TY - GEN
T1 - STAFF
T2 - 55th Annual Design Automation Conference, DAC 2018
AU - Gupta, Ujjwal
AU - Babu, Manoj
AU - Ayoub, Raid
AU - Kishinevsky, Michael
AU - Paterna, Francesco
AU - Ogras, Umit
N1 - Funding Information:
This work was supported partially by Semiconductor Research Corporation (SRC) task 2721.001 and National Science Foundation (NSF) grant CNS-1526562.
Funding Information:
In this ,paper ew esentpr the AFFST orkwframe that guarantees ,stability performs online efeatur selection with linear x-comple ,ity and dynamically changes the forgetting .factor eW valuatee AFFST by edictingpr the equencyfr sensitivity of a graphics unit in a cialcommer Intel platform. The oposeprd orkwframe videsopr fast tracking with up to edcompar to xistinge state-of-the-art techniques. Acknowledgements: This orkw was supported partially by Semiconductor chResear Corporation (SRC) task 2721.001 and National Science Foundation (NSF) grant CNS-1526562.
Publisher Copyright:
© 2018 Association for Computing Machinery.
PY - 2018/6/24
Y1 - 2018/6/24
N2 - Dynamic resource management techniques rely on power consumption and performance models to optimize the operating frequency and utilization of processing elements, such as CPU and GPU. Despite the importance of these decisions, many existing approaches rely on fixed power and performance models that are learned online. However, o.ine models cannot guarantee accuracy when workloads differ signifucantly from the training available at design time. This paper presents an online learning framework (STAFF) that constructs adaptive run-time models for stationary and non-stationary workloads. STAFF is the first framework that (1) guarantees stability while quickly adapting to workload changes, (2) performs online feature selection with linear complexity, and (3) adapts to new model coe.cients by employing adaptively varying forgetting factor, all at the same time. Experiments on an Intelr CoreTM i5 6th generation platform demonstrate up to 6. improvement in the performance prediction accuracy compared to existing techniques.
AB - Dynamic resource management techniques rely on power consumption and performance models to optimize the operating frequency and utilization of processing elements, such as CPU and GPU. Despite the importance of these decisions, many existing approaches rely on fixed power and performance models that are learned online. However, o.ine models cannot guarantee accuracy when workloads differ signifucantly from the training available at design time. This paper presents an online learning framework (STAFF) that constructs adaptive run-time models for stationary and non-stationary workloads. STAFF is the first framework that (1) guarantees stability while quickly adapting to workload changes, (2) performs online feature selection with linear complexity, and (3) adapts to new model coe.cients by employing adaptively varying forgetting factor, all at the same time. Experiments on an Intelr CoreTM i5 6th generation platform demonstrate up to 6. improvement in the performance prediction accuracy compared to existing techniques.
UR - http://www.scopus.com/inward/record.url?scp=85053682642&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85053682642&partnerID=8YFLogxK
U2 - 10.1145/3195970.3196122
DO - 10.1145/3195970.3196122
M3 - Conference contribution
AN - SCOPUS:85053682642
SN - 9781450357005
T3 - Proceedings - Design Automation Conference
BT - Proceedings of the 55th Annual Design Automation Conference, DAC 2018
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 24 June 2018 through 29 June 2018
ER -