Sharif Digital Repository

Exponential growth of traffic and bandwidth demands in current data center networks, requires low-latency high-throughput interconnection networks, considering power consumption. Furthermore, increasing multicast intensive applications, alongside conventional unicast applications, arises power efficient communication in today’s data center networks as the main design challenge. Addressing these demands, optical networks suggest several benefits as well as circumventing most disadvantages of electrical networks. In this thesis, we propose an all-optical scalable architecture, for communicating intra-data centers. This architecture utilizes passive optical devices and enables optical circuit... 

Data center performance is a function of three features; bandwidth, latency, and reliability. The issue of increasing bandwidth and reducing power consumption for scalability is one of the things that can be accessed with optical technology and it has been used in recent designs. However, the problem of fault tolerance and dealing with failures in optical interconnection networks have been raised less. Data center failures fall into two categories of links and switches. The design of these networks is mainly done at two levels: the hardware design and the design of routing methods. Also, some parts may be added to control and establish connections between the routing and network hardware... 

In this research, an optical processing method for DNA classification is presented in order to overcome the problems in the previous methods. With improving in the operational capacity of the sequencing process, which has increased the number of genomes, comparing sequences with a complete database of genomes is a serious challenge to computational methods. Most current classification programs suffer from either slow classification speeds, large memory requirements, or both. To achieve high speed and accuracy at the same time, we suggest using optical processing methods. The performance of electronic processing-based computing, especially in the case of large data processing, is usually... 

It is well known that data centres consume high amounts of energy, which has become a major concern in the field of cloud computing. Therefore, energy consumption could be reduced by using intelligent mechanisms work to adapt the set of network components to the total traffic volume. SDN is an efficient way to do so because it has many benefits over traditional approaches, such as centralised management, low capex, flexibility, scalability and virtualisation of network functions. In our work will we use the heuristic energy-aware routing (HEAR) model, which is composed of the proposed heuristic algorithm and the energy-aware routing algorithm. This work identifies the unused links and... 

Abstract: In this thesis, we tackled the problem of nonlinear activation function in optical artificial neural networks, and in particular in convolutional and recurrent neural networks. In the end, we propose an all-optical recurrent neural network in free-space optics for the first time. Artificial neural networks are a branch of artificial intelligence, which can be adopted to solve a wide variety of problems. While very powerful, these networks can be very power hungry and slow when it comes to solving very complicated problems. Optical versions of these networks bring the promise of solving both of these issues and provide a fast and power efficient platform for these networks. However,... 

Protein, as the functional unit of the cell, plays a vital role in its biological function. With the advent of advanced sequencing techniques in recent years and the consequent exponential growth of the number of protein sequences extracted from diverse biological samples, their analysis, comparison, and classification have faced a considerable challenge. Existing methods for comparing proteins divide into two categories: methods based on alignment and alignment-free. Although alignment-based methods are among the most accurate methods, they face inherent limitations such as poor analysis of protein groups with low sequence similarity, time complexity, computational complexity, and memory... 

During the recent years, the emerging technology of photonic, on-chip communication has been widely explored to resolve the problems of traditional electrical communication in terms of performance and power consumption. Micro-Resonator (MR) as the photonic building block for optical comunication is very sensitive to temperature and process variation. Variation of resonance wavelength of a micro-resonator as a result of temperature fluctuations causes tremendous error for optical data routing through the waveguides, and reduce the reliability of network. A temperature-aware routing algorithm enables the network to tolerate fault and continue working properly. Temperature variation of MRs is a... 

Global on chip communication becomes a critical power bottleneck in high performance many core architectures. The importance of power dissipation in networks-on-chip along with power reduction capability of on-chip nanophotonic interconnects has made optical network on chip a novel technology. Major advantages like high bandwidth, light speed latency and low power consumption, provide a promising solution for future of communications in many core architectures. However, the basic elements that are embedded in optical networks on chip are extremely temperature sensitive. This would lead to change in the physical characteristics of nanophotonic elements which may cause failure in network on... 

The classification of biological sequences is an open issue for a variety of data sets, such as viral and metagenomics sequences. Therefore, many studies utilize neural network tools, as the well-known methods in this field, and focus on designing customized network structures. However, a few works focus on more effective factors, such as input encoding method or implementation technology, to address accuracy and efficiency issues in this area. Therefore, in this work, we propose an image-based encoding method, called as WalkIm, whose adoption, even in a simple neural network, provides competitive accuracy and superior efficiency, compared to the existing classification methods (e.g. VGDC,... 

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... 

Due to the increasing volume of data in various fields, existing electronic processors face a major challenge. While processing power has increased, solving complex problems in a timely manner remains a major challenge for today's processors. Neuromorphic engineering offers a potential solution by looking to processors found in nature, such as the human brain. This field of research involves investigating natural processors and designing new ones based on these models. To address issues related to manufacturing and integrating transistors, increasing processor costs, and the limitations of Moore's law, it is possible to use analog signals, such as sound or light, instead of electrical... 

