Sharif Digital Repository

Flameless combustion is among the newest combustion technologies that is used for reducing the pollutants in recent couple of decades. Within this combustion regime, diminution of super hot points throughout the combustion chamber along with decline in production of pollutant gases occur simultaneously. This kind of combustion often is created by applying two processes; first by preheating the fuel and/or oxydizer flows at the inlet to tempratures over the self-ignition level; and second with diluting the whole amount of fuel and oxidizer in the chamber. These two happen by using hot exhaust gas of the same chamber (via feedback) or of the previous section (in multi-section chambers).... 

Tracking is one of the old and still not thoroughly solved problems in machine vision. Its importance lies on its many applications. These applications vary from security surveillance to examining the motion pattern of atomic particles. There is not a tracker which has acceptable results in all situations, yet. A tracker faces many difficulties such as change in illumination and occlusion. In past, tracking was done by using filters or optical flows. By use of the advances in machine learning and pattern recognition, many models have been proposed to accomplish tracking by using these new learning methods. In this dissertation, we proposed a new tracking method which utilizes sparse... 

Flameless combustion in combustion technology is an innovative method with low nox production that few years provided it passes. In this paper the effects of various parameters on flameless combustion chamber with separate fuel and air jets, has been modeled numerically with the help of Fluent 6.3 software. Differences between traditional combustion and flameless combustion were determined and the flame structure has been studied in different scenarios. In numerical modeling, approach were used to modeling turbulence and eddy dissipation concept (EDC) approach use to modeling combustion and turbulence interaction effects. A global Two-step reaction mechanism for propane and a global... 

A social network represents a set of entities and their relationships. Telecommunication networks, online social networks, and paper citation networks are some examples of networks in real world. Nowadays, analysis of social networks is an interesting research area with important applications. Particularly, managers of the social networks and the decision makers often require intelligent decision support for futures study in these social systems. The demanded decision support systems make it possible to define the desired social problem and to analyze the ”what-if scenarios.“ Computer simulation is an appropriate approach toward such decision support systems. In this approach, the desired... 

In a variety of applications, we deal with event sequences which we require to model the time of those. Event sequences modelling is vital in a variety of applications such as electronic commerce, social networks, health information; For instance, in the context of social networks entrance time, the act of like or comment can be regarded as an event sequence. Point processes are the framework for modelling event sequences, in which designer use prior knowledge and different assumptions (which is not necessarily true) to set the functional form of intensity function. That functional form may not be sufficient enough to model event sequences. In this project, we have used a deep nonlinear... 

Deep learning methods have become more popular in the past years. These methods use complex network architectures to model rich, hierarchical datasets. Although most of the research has been centered around Discriminative models, however, recently a lot of research is focused on Deep Generative Models. Two of the pioneering models in this field are Generative Adversarial Networks and Variational Auto-Encoders. In addition, knowing the structure of data helps models to search in a narrower hypothesis space. Most of the structure in datasets are models using Probabilistic Graphical Models. Using this structural information, one can achieve better parameter estimations. In the case of... 

Trip demand analysis is among the most important parts in comprehensive transportation studies. Developing four stage models including trip generation, trip distribution, mode choice, and assignment are so common in demand analyzing. In this study, trip generation models have been studied and to analyze data, two different methods including OLR and GWR are utilized. This study aims to examine relationship between aggregate zones in order to forecast future trips. Moreover, results of modeling trip generation with two different methods, OLR and GWR, are compared. Comparing different levels of aggregation reveals that as the number of zones decreases, the results get better. Decreasing number... 

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... 

One of the most crucial aspects of financial risk management is risk measurement. Advanced AI-based solutions can provide the proper tools for assessing global markets, given the complexity of the global economy and the violation of typical modeling assumptions. A new strategy for quantifying stock portfolio risk based on one of the machine learning models known as Variational Autoencoders is provided in this dissertation. The suggested method is a generative model that can learn the stocks' dependency structure without relying on assumptions about stock return covariance and produce various market scenarios using cross-sectional stock return data with a higher signal-to-noise ratio. We... 

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... 

