Projects

Further Interested Projects: (suitable for BS/MS/PhD Students, titles subject to revision based on the scope of the project for each level of students)

Research Project 1: Exploration of Structured Grid Computation on Many-Core Architectures

Recent advances in many-core architectures opened a new opportunity in harnessing their computing power for general purpose computing. However, efficient utilization of these architectures is possible with expert knowledge of the architecture and code optimizations. Several optimization techniques, tools and libraries have been proposed for these architectures focusing on diverse applications domains such as numerical linear algebra, linear iterative solvers, machine learning algorithms, image processing, fluid-flow dynamics and reservoir simulations. In this thesis, we will be focusing on accelerating computational science applications such as the class of structured grid computation (SGC). Most of the SGC simulation time is spent in repeatedly solving a large sparse linear system. Aggressive code optimizations are needed to achieve higher simulation accuracy by scaling up the application. Following are the key objectives of this work:

• Discuss the mismatch between current compiler techniques and the requirements for implementing efficient iterative linear solvers.
• Identify the basic and domain specific optimizations.
• Explore the approaches used by computational scientists to program SGCs such as manual optimizations, utilization of numerical libraries, and use of a combination of integrated libraries, user annotations to define regularity in data structure and algorithm properties and auto-tuning.

Identify the main optimization functionalities for an integrated library to ease the process of defining complex SGC data structure and optimizing solver code using intelligent, high-level interface, and domain specific annotations.

Research Project 2: Evaluation of Sparse Linear Algebra Operators on Many-Core Architectures

The abundant data parallelism available in many-core architectures is needed to improve accuracy in scientific and engineering simulation (SES). In many cases, most of the simulation time is spent in a linear solver involving sparse matrix-vector multiply (SpMV). In petroleum forward Oil and Gas reservoir simulation (FRS), the application of a stencil relationship to a structured grid leads to a family of generalized Heptadiagonal (GH) sparse solver matrix with some regularity and structural uniqueness. To solve the large scale systems of linear equation one may use direct and iterative approaches. Direct methods are more reliable, more accurate but require large memory space and generally lacks scalability in many-core due to their sequential nature. Iterative methods have abundant data parallelism with large sparse systems of linear equations. Among the most popular iterative methods to solve large sparse linear system is Krylov subspace solvers and BiCGStab technique is one example solver. The most frequent operators are the matrix vector multiplication, vector inner product, and vector addition and scaling. The SpMV where the multiplication occurs between large sparse matrices with dense vectors is considered one of the most important computational and time-consuming kernels for a wide range of engineering and scientific applications from structural mechanics to quantum physics to fluid dynamics. Thus, to achieve higher performance on this operation, it is important to choose a sparse matrix storage format and utilize the underlying architecture carefully. Following are the key objectives of this work:

• Exploration of current sparse matrix storage formats on many-core architectures using generalized hepta-diagonal sparse solver matrix.
• Study and evaluate various libraries for sparse matrix operations on many-core architectures
• Enhance one of the library to add sparse matrix formats in linear operators
• Performance evaluation of the implemented linear operators using the well-known linear iterative solvers such as BiCG-Stab.

Research Project 3: Exploration of Deep Learning Algorithms using Parallel Programming Models

Deep learning is part of a broader family of machine learning methods based on learning representations of data. An observation (e.g., an image) can be represented in many ways such as a vector of intensity values per pixel, or in a more abstract way as a set of edges, regions of particular shape, etc. Some representations make it easier to learn tasks (e.g., face recognition or facial expression recognition) from examples. One of the promises of deep learning is replacing handcrafted features with efficient algorithms for unsupervised or semi-supervised feature learning and hierarchical feature extraction. Various deep learning architectures such as deep neural networks, convolutional deep neural networks, deep belief networks and recurrent neural networks have been applied to fields like computer vision, automatic speech recognition, natural language processing, audio recognition and bioinformatics where they have been shown to produce state-of-the-art results on various tasks. Deep learning can achieve outstanding results in various fields. However, it requires so significant computational power that massively parallel processors and/or numerous computers are often required for the practical application.

In this project, the students are expected to work on the following objectives:

• Study various deep learning algorithms (any 3), their parallel implementations using different parallel programming models, and their applications.
• Implement the above selected algorithms using any two parallel programming models/frameworks/libraries such as pthreads, CUDA, OpenMP, OpenACC, OpenCL, MPI or combination of these with best possible optimizations.
• Performance evaluation of the above implementations in terms of execution time and cost.

Research Project 4: Exploration of Deep Learning Software on Parallel Architectures

Deep learning is a rapidly growing segment of artificial intelligence. It is increasingly used to deliver near-human level accuracy for image classification, voice recognition, natural language processing, sentiment analysis, recommendation engines, and more. Applications areas include facial recognition, scene detection, advanced medical and pharmaceutical research, and autonomous, self-driving vehicles. Deep learning uses neural networks (DNNs) many layers deep and large datasets to teach computers how to solve perceptual problems, such as detecting recognizable concepts in data, translating or understanding natural languages, interpreting information from input data, and more. Deep learning is used in the research community and in industry to help solve many big data problems such as computer vision, speech recognition and natural language processing. Practical examples include vehicle, pedestrian and landmark identification for driver assistance; image recognition; speech recognition; natural language processing; neural machine translation and cancer detection. Various software solutions has been provided in the market for deep learning using different parallel computing architectures including programming language extensions, libraries, frameworks.

In this project, the students are expected to work on the following objectives:

• Study and explore various software tools/frameworks (any 3) for deep learning on parallel architectures
• Develop a case study using the above selected tools in a specific application area such as computer vision, speech recognition, face recognition, etc.
• Provide the evaluations of the selected frameworks on large datasets in terms of execution time, accuracy, and implementation cost.

Research Project 5: Big Data Analytics using Many-Core Architectures

Recent advances in many-core architectures opened a new opportunity in harnessing their computing power for general purpose computing. However, efficient utilization of these architectures is possible only with expert knowledge of the architecture and code optimizations. Several optimization techniques, tools and libraries have been proposed for these architectures focusing on diverse applications domains such as numerical linear algebra, linear iterative solvers, machine learning algorithms, image processing, fluid-flow dynamics and reservoir simulations. Big data analytics has the goal to analyse massive datasets, which increasingly occur in web-scale business intelligence problems. The common strategy to handle these workloads is to distribute the processing utilizing massive parallel analysis systems or to use big machines able to handle the workload. Today, many different hardware architectures apart from traditional CPUs can be used to process data. GPUs (Graphics Processing Units), MICs (Many Integrated Cores) or FPGAs (Field Programmable Gate Arrays) etc. are usually employed as co-processors to accelerate query execution. Such hardware architectures offer the processing capability to distribute the workload among the CPU and other processors, and enable systems to process bigger workloads.

In this project, the students are expected to work on the following objectives:

• Study and review the latest trends of parallel computing using massively parallel processors in big data analytics
• Explore various software tools/frameworks to be useful for big data analytics using parallel computing
• Develop a case study using these tools (any 3) to perform data processing and analysis on the real datasets