Motif finding problem in biology is a search for repeated patterns to reveal information about gene expression, one of the most complex subsystems in genomics. ChIP-seq technology abled researchers to investigate location of protein-DNA interactions but analyzing downstream results of such experiments to find actual regulatory signals in genome is challenging. For many years, applications of motif finding had models based on limiting assumption as an exchange for lower computational complexity. Results: AKAGI program is build upon upgraded methods and new general models to investigate statistical and experimental evidences for accurately finding significant patterns among biological... 

Convolutional neural networks (CNNs) are at the heart of several machine learning applications, while they suffer from computational complexity due to their large number of parameters and operations. Recently, all-optical implementation of the CNNs has achieved many attentions, however, the recently proposed optical architectures for CNNs cannot fully utilize the tremendous capabilities of optical processing, due to the required electro-optical conversions in-between successive layers. Therefore, in our first study, we proposed OP-AlexNet which has five convolutional layers and three fully connected layers. Array of 4f optical correlators is considered as the optical convolutional layer,... 

The Major Histocompatibility Complex (MHC) binds to the derived peptides from pathogens to present them to killer T cells on the cell surface. Developing computational methods for accurate, fast, and explainable peptide-MHC binding prediction can facilitate immunotherapies and vaccine development. Various deep learning-based methods rely on feature extraction from the peptide and MHC sequences separately and ignore their valuable binding information. This paper develops a capsule neural network-based method to efficiently capture and model the peptide-MHC complex features to predict the peptide- MHC class I binding. Various evaluations over multiple datasets using popular performance metrics... 

As sequencing technologies have advanced in the field of single cells, it has become possible to investigate complex and rare cell populations and discover regulatory relationships between genes. The detection of rare cells has been greatly facilitated by this technology. However, due to the large volume of data and the complex and uncertain distribution of data, as well as the high rate of technical zeros, the analysis of single cell data clusters remains a computational and statistical challenge. Dimensionality reduction is a significant component of big data analysis. Machine learning methods provide the possibility of better analysis by reducing the non-linear dimensions of data. A graph... 

Biological sequences, such as DNA, RNA, and proteins, contain vital information and activities of living organisms. Any changes or mutations in these sequences can alter the structure of proteins, which are the basis of all cell activities. This can cause disruptions in cellular functions. With advancements in cell data extraction technology, the amount of genetic data has increased, leading to a need for efficient methods to identify and study mutations. One of the primary tools for this work is biological sequence comparators. Despite the different types of sequence comparators available, there is still a need for an algorithm that is fast, accurate, scalable, and uses few resources.... 

Alignment of genetic sequences is a fundamental part of genetic and bio-science. Alignment of DNA and protein sequences has an effective role in accelerating and simplifying problems in Bioinformatics like predicting protein interactions. Smith-Waterman algorithm is a precise algorithm for performing local alignment, suffering from computation complexity. There are some implementations on CPU, GPU, and FPGA platforms in order to reduce the run time of this algorithm. FPGA implementation is considered because of low power consumption and high degree of parallelism. With using pipeline and hardware redundancy techniques, various architectures have been proposed and implemented. In the best... 

The ever-expanding growth of internet traffic enforces deployment of massive Data Center Networks (DCNs) supporting high performance communications. Optical switching is being studied as a promising approach to fulfill the surging requirements of large scale data centers. Data center networks include hundreds of thousands of nodes that require flexible, high bandwidth, and low power infrastructures for their communications. Optical interconnection networks supply the required bandwidth, consuming much lower power compared to their electrical counterparts. The heterogeneous nature of data centers’ traffic needs a flexible network architecture which can be dynamically configured according to... 

Integrating hundreds of cores on a single chip necessitates high performance interconnection network. Due to great delay and power consumption of electrical networks in global connections, their use might be limited in future. Optical networks-on-chip are introduced recently as a proper alternative for electrical networks by providing high bandwidth, low latency, and low power consumption. Despite these outstanding advantages, optical component behavior is very sensitive to temperature fluctuations. Specifically, thermal variations affect refractive index of semiconductor, which in turn change resonating wavelength of micro-ring resonator. Therefore, the switch cannot operate properly and... 

In this research, two novel optical methods have been proposed for DNA local sequence alignment. The proposed methods benefit from algorithms and methods in computer field and ability of parallelism in optical wave to achieve a low-cost process and propose an easy understanding output in DNA local sequence alignment procedure. The first method is built upon moiré matching technique which is extended by proposed HAPPOC scheme using amplitude, phase, and polarization of optical wave. For analyzing the extended moiré output, a novel 3D Artificial Neural Network is designed and developed by optical structure. The second structure, as named HAWPOD method, is based on DV-Curve method. The HAWPOD...