Drug-target binding affinity prediction is one of the most important and vital part of drug discovery. The computational methods to predict binfing affinity is a standing challenge in drug discovery. State-of-the-art models are usually based on supervised machine learning with known label information. It is expensive and time-consuming to collect labeled data. This thesis proposes a semi-supervised model based on convolutional GAN (Generative adversarial networks). The model consists of two Gans and Two CNN blocks for feature extraction and fully connected layers for prediction. Gan can learn protein and drug features from unlabeled data. We evaluate the performance of our method using four... 

In this study, we aim to learn representations from microscopic cell images that effectively capture the features of drugs affecting the cells, allowing us to identify effective drugs for treating a disease. We employ two parallel learning paths using predictive and generative models. Specifically, we have achieved a predictive model on the RxRx19a dataset that, unlike previous models, is interpretable, optimized, and robust to dras- tic changes in drug properties. Additionally, we have developed the first generative model on this dataset, which not only generates high-quality images but also discovers a meaningful latent space. This latent space divides the representation into relevant... 

Aautomatic image annotation that assigns some labels to input images and provides a textual description for the contents of images has become an active field in machine vision community. To design an annotation system, we need a dataset that contains images and labels for them. However, a large amount of manual efforts is required to annotate all images in a dataset. To reduce the demand of annotation systems on the labeled images, one solution is to exploit useful information embedded into the unlabeled images and incorporate them into learning process. In machine learning community, semi-supervised learning (SSL) has been introduced with the aim of incorporating unlabeled samples into the... 

By the improvement of machine learning methods specially the Deep Learning in the last decade, there were expanding usage of these methods in Language Modeling task. As the essence of a language model is more basic, recently huge networks are trained with language model objective but fine-tuned on target tasks such as Question Answering, Sentiment Analysis and etc. which is a promising sign of its importance and usage in even other NLP tasks. However, this task still has severe problems. The Teacher Forcing based methods, suffer from the so-called exposure bias problem which is due to the train/test procedure discrepancy. Some solutions such as using Reinforcement Learning which has high... 

Representation learning refers to mapping the input data to another space, usually with lower dimensions than the input space. This task can be helpful in improving the performance of methods in downstream tasks, compression, and improving sample generation in generative models. Representation learning is a problem connected to information theory, and information theory's concepts and quantities are used widely in representation learning models. Besides, the representation learning problem is closely related to latent variable generative models. These models usually learn useful representations in their process of training, implicitly or explicitly. So, the usage of latent variable... 

Recently, a new generation of machine learning tasks, namely data generation, was born by emerging deep networks and modern methods for training neural networks on one hand, and the growth of available training data for training these networks on the other hand. Although distribution estimation and sampling were well-known problems in the science of statics, deep generative models can properly generate samples from real world distributions that common statistical methods fail in them e.g., image and music generation.Due to these improvements in deep generative models, researchers have recently tried to propose deep generative models for datasets with complex structures. These structured... 

In recent years, deep generative models have achieved incredible successes in various fields, including graph generation. Due to the advances made in graph generation by deep generative models, these methods have shown numerous applications from drug discovery and molecular graph generation to modeling social and citation network graphs. Graph generation is an approach to discovering and exploring new graph structures and has been attracting growing attention. One of the most challenging applications of deep graph generative models is molecular graph generation since it requires not only generating chemically valid molecular structures but also optimizing their chemical properties in the... 

Nowadays, there exist many real networks with distinctive features in comparison with random networks. Social networks, collaboration networks, citation networks, protein networks and communication networks are some example of complex network classes. Nowadays these networks are widespread and have many applications and the study of complex networks is an important research area. In many applications, the “synthetic networks generation” is one of the first levels of complex networks analysis. This level has many applications such as simulation and extrapolation. Many generative models are proposed for complex network modeling in recent years. By the use of these models, synthetic networks... 

In this thesis Generative Models in Deep Learning are discussed, especially Generative Models which are based on latent variables. Deep Generative Models have key role in developing Artificial Intelligence, particularly in developments of general cognition and perception in AI. In this thesis, this role for Generative Models and their applications in cognition development, and also the mathematical foundation of generative models are discussed