Dynamic Power Management under Uncertain Information

Jung, Hwisung; Pedram, Massoud

doi:10.1109/date.2007.364434

Cited by 16 publications

(15 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To be more realistic, we consider in this work that the SR state cannot be directly obtained by the PM. In contrast to previous work on POMDP [7] [8], the PM has no prior knowledge of the characteristics of the SR. Therefore, workload prediction has to be incorporated to provide partial information to the PM so that the PM can learn in the observation domain of the SR.…”

Section: System Modelmentioning

confidence: 74%

“…Other enhancements include time-indexed semi-MDP of Simunic et al [6]. To cope with uncertainties in the underlying hardware state, DPM policies based on partially observable Markov decision process (POMDP) have been proposed in [7] and [8]. Note that in the aforesaid stochastic DPM approaches, request inter-arrival times and system service times are modeled as stationary processes that satisfy certain probability distributions.…”

Section: Introductionmentioning

confidence: 98%

See 1 more Smart Citation

Deriving a near-optimal power management policy using model-free reinforcement learning and Bayesian classification

Wang

Xie

Ammari

et al. 2011

Proceedings of the 48th Design Automation Conference

Self Cite

View full text Add to dashboard Cite

To cope with the variations and uncertainties that emanate from hardware and application characteristics, dynamic power management (DPM) frameworks must be able to learn about the system inputs and environment and adjust the power management policy on the fly. In this paper we present an online adaptive DPM technique based on model-free reinforcement learning (RL), which is commonly used to control stochastic dynamical systems. In particular, we employ temporal difference learning for semiMarkov decision process (SMDP) for the model-free RL. In addition a novel workload predictor based on an online Bayes classifier is presented to provide effective estimates of the workload states for the RL algorithm. In this DPM framework, power and latency tradeoffs can be precisely controlled based on a user-defined parameter. Experiments show that amount of average power saving (without any increase in the latency) is up to 16.7% compared to a reference expert-based approach. Alternatively, the per-request latency reduction without any power consumption increase is up to 28.6% compared to the expert-based approach.

show abstract

Section: System Modelmentioning

confidence: 74%

Section: Introductionmentioning

confidence: 98%

Deriving a near-optimal power management policy using model-free reinforcement learning and Bayesian classification

Wang

Xie

Ammari

et al. 2011

Proceedings of the 48th Design Automation Conference

Self Cite

View full text Add to dashboard Cite

show abstract

“…The workload variation has a significant impact on the system performance and power consumption. Thus, a robust power management technique must consider the uncertainty and variability that emanate from the environment, hardware and application characteristics [11] and must be able to interact with the environment to obtain information that can be processed to produce optimal policies.…”

Section: Introductionmentioning

confidence: 99%

Extending the Service Lifetime of a Battery-Powered System Supporting Multiple Active Modes

Triki¹

2017

IOSR JEEE

View full text Add to dashboard Cite

show abstract

“…Stochastic methods are naturally selected for modeling and optimization of such system. Stochastic models such as Markov decision process ( [4]), SemiMarkov decision process ( [5]), and Partially Observable Markov Decision process ( [6], [7]) have been investigated by previous DPM research. All of these models assume that the workload is intrinsically Markovian and this embedded Markov model can be reconstructed (or trained) based on the given observation sequence.…”

Section: Introductionmentioning

confidence: 99%

“…The authors of [6] consider an integrated circuit under process, voltage and temperature variations as a partially observable system. POMDP is applied to search for the optimal power management policy of such system.…”

Section: Introductionmentioning

confidence: 99%

A Framework of Stochastic Power Management Using Hidden Markov Model

Tan

Qiu

2008

2008 Design, Automation and Test in Europe

View full text Add to dashboard Cite

-The effectiveness of stochastic power management relies on the accurate system and workload model and effective policy optimization. Workload modeling is a machine learning procedure that finds the intrinsic pattern of the incoming tasks based on the observed workload attributes. Markov Decision Process (MDP) based model has been widely adopted for stochastic power management because it delivers provable optimal policy. Given a sequence of observed workload attributes, the hidden Markov model (HMM) of the workload is trained. If the observed workload attributes and states in the workload model do not have one-to-one correspondence, the MDP becomes a Partially Observable Markov Decision Process (POMDP). This paper presents a framework of modeling and optimization for stochastic power management using HMM and POMDP. The proposed technique discovers the HMM of the workload by maximizing the likelihood of the observed attribute sequence. The POMDP optimization is formulated and solved as a quadraticly constrained linear programming (QCLP). Compared with traditional optimization technique, which is based on value iteration, the QCLP based optimization provides superior policy by enabling stochastic control.

show abstract

Dynamic Power Management under Uncertain Information

Cited by 16 publications

References 9 publications

Deriving a near-optimal power management policy using model-free reinforcement learning and Bayesian classification

Deriving a near-optimal power management policy using model-free reinforcement learning and Bayesian classification

Extending the Service Lifetime of a Battery-Powered System Supporting Multiple Active Modes

A Framework of Stochastic Power Management Using Hidden Markov Model

Contact Info

Product

Resources

About