Sharif Digital Repository

Multiple object tracking refers to the detection and following of target object classes in video sequences. In this task, all objects belonging to the target classes in the video are detected simultaneously in each frame, and a unique ID is assigned to each of them throughout the video. In recent years, the use of graph neural networks for solving this problem has received significant attention because these models are suitable tools for discovering and improving the relationships between objects in the scene, which can greatly assist in better object pairing. However, there are various challenges to using graph neural networks, the most important of which is the limitation of input graph... 

Forecasting and analysing financial markets has always been an interesting research topic for fields ranging from financial sciences to mathematics and statistics. With the rapid development of artificial intelligence in the recent years, there has been a growing interest in using deep neural networks to predict market future trends. The price in these markets is determined by mechanisms of demand and supply. When there is a tendancy to buy a stock, there will be an increase in demand resulting a positive growth for price. On the other hand, when a large group of investors decide to sell their assets, market will experience an increase in supply and subsequently the prices drop. Availability... 

There has been a noticable surge in the usage of machine learning techniques in various fields, such as security related fields. With this growing pace of using machine learning to solve various problems, securing these models against attackers has become one of the main topics of machine learning literatures. Recent work has shown that in an adversarial environment, machine learning models are vulernable, and attackers can create carefully crafted inputs to fool the models. With the advent of deep neural networks, many researchers have used deep neural networks for the task of malware detection, and they have achieved impresive results. Finding the vulnerlabilities of these models is an... 

We study the problem of experiment design to learn causal structures from interventional data. We consider an active learning setting in which the experimenter decides to intervene on one of the variables in the system in each step and uses the results of the intervention to recover further causal relationships among the variables. The goal is to fully identify the causal structures with minimum number of interventions. We present the first deep reinforcement learning based solution for the problem of experiment design. In the proposed method, we embed input graphs to vectors using a graph neural network and feed them to another neural network which outputs a variable for performing... 

Analysis of people appearance variation in multi-camera networks for person re-identification or person retrieval is a very challenging problem due to the many intra-class variations between different cameras. Like any problem in the field of machine vision, it is generally divided into two parts. The first part is feature extraction and the second part is feature matching for person retrieval. So far, various methods have been proposed for the extraction of discriminative features, which are generally divided into three categories: stripe-based, patch-based, and body-based methods. However, methods based on stripes, although simpler, have performed better due to their greater compatibility... 

Diverse personal names are mentioned in everyday news but news agencies do not separate entities with same or equal names. This could make irrelevant news appear while searching an ambiguous name. Personal Name Disambiguation in news seeks to partition a significant amount of news to distinct classes each of which belongs to a single entity in the real world. In this thesis, which up to the researcher is the first of its kind at least in Persian, researcher gained opportunity of using FarsiYar News Dataset and to be specific 50,000 of news in FarsNews dataset which were published in the year 1397. First of all, a database was built using these news data and then the nonstructured news were... 

Regarding the ever-increasing usage of text generation and analysis in Natural Language Processing field, Language Modeling and Masked Language Modeling have been recently one of the most frequent tasks. Besides, many pretrained models such as BERT have been proposed due to the lack of rich datasets and computational resources among researchers. These models can be finetuned on other datasets in downstream tasks. Although these Transformer-based deep neural networks have performed perfectly in many problems, they still have some shortcomings in a few tasks.Furthermore, structured data like graphs have been recently used extensively in Natural Language Processing and researchers have taken... 

Due to the emergence of social networks in recent years and people’s usage of them for expressing their thoughts and emotions, there are lots of user data in these networks. The development of social networks has created a good opportunity for organizations and people to extract user profiles from social networks. Hence, user profiling has become an interesting problem for researchers. Predicting users’ occupational class is one of the main problems in this field. Most of the existing related works use only textual features of users, whereas users’ relations graph can give useful information about users. In this research, we propose a model based on Graph Neural Networks (GNNs) to predict... 

With the advent of graph neural network networks, a new class of models has emerged that has a high ability to learn powerful representations. Also, there have been popular probabilistic latent variable models for representation learning and solving graph problems. Graph neural networks do not necessarily provide meaningful representation in their hidden layers and also do not have the ability to estimate uncertainty. Learning probabilistic models is usually a slow process and there is no specific way to add general features to these models. Therefore, recently, a combination of neural network models and probabilistic network models have been developed that can partially answer these... 

