GAE-LCT: A Run-Time GA-Based Classifier Evolution Method for Hardware LCT Controlled SoC Performance-Power Optimization

Anmol Surhonne; Nguyen Anh Vu Doan; Florian Maurer; Thomas Wild; Andreas Herkersdorf

doi:10.1007/978-3-031-21867-5_18

GAE-LCT: A Run-Time GA-Based Classifier Evolution Method for Hardware LCT Controlled SoC Performance-Power Optimization

Anmol Surhonne, Nguyen Anh Vu Doan, Florian Maurer, Thomas Wild, Andreas Herkersdorf

Chair of Integrated Systems

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

2 Scopus citations

Abstract

Learning classifier tables (LCTs) are classifier based and lightweight hardware reinforcement learning building blocks which inherit the concepts of learning classifier systems. LCTs are used as a per-core low level controllers to learn and optimize potentially conflicting objectives e.g. achieving a performance target under a power budget. A supervisor is used at the system level which translate system and application requirements into objectives for the LCTs. The classifier population in the LCTs has to be evolved in run-time to adapt to the changes in the mode, performance targets, constraints or workload being executed. Towards this goal, we present GAE-LCT, a genetic algorithm (GA) based classifier evolution for hardware learning classifier tables. The GA uses accuracy to evolve classifiers in run-time. We introduce extensions to the LCT to enable accuracy based genetic algorithm. The GA runs as a software process on one of the cores and interacts with the hardware LCT via interrupts. We evaluate our work using DVFS on an FPGA using Leon3 cores. We demonstrate GAE-LCT’s ability to generate accurate classifiers in run-time from scratch. GAE-LCT achieves 5% lower difference to IPS reference and 51.5% lower power budget overshoot compared to Q-table while requiring 75% less memory. The hybrid GAE-LCT also requires 12 times less software overhead compared to a full software implementation.

Original language	English
Title of host publication	Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings
Editors	Martin Schulz, Carsten Trinitis, Nikela Papadopoulou, Thilo Pionteck
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	271-285
Number of pages	15
ISBN (Print)	9783031218668
DOIs	https://doi.org/10.1007/978-3-031-21867-5_18
State	Published - 2022
Event	35th International Conference on Architecture of Computing Systems, ARCS 2022 - Heilbronn, Germany Duration: 13 Sep 2022 → 15 Sep 2022

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13642 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	35th International Conference on Architecture of Computing Systems, ARCS 2022
Country/Territory	Germany
City	Heilbronn
Period	13/09/22 → 15/09/22

Keywords

DVFS
Genetic algorithms
Learning classifier systems
Model-free control
Reinforcement learning
Run-time management

Access to Document

10.1007/978-3-031-21867-5_18

Cite this

Surhonne, A., Doan, N. A. V., Maurer, F., Wild, T., & Herkersdorf, A. (2022). GAE-LCT: A Run-Time GA-Based Classifier Evolution Method for Hardware LCT Controlled SoC Performance-Power Optimization. In M. Schulz, C. Trinitis, N. Papadopoulou, & T. Pionteck (Eds.), Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings (pp. 271-285). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13642 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-21867-5_18

Surhonne, Anmol ; Doan, Nguyen Anh Vu ; Maurer, Florian et al. / GAE-LCT : A Run-Time GA-Based Classifier Evolution Method for Hardware LCT Controlled SoC Performance-Power Optimization. Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings. editor / Martin Schulz ; Carsten Trinitis ; Nikela Papadopoulou ; Thilo Pionteck. Springer Science and Business Media Deutschland GmbH, 2022. pp. 271-285 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{33a1c2fad88a43aa9805ef50b601e511,

title = "GAE-LCT: A Run-Time GA-Based Classifier Evolution Method for Hardware LCT Controlled SoC Performance-Power Optimization",

abstract = "Learning classifier tables (LCTs) are classifier based and lightweight hardware reinforcement learning building blocks which inherit the concepts of learning classifier systems. LCTs are used as a per-core low level controllers to learn and optimize potentially conflicting objectives e.g. achieving a performance target under a power budget. A supervisor is used at the system level which translate system and application requirements into objectives for the LCTs. The classifier population in the LCTs has to be evolved in run-time to adapt to the changes in the mode, performance targets, constraints or workload being executed. Towards this goal, we present GAE-LCT, a genetic algorithm (GA) based classifier evolution for hardware learning classifier tables. The GA uses accuracy to evolve classifiers in run-time. We introduce extensions to the LCT to enable accuracy based genetic algorithm. The GA runs as a software process on one of the cores and interacts with the hardware LCT via interrupts. We evaluate our work using DVFS on an FPGA using Leon3 cores. We demonstrate GAE-LCT{\textquoteright}s ability to generate accurate classifiers in run-time from scratch. GAE-LCT achieves 5% lower difference to IPS reference and 51.5% lower power budget overshoot compared to Q-table while requiring 75% less memory. The hybrid GAE-LCT also requires 12 times less software overhead compared to a full software implementation.",

keywords = "DVFS, Genetic algorithms, Learning classifier systems, Model-free control, Reinforcement learning, Run-time management",

author = "Anmol Surhonne and Doan, {Nguyen Anh Vu} and Florian Maurer and Thomas Wild and Andreas Herkersdorf",

