Article Preview
TopIntroduction
The main innovation in this research is to exploit neural network inversion as a machine learning technique to recommend architecture for solving MF problem. The recommended architecture executes “SKIP Brute-Force” by (Faheem, 2010) to solve planted MF (15, 4). The recommended architecture aims to maximize speedup and minimizes power consumption within given time and power constraints. The recommended architecture employs different homogeneous parallel paradigms with several hardware architectures: TESLA K20 Kepler GPUs ACCELERATOR, each GPU has 13 Multiprocessors each of 192 CUDA Cores which using CUDA and MPI as hybrid programing paradigm and CPU Multi-core based on XEON dual physical processor, each has 12 core which using hybrid MPI and OpenMP programming paradigm
The problem parameters are:
- •
Number of sequences (T): varies from 25 up to 2400.
- •
Sequence Length (N): varies from 20 up to 240.
- •
Problem Size (NxT): varies from 6 K up to 48 K.
The combination between various input problem size and different hardware architecture configuration generates 140 running case, each contains: T, N, problem size, specified architecture, power consumption and execution time, the exploited architectures are shown in Table 1, training dataset are described in tables from Table 2 to Table 8
Table 1.
Recommending System exploited architectures
| Architecture specification |
| CPU Based architecture | 16*24 XEON dual physical processor |
| 8*24 XEON dual physical processor |
| 4*24 XEON dual physical processor |
| 2*24 XEON dual physical processor |
| 1*24 XEON dual physical processor |
| GPU Based Architecture | 1 NVidia TESLA K20 Kepler GPU CUDA Core |
| 2 NVidia TESLA K20 Kepler GPU CUDA Core |
Sequencing technology recent developments permit efficient and cost-effective acquisition of genomic data (Xiong, Zhongming, Arnold, and Yu, 2009). DNA motifs usually are considered as Transcription Factor Binding Sites (TFBS) where proteins are attached to regulate the expression of genes (Yu, Mani, Cao, and Brenner, 2010; Hu, Yu, Taylor, Chinnaiyan, Qin, 2010). Although there are several algorithms available to tackle this problem, MF is recognized as NP- complete problem (Nondeterministic Polynomial Time Order Problem). There is also different Accelerators, (S/W) and (H/W), have been developed to accelerate MF algorithms.