Extended scaled neural predictor for improved branch prediction

Zhou, Zihao; Kejriwal, Mayank; Miikkulainen, Risto

doi:10.1109/ijcnn.2013.6707059

Cited by 2 publications

(16 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Seznec [32] uses multiple predictor tables, hashing into them using various ranges of the path history. Zhou et al [41] looks deeper into the findings of Seznecs paper and brings some key insights into the hashing and parameter configurations.…”

Section: Path-based Historymentioning

confidence: 99%

“…Using this implementation, when the perceptrons outcome is around zero, it is a sign that the perceptron is not trained enough to make confident predictions. Zhou et al [41] studied a mechanism for dynamically changing the training threshold through the use of a saturating counter.…”

Section: Training Thresholdmentioning

confidence: 99%

“…They proposed using scaling of individual weights by predetermined coefficients based on the history position. Zhou et al [41] studied the effect of various implementations of scaling coefficients.…”

Section: Scaling Coefficientsmentioning

confidence: 99%

“…Jimenez and Lin [14] use a trace-based simulation technique, and report their compiler information. Zhou et al [41] use SimpleScalar for their simulations, but did not disclose any compiler information regarding their benchmarks. We use two statistics to determine the effectiveness of a configuration: memory footprint and branch direction-prediction rate.…”

Section: Simulation Environmentmentioning

confidence: 99%

“…We use branch prediction as a case study in order to quantify the learning and predictive characteristics of our design. Using branch prediction, we are able to compare our perceptron design in terms of memory footprint and prediction performance to the state-of-the-art neural branch predictors [12,13,14,41].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Balancing the learning ability and memory demand of a perceptron-based dynamically trainable neural network

et al. 2018

View full text Add to dashboard Cite

Artificial neural networks (ANNs) have become a popular means of solving complex problems in prediction based applications such as image and natural language processing. Two challenges prominent in the neural network domain are the practicality of hardware implementation and dynamically training the network. In this study we address these challenges with a development methodology that balances the hardware footprint and the quality of the ANN. We use the well-known perceptron-based branch prediction problem as a case study for demonstrating this methodology. This problem is perfect to analyze dynamic hardware implementations of ANNs because it exists in hardware and trains dynamically. Using our hierarchical configuration search space exploration, we show that we can decrease the memory footprint of a standard perceptron-based branch predictor by 2.3x with only a 0.6% decrease in prediction accuracy.

show abstract

Section: Path-based Historymentioning

confidence: 99%

Section: Training Thresholdmentioning

confidence: 99%

Section: Scaling Coefficientsmentioning

confidence: 99%

Section: Simulation Environmentmentioning

confidence: 99%