Menu

Thesis Project for SCI-Arc Edge Synthetic Landscapes

IEEE Transactions on Parallel and Distributed Systems, 2020

Despite foreseeing tremendous speedups over conventional deep neural networks, the performance advantage of binarized neural networks (BNNs) has merely been showcased on general-purpose processors such as CPUs and GPUs. In fact, due to being unable to leverage bit-level-parallelism with a word-based architecture, GPUs have been criticized for extremely low utilization (1%) when executing BNNs. Consequently, the latest tensorcores in NVIDIA Turing GPUs start to experimentally support bit computation. In this work, we look into this brand new bit computation capability and characterize its unique features. We show that the stride of memory access can significantly affect performance delivery and a data-format co-design is highly desired to support the tensorcores for achieving superior performance than existing software solutions without tensorcores. We realize the tensorcore-accelerated BNN design, particularly the major functions for fully-connect and convolution layers-bit matrix multiplication and bit convolution. Evaluations on two NVIDIA Turing GPUs show that, with ResNet-18, our BTC-BNN design can process ImageNet at a rate of 5.6K images per second, 77% faster than state-of-the-art. Our BNN approach is released on https://github.com/pnnl/TCBNN.

2010

In this paper, a new lossless data compression method that is based on digram coding is introduced. This data compression method uses semi-static dic- tionaries: All of the used characters and most frequently used two character blocks (digrams) in the source are found and inserted into a dictionary in the first pass, compression is performed in the second pass. This two-pass structure is repeated several times and in every iteration particular number of elements is inserted in the dictionary until the dictionary is filled. This algorithm (ISSDC: Iterative Semi- Static Digram Coding) also includes some mechanisms that can decide about total number of iterations and dictionary size whenever these values are not given by the user. Our experiments show that ISSDC is better than LZW/GIF and BPE in com- pression ratio. It is worse than DEFLATE in compression of text and binary data, but better than PNG (which uses DEFLATE compression) in lossless compression of simple images.

2012 11th International Symposium on Parallel and Distributed Computing, 2012

An efficient memory bandwidth utilization for GPU accelerators is crucial for memory bound applications. In medical imaging, the performance of many kernels is limited by the available memory bandwidth since only a few operations are performed per pixel. For such kernels only a fraction of the compute power provided by GPU accelerators can be exploited and performance is predetermined by memory bandwidth. As a remedy, this paper investigates the optimal utilization of available memory bandwidth by means of increasing in-flight memory transactions. Instead of doing this manually for different GPU accelerators, the required CUDA and OpenCL code is automatically generated from descriptions in a Domain-Specific Language (DSL) for the considered application domain. Moreover, the DSL is extended to also support global reduction operators. We show that the generated target-specific code improves bandwidth utilization for memory-bound kernels significantly. Moreover, competitive performance compared to the GPU back end of the widely used image processing library OpenCV can be achieved.

2001

Quito, 2020

This research focuses on the development of an application using Python whose functions are to select points manually and generate effective and optimal trajectories for a KUKA KR20 robotic arm through the implementation of artificial vision. This research uses any type of camera device, the calibration and distortion correction processes are performed to improve the accuracy of the method. Using stereo vision, 3D reconstruction is performed with the use of two perfectly aligned images. Through reconstruction, the depth of any point captured is calculated, to then triangulate and re project the point obtaining its 3D coordinates. These calculated coordinates are used as inputs to the program that generates the KRL code. The code generated has the complete information of the action to perform, such as the type of the trajectory, the coordinates of each point in the space and the velocities with which the arm reaches those points.

Communications in Nonlinear Science and Numerical Simulation, 2012

Background and objectives: Whole transcriptome gene expression data from microarray and next-generation sequencing (NGS) have made possible for large-scale data analysis for biomarker discovery and pathogenesis. IgA nephropathy (IgAN) is the most popular nephritis in eastern Asia and in the world, but its pathogenesis has not been fully elucidated. We hereby study the gene cluster expression abnormality of IgAN compared to other kidney diseases and healthy controls using tissue microarray data.

This article addresses contemporary efforts to understand how the earliest practitioners of philosophy conceived of the philosophic life. It argues that, for Plato, the concept of bios was a central, animating, and structuring object of philosophic inquiry. Concentration on the imagery Plato employed to draw bios into the purview of philosophic contemplation and choice points to interpretative avenues that further the aim of treating the dialogues as complex, integrated wholes, and offers a new approach to the question of the status of image-making in them. The article concludes with thoughts on how an exploration of bios might extend beyond Plato to Aristotle, via an examination of his treatment of the range of human and animal bioi, suggesting that such an examination clarifies the relationship between his analysis of the polis-dwelling animal and his broader investigation of living beings as such.