A review on accelerating scientific computations using the Conjugate Gradient method

Debnath, Shreyasee; Tamuli, Manashwi; Ray, Ashok K; Trivedi, Gaurav

doi:10.1109/edcav.2015.7060557

Cited by 4 publications

(2 citation statements)

References 28 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Several works have been implemented on FPGAs to accelerate the processing of PCG. Specifically, in [10], Debnath et al present a comparative analysis of multiple implementations of the conjugate gradient method on various platforms suitable for high performance computing, such as FPGAs and GPUs. They conclude that FPGAs and GPUs are much more efficient than CPUs in calculating the conjugate gradient.…”

Section: Related Workmentioning

confidence: 99%

Preconditioned Conjugate Gradient Acceleration on FPGA-Based Platforms

Malakonakis¹,

Isotton²,

Miliadis

et al. 2022

Electronics

View full text Add to dashboard Cite

Reconfigurable computing can significantly improve the performance and energy efficiency of many applications. However, FPGA-based chips are evolving rapidly, increasing the difficulty of evaluating the impact of new capabilities such as HBM and high-speed links. In this paper, a real-world application was implemented on different FPGAs in order to better understand the new capabilities of modern FPGAs and how new FPGA technology improves performance and scalability. The aforementioned application was the preconditioned conjugate gradient (PCG) method that is utilized in underground analysis. The implementation was done on four different FPGAs, including an MPSoC, taking into account each platform’s characteristics. The results show that today’s FPGA-based chips offer eight times better performance on a memory-bound problem than 5-year-old FPGAs, as they incorporate HBM and can operate at higher clock frequencies.

show abstract

Section: Related Workmentioning

confidence: 99%

Preconditioned Conjugate Gradient Acceleration on FPGA-Based Platforms

Malakonakis¹,

Isotton²,

Miliadis

et al. 2022

Electronics

View full text Add to dashboard Cite

show abstract

“…The overall learning process of an ANN happens when the output of a multi-layer perceptron is compared to the training data and then an adjustment of the weights of each neuron is carried out using a training algorithm. There exist different types of these algorithms: the gradient descentbased method, 37 genetic algorithms, 38 simulated annealing, 39 the conjugate gradient method, 40 the Levenberg-Marquardt algorithm, 41 and the Particle Swarm algorithm. 42 In general, training algorithms for an ANN can be classified as (a) derivative-based algorithms or (b) derivative-free algorithms.…”

Section: Artificial Neural Networkmentioning

confidence: 99%

An Artificial Neural Networks approach to predict low-velocity impact forces in an elastomer material

Rodríguez-Sánchez

2020

SIMULATION

View full text Add to dashboard Cite

The study of the impact phenomenon on rubber-like materials has been traditionally related to lumped parameter modeling or discrete Finite Element models that require experimentation associated with the material behavior at a level of constitutive modeling, and additional testing to validate their operation in case of engineering applications. This article presents an Artificial Neural Network approach to predict and simulate the low-velocity impact force in a thermoplastic elastomer material. Neural network models were trained and validated with experimental data obtained from impact tests in a modified Charpy apparatus. An experimental setup and a data acquisition procedure were set out to record the impact forces on elastomer specimens. The coefficient of determination R2, the Root Mean Square Error, and the Maximum Absolute Error measures were implemented as error functions to evaluate the performance of the neural networks regarding experimental data. Results show that the proposed method helps to predict and derive impact force curves within the range of the training data, since errors below 1% regarding experimental values were obtained. The results also demonstrate that the neural networks can simulate impact force curves within the range of the experimental values without the need to involve parameters of material strain-rate sensitivity. In addition, the approach was tested in another material, and the corresponding results show good prediction capabilities since errors below 1% were obtained. Therefore, it is concluded that the presented artificial neural models, and the approach, could be useful to create solution spaces for low-velocity impact responses of thermoplastic elastomers.

show abstract