note = "Publisher Copyright: {\textcopyright} 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 35th International Conference on Architecture of Computing Systems, ARCS 2022 ; Conference date: 13-09-2022 Through 15-09-2022",

year = "2022",

doi = "10.1007/978-3-031-21867-5_18",

language = "English",

isbn = "9783031218668",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "271--285",

editor = "Martin Schulz and Carsten Trinitis and Nikela Papadopoulou and Thilo Pionteck",

booktitle = "Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings",

}

Surhonne, A, Doan, NAV, Maurer, F, Wild, T & Herkersdorf, A 2022, GAE-LCT: A Run-Time GA-Based Classifier Evolution Method for Hardware LCT Controlled SoC Performance-Power Optimization. in M Schulz, C Trinitis, N Papadopoulou & T Pionteck (eds), Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13642 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 271-285, 35th International Conference on Architecture of Computing Systems, ARCS 2022, Heilbronn, Germany, 13/09/22. https://doi.org/10.1007/978-3-031-21867-5_18

GAE-LCT: A Run-Time GA-Based Classifier Evolution Method for Hardware LCT Controlled SoC Performance-Power Optimization. / Surhonne, Anmol; Doan, Nguyen Anh Vu; Maurer, Florian et al.
Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings. ed. / Martin Schulz; Carsten Trinitis; Nikela Papadopoulou; Thilo Pionteck. Springer Science and Business Media Deutschland GmbH, 2022. p. 271-285 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13642 LNCS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - GAE-LCT

T2 - 35th International Conference on Architecture of Computing Systems, ARCS 2022

AU - Surhonne, Anmol

AU - Doan, Nguyen Anh Vu

AU - Maurer, Florian

AU - Wild, Thomas

AU - Herkersdorf, Andreas

PY - 2022

Y1 - 2022

N2 - Learning classifier tables (LCTs) are classifier based and lightweight hardware reinforcement learning building blocks which inherit the concepts of learning classifier systems. LCTs are used as a per-core low level controllers to learn and optimize potentially conflicting objectives e.g. achieving a performance target under a power budget. A supervisor is used at the system level which translate system and application requirements into objectives for the LCTs. The classifier population in the LCTs has to be evolved in run-time to adapt to the changes in the mode, performance targets, constraints or workload being executed. Towards this goal, we present GAE-LCT, a genetic algorithm (GA) based classifier evolution for hardware learning classifier tables. The GA uses accuracy to evolve classifiers in run-time. We introduce extensions to the LCT to enable accuracy based genetic algorithm. The GA runs as a software process on one of the cores and interacts with the hardware LCT via interrupts. We evaluate our work using DVFS on an FPGA using Leon3 cores. We demonstrate GAE-LCT’s ability to generate accurate classifiers in run-time from scratch. GAE-LCT achieves 5% lower difference to IPS reference and 51.5% lower power budget overshoot compared to Q-table while requiring 75% less memory. The hybrid GAE-LCT also requires 12 times less software overhead compared to a full software implementation.

AB - Learning classifier tables (LCTs) are classifier based and lightweight hardware reinforcement learning building blocks which inherit the concepts of learning classifier systems. LCTs are used as a per-core low level controllers to learn and optimize potentially conflicting objectives e.g. achieving a performance target under a power budget. A supervisor is used at the system level which translate system and application requirements into objectives for the LCTs. The classifier population in the LCTs has to be evolved in run-time to adapt to the changes in the mode, performance targets, constraints or workload being executed. Towards this goal, we present GAE-LCT, a genetic algorithm (GA) based classifier evolution for hardware learning classifier tables. The GA uses accuracy to evolve classifiers in run-time. We introduce extensions to the LCT to enable accuracy based genetic algorithm. The GA runs as a software process on one of the cores and interacts with the hardware LCT via interrupts. We evaluate our work using DVFS on an FPGA using Leon3 cores. We demonstrate GAE-LCT’s ability to generate accurate classifiers in run-time from scratch. GAE-LCT achieves 5% lower difference to IPS reference and 51.5% lower power budget overshoot compared to Q-table while requiring 75% less memory. The hybrid GAE-LCT also requires 12 times less software overhead compared to a full software implementation.

KW - DVFS

KW - Genetic algorithms

KW - Learning classifier systems

KW - Model-free control

KW - Reinforcement learning

KW - Run-time management

UR - http://www.scopus.com/inward/record.url?scp=85144828406&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-21867-5_18

DO - 10.1007/978-3-031-21867-5_18

M3 - Conference contribution

AN - SCOPUS:85144828406

SN - 9783031218668

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 271

EP - 285

BT - Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings

A2 - Schulz, Martin

A2 - Trinitis, Carsten

A2 - Papadopoulou, Nikela

A2 - Pionteck, Thilo

PB - Springer Science and Business Media Deutschland GmbH

Y2 - 13 September 2022 through 15 September 2022

ER -

Surhonne A, Doan NAV, Maurer F, Wild T, Herkersdorf A. GAE-LCT: A Run-Time GA-Based Classifier Evolution Method for Hardware LCT Controlled SoC Performance-Power Optimization. In Schulz M, Trinitis C, Papadopoulou N, Pionteck T, editors, Architecture of Computing Systems - 35th International Conference, ARCS 2022, Proceedings. Springer Science and Business Media Deutschland GmbH. 2022. p. 271-285. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-21867-5_18