A scalable storage system for structured data based on higher order index array

Omar, Mehnuma Tabassum; Hasan, K. M. Azharul

doi:10.1145/3006299.3006333

Cited by 6 publications

(1 citation statement)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This scheme uses the concept of dimension conversion where n-dimension is converted into two-dimension [3]. To compute the SHDP, we use Scalable Tensor Structure (STS) that can scale (extension or reduction) the dimension size during runtime for the inclusion of the new tensor elements in future [4]. The structure is chunked in such a way that the task parallel scheme maintains simplicity for the parallel tensor computation [5,6].…”

Section: Introductionmentioning

confidence: 99%

Parallel Computation of Hadamard Product for Dynamic Dataset on Google Colab

Shariar

Hasan

2021

ICT With Intelligent Applications

Self Cite

View full text Add to dashboard Cite

Hadamard product, the element-wise product of tensors, is an important tool for many real-life applications such as machine LSTM, multimodal learning, object recognition, and segmentation. In this paper, we propose a faster processing of hadamard product by the parallel computation on graphics processing unit (GPU) for dynamic data. Most of the computations for real-life data are multidimensional and dynamic which expand during run time. In this scheme, we use dynamic memory allocation that allows further expansion of the dimension size and high memory utilization based on the notion of extendable array. Since high dimensionality is difficult to handle, we use a dimension conversion technique where the n-dimensional array structure is transformed into a 2-dimensional structure. Furthermore, to handle the future data, we apply the converted extendable array structure to get Scalable Hadamard Product (SHDP) for dynamic dataset. Then, we implement a parallel algorithm for the parallel processing of the SHDP that partitions and distributes the element-wise product during runtime. We utilize the maximum available GPU threads for the faster performance of the hadamard product. We found good speed-up for the proposed partitioning while applying to "Google Colab".

show abstract

Section: Introductionmentioning

confidence: 99%