Graphs are a language to describe and analyze connections and relations. Recent developments have increased graphs' applications in real-world problems such as social networks, researchers' collaborations, and chemical compounds. Now that we can extract graphs from real life, how can we model and generate graphs similar to a set of known graphs or that are very likely to exist but haven't been discovered yet? Therefore, this research will focus on the problem of graph generation. In graph generation, a set of graphs is a training dataset, and the goal of the thesis is to present an improved deep generative model to learn the training data's distribution, structure, and features.Identifying... 

EEG brain signals have gained particular attention among researchers in the field of brain signal processing due to their easy and cheap recording, high temporal resolution, and non-invasiveness. On the other hand, defects such as high vulnerability to various types of noise and artifacts have caused the main challenge before processing them to improve the signal-to-noise ratio and the interpretability of brain connectivity obtained from them. In order to solve these challenges, two important problems of "separation of desired and undesired signal subspace" and "functional and effective connectivity analysis" have been raised, respectively. In solving both problems, EEG signals are usually... 

The primary purpose of the present research is development and implementation of advanced state estimation and control techniques for space rendezvous and docking. To achieve this aim, the present study has first investigated the use of graph neural networks (GNNs) to filter out the noise of the sensor data in the state estimation process. The measurement package is consisted of a gyroscope, star trackers, and a GPS sensor providing inputs to the GNNs. The obtained results showed that the use of GNNs significantly improves the accuracy of the state estimation compared to traditional methods. In addition, the study has focused on developing advanced control techniques for spacecraft position... 

Stock market prediction has always been a challenging task. Due to its stochastic nature, naive models cannot help solve the problem. In the past, Statistical models were used, however nowa- days with the rise of deep learning and more complex models, aggregating data, in order to pre- dict the stock price, has become feasible. Moreover, the emergence of social networks enables researchers to design models for stock prediction.Researchers used recurrent networks and word vector representations to solve this problem. However, recently newer models such as generative models based on VAEs and attention have gained interest. Newer models also don’t rely on a single data source and use multiple... 

Around world and the networks within it can be modeled in various templates. Graph structure is one of those templates in which objects and relations may have more than one types. We call this phenomenon "heterogeneity".Heterogeneity makes the networks hard to model and that is why the proposed methods for modeling the networks assumed the network structures homogeneous. This assumption may cause data loss due to ignoring the variety of types in network objects and relations and this loss can lessen the accuracy of data mining tasks.To tackle the challenge of data loss in the mentioned assumption, learning representations for heterogeneous information networks (HINs) was introduced. HINs... 

A drug can be defined as a substance made to prevent disease, cure a specific symptom, relieve pain, and reduce anomalies in the body. The process of drug designing is so laborious, complex, costly, and time-consuming that chance of failure during the lab experiment stages is high. These challenges have persuaded researchers to find new usage for existing drugs, referred to as drug repurposing, with the main advantage of reducing cost, risk, and time. To this aim, computational methods have been applied to discover hidden pharmaceutical capabilities of drugs in terms of predicting whether a particular drug can interact with a particular protein.Graph Neural Networks (GNNs) have recently... 

In network science, link prediction is one of the essential tasks that has been neglected. One important application of link prediction in telecommunication networks is analyzing the user's consumption pattern to provide better service. This project aims to predict future links with applications to call data using the users' call history. In previous research, there are two main approaches: 1) heuristic-based approach, and 2) deep-learning-based approach, such as graph neural networks. These methods are mainly used for processing static graphs, and therefore, we cannot generalize them to dynamic graphs. But there are many graphs which are dynamic in nature. For instance, call data records... 

Graph signals are sets of values residing on sets of nodes that are connected via edges. Graph Neural Networks (GNNs) are a type of machine learning model for working with graph-structured data, such as graph signals. GNNs have applications in graph classification, node classification, and link prediction. They can be thought of as learnable filters. In this thesis, our focus is on graph filters and enhancing the performance of GNNs. In the first part, we aim to reduce computational costs in graph signal processing, particularly in graph filters. We explore methods to transform signals to the frequency domain with lower computational cost. In the latter part, we examine regulations in... 

Recognizing people mentioned in the text of narrations, is of the main importance in validating narrations in narration analysis. The problem naturally reduces to two sub-problems: Firstly, named entities related to narrators in the text should be determined. Secondly, if they are ambigous, the purpose of each should be clearified. In this resaerch, it is tried to apply pre-trained language models in soultion of the first sub-problem. For the purpose of entity linking with entities in a KB, graph-based model is suggested