A domain-specific supercomputer for training deep neural networks

Jouppi, Norman P.; Yoon, Doe Hyun; Kurian, George Thomas; Li, Sheng; Patil, Nishant; Laudon, James; Young, Cliff; Patterson, David A.

doi:10.1145/3360307

Cited by 192 publications

(90 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…They show that when we use larger systolic arrays (approaching the baseline size) the average PE array utilization always decreases. This behaviour has been documented in prior studies [20], [24] and occurs due to difficulties when mapping GEMM operations with irregular shapes. Input matrices fall into this category, when they have a dimension smaller than the width of the PE array.…”

Section: Resultssupporting

confidence: 66%

Design Space Exploration of Accelerators and End-to-End DNN Evaluation with TFLITE-SOC

Agostini

Dong

Karimi

et al. 2020

2020 IEEE 32nd International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD)

View full text Add to dashboard Cite

Section: Resultssupporting

confidence: 66%

Design Space Exploration of Accelerators and End-to-End DNN Evaluation with TFLITE-SOC

Agostini

Dong

Karimi

et al. 2020

2020 IEEE 32nd International Symposium on Computer Architecture and High Performance Computing (SBAC-PAD)

View full text Add to dashboard Cite

“…Computationally, how do we scale the training, testing and deployment of complex PIML models on large datasets efficiently, so that they perform well in a rapidly changing computational landscape [130]?…”

Section: Synthesis and Outlookmentioning

confidence: 99%

Physics-informed machine learning: case studies for weather and climate modelling

Kashinath

Mustafa

Albert

et al. 2021

Phil. Trans. R. Soc. A.

280

143

View full text Add to dashboard Cite

Machine learning (ML) provides novel and powerful ways of accurately and efficiently recognizing complex patterns, emulating nonlinear dynamics, and predicting the spatio-temporal evolution of weather and climate processes. Off-the-shelf ML models, however, do not necessarily obey the fundamental governing laws of physical systems, nor do they generalize well to scenarios on which they have not been trained. We survey systematic approaches to incorporating physics and domain knowledge into ML models and distill these approaches into broad categories. Through 10 case studies, we show how these approaches have been used successfully for emulating, downscaling, and forecasting weather and climate processes. The accomplishments of these studies include greater physical consistency, reduced training time, improved data efficiency, and better generalization. Finally, we synthesize the lessons learned and identify scientific, diagnostic, computational, and resource challenges for developing truly robust and reliable physics-informed ML models for weather and climate processes. This article is part of the theme issue ‘Machine learning for weather and climate modelling’.

show abstract

“…Jouppi et al divided the optimization process into two stages: first simplifying the optimization to TSP and then solving TSP. is is a relatively new method and achieved good results [18]. Huang et al proposed a new path merging method and path length saving value calculation formula, improved the classical saving method to solve the optimization, and improved the performance of the original algorithm [19].…”

Section: Related Workmentioning

confidence: 99%

[Retracted] International Trade Path with Multi‐Polarization based on Multidirectional Mutation Genetic Algorithm Enabled Neural Network

Zhang

Chong

Abdullah

et al. 2021

Computational Intelligence and Neuroscience

View full text Add to dashboard Cite

At present, the development speed of international trade cannot catch up with the economic development speed, and the insufficient development speed of international trade will directly affect the rapid development of national economy. In order to solve the problem of international trade, the overall optimal scheduling of trade vehicles and the optimal planning of trade transportation path are very important to improve enterprise services, reduce enterprise costs, increase enterprise benefits, and enhance enterprise competitiveness. The second development of the program is based on the programming interface provided by Baidu map. This paper proposes a neural network algorithm for genetic optimization of multiple mutations, which overcomes the shortcoming of traditional genetic algorithm population “ten” character distribution by mixing multiple coding methods, and enhances the local search ability of genetic algorithm by introducing a new large-mutation small-range search population. The example application shows that the optimization method can realize the optimization of international trade path under real road conditions and greatly improve the work efficiency of actual trade.

show abstract

A domain-specific supercomputer for training deep neural networks

Abstract: Google's TPU supercomputers train deep neural networks 50x faster than general-purpose supercomputers running a high-performance computing benchmark.

Cited by 192 publications

References 8 publications

Design Space Exploration of Accelerators and End-to-End DNN Evaluation with TFLITE-SOC

Design Space Exploration of Accelerators and End-to-End DNN Evaluation with TFLITE-SOC

Physics-informed machine learning: case studies for weather and climate modelling

[Retracted] International Trade Path with Multi‐Polarization based on Multidirectional Mutation Genetic Algorithm Enabled Neural Network

Contact Info

Product

Resources

About