrains to engineers and provide designers with the freedom to design components. Nowadays, one of the primary goals of all industries is to utilize as few raw materials as possible; this way they can deal with the shortage of raw materials and improve their efficiency. Consequently, they implement topology optimization algorithms to design and produce their components. However, topology optimization algorithms result in complicated geometries that can only be fabricated by AM.... 

Following dramatic changes after using deep learning method as a solution for Natural Language Processing tasks, Transformer architecture get popular. Based on that, then BERT Language model presented and get state-of-the-art as a solution for a lot of language processing tasks. It was a turning point in Natural Language Processing field. Also, in cross-lingual methods research line motivated by developing a common space for representation of language units, e.g. words, sentences, in more that one language, get some remarkable improvements. However, for languages distant from English such as Persian or Arabic the methods' performance was not clear. In this work, we performed some innovative... 

There are currently about 50 million people with Alzheimer's disease in the world, and this number is about 700 thousand in Iran. The symptoms of the disease include decreased awareness, disinterest in unfamiliar subjects, increased distraction, speech problems, and etc. which gradually leads to an absolute inability to perform daily activities and completely mute. The disease belongs to the category of neurological disorders and is the most common type of dementia for which no treatment has been offered so far. However, if the disease is diagnosed in its early stage, a series of pharmacological and behavioral therapy approaches can be prescribed to reduce the pace or progression of the...