An Energy-aware Online Learning Framework for Resource Management in Heterogeneous Platforms

Mandal, Sumit K.; Bhat, Ganapati; Doppa, Janardhan Rao; Pande, Partha Pratim; Ogras, Ümit Y.

doi:10.1145/3386359

Cited by 26 publications

(12 citation statements)

References 35 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To tackle this challenge, we employ the epsilon-greedy algorithm. This algorithm is one of the commonly-adopted algorithms [62,79,89,91] due to its effectiveness and simplicity (Section 4.2) -it achieves comparable prediction accuracy to other complex algorithms, such as Exp3, Softmax, UCB, and Thompson Sampling, with lower overhead [20,116].…”

Section: Autoflmentioning

confidence: 99%

AutoFL: Enabling Heterogeneity-Aware Energy Efficient Federated Learning

Kim¹,

Wu²

2021

Preprint

View full text Add to dashboard Cite

Federated learning enables a cluster of decentralized mobile devices at the edge to collaboratively train a shared machine learning model, while keeping all the raw training samples on device. This decentralized training approach is demonstrated as a practical solution to mitigate the risk of privacy leakage. However, enabling efficient FL deployment at the edge is challenging because of non-IID training data distribution, wide system heterogeneity and stochastic-varying runtime effects in the field. This paper jointly optimizes time-to-convergence and energy efficiency of state-of-the-art FL use cases by taking into account the stochastic nature of edge execution. We propose AutoFL by tailor-designing a reinforcement learning algorithm that learns and determines which K participant devices and per-device execution targets for each FL model aggregation round in the presence of stochastic runtime variance, system and data heterogeneity. By considering the unique characteristics of FL edge deployment judiciously, AutoFL achieves 3.6 times faster model convergence time and 4.7 and 5.2 times higher energy efficiency for local clients and globally over the cluster of K participants, respectively.

show abstract

Section: Autoflmentioning

confidence: 99%

AutoFL: Enabling Heterogeneity-Aware Energy Efficient Federated Learning

Kim¹,

Wu²

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…On the other hand, if it keeps exploring all possible actions, the convergence might get slower. To solve this problem, we employ epsilon-greedy algorithm, which is one of the widely adopted randomized greedy algorithms in this domain [64,71,74], due to its simplicity and effectiveness (Section 4. Low Latency Overhead: For the real-time inference exe-cution on the energy constrained edge devices, latency overhead is also one of the crucial factors.…”

Section: Autoscalementioning

confidence: 99%

“…Gaudette et al proposed to use arbitrary polynomial chaos expansions to consider the impact of various sources of uncertainties on mobile user experience [29]. Other works explored the use of reinforcement learning to handle runtime variance for web browsers, for latency-critical cloud services, and for CPUs [12,64,71].…”

Section: Related Workmentioning

confidence: 99%

AutoScale: Optimizing Energy Efficiency of End-to-End Edge Inference under Stochastic Variance

Kim,

2020

Preprint

View full text Add to dashboard Cite

Deep learning inference is increasingly run at the edge. As the programming and system stack support becomes mature, it enables acceleration opportunities within a mobile system, where the system performance envelope is scaled up with a plethora of programmable co-processors. Thus, intelligent services designed for mobile users can choose between running inference on the CPU or any of the co-processors on the mobile system, or exploiting connected systems, such as the cloud or a nearby, locally connected system. By doing so, the services can scale out the performance and increase the energy efficiency of edge mobile systems. This gives rise to a new challenge-deciding when inference should run where. Such execution scaling decision becomes more complicated with the stochastic nature of mobile-cloud execution, where signal strength variations of the wireless networks and resource interference can significantly affect real-time inference performance and system energy efficiency. To enable accurate, energy-efficient deep learning inference at the edge, this paper proposes AutoScale. AutoScale is an adaptive and light-weight execution scaling engine built upon the customdesigned reinforcement learning algorithm. It continuously learns and selects the most energy-efficient inference execution target by taking into account characteristics of neural networks and available systems in the collaborative cloudedge execution environment while adapting to the stochastic runtime variance. Real system implementation and evaluation, considering realistic execution scenarios, demonstrate an average of 9.8 and 1.6 times energy efficiency improvement for DNN edge inference over the baseline mobile CPU and cloud offloading, while meeting the real-time performance and accuracy requirement.

show abstract

“…Power consumption has been one of the primary design considerations for more than a decade [10][11][12]. Hence, energy-efficient techniques have been widely studied to harness the processing power within available power and thermal budgets [13][14][15][16][17][18].…”

Section: Introductionmentioning

confidence: 99%

“…Therefore, the power consumption of major PEs should be modeled accurately. Then, dynamic thermal and power management (DTPM) algorithms can utilize these models to control each PE more effectively [11,[23][24][25][26][27][28].…”

Section: Introductionmentioning

confidence: 99%

Per-Core Power Modeling for Heterogenous SoCs

et al. 2021

Self Cite

View full text Add to dashboard Cite

State-of-the-art mobile platforms, such as smartphones and tablets, are powered by heterogeneous system-on-chips (SoCs). These SoCs are composed of many processing elements, including multiple CPU core clusters (e.g., big.LITTLE cores), graphics processing units (GPUs), memory controllers and other on-chip resources. On the one hand, mobile platforms need to provide a swift response time for interactive apps and high throughput for graphics-oriented workloads; on the other hand, the power consumption must be under tight control to prevent high skin temperatures and energy consumption. Therefore, commercial systems feature a range of mechanisms for dynamic power and temperature control. However, these techniques rely on simple indicators, such as core utilization and total power consumption. System architects are typically limited to the total power consumption, since multiple resources share the same power rail. More importantly, most of the power rails are not exposed to the input/output pins. To address this challenge, this paper presents a thorough methodology to model the power consumption of major resources in heterogeneous SoCs. The proposed models utilize a wide range of performance counters to capture the workload dynamics accurately. Experimental validation on a Nexus 6P phone, powered by an octa-core Snapdragon 810 SoC, showed that the proposed models can estimate the power consumption within a 10% error margin.

show abstract

An Energy-aware Online Learning Framework for Resource Management in Heterogeneous Platforms

Cited by 26 publications

References 35 publications

AutoFL: Enabling Heterogeneity-Aware Energy Efficient Federated Learning

AutoFL: Enabling Heterogeneity-Aware Energy Efficient Federated Learning

AutoScale: Optimizing Energy Efficiency of End-to-End Edge Inference under Stochastic Variance

Per-Core Power Modeling for Heterogenous SoCs

Contact Info

Product

Resources

About