Computation Offloading for Machine Learning in Industrial Environments

Guo, Mian; Mukherjee, Mithun; Liang, Gen; Zhang, Jinyou

doi:10.1109/iecon43393.2020.9255001

Cited by 8 publications

(7 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…where E(i) follows (6), N opt is the number of rounds required to repeat until achieving the model accuracy, which could be approximated by N opt = α/E[M succ (i)] according to [30], where α is a factor parameter related to training accuracy. Similarly, the learning delay of a learning process is defined by…”

Section: Energy Consumption and Learning Delaymentioning

confidence: 99%

See 1 more Smart Citation

Dynamic Federated Learning for GMEC With Time-Varying Wireless Link

et al. 2021

View full text Add to dashboard Cite

Smart grid applications, such as predicting energy consumption, grid user behavior analysis and predicting energy theft, etc., are data-driven applications that require machine learning with a wealth of data generated from Internet of Things (IoT) based metering devices. However, traditional methods of uploading this huge data to the remote cloud for data analytics may be low efficient due to the non-negligible network transmission delay. By deploying a number of computing-enabled devices at the network edge, edge computing supports the implementation of machine learning close to the power grid environment. Considering the limited computing resources of edge devices and non-independent and identical (non-IID) data source, federated learning is a feasible edge computing based machine learning model. In federated learning, distributed mobile clients and a federated server collaborate to perform machine learning. Generally, the more clients to join the federated learning, the faster to obtain learning convergence and the higher resource utility. However, the communications between clients and the server in training rounds of federated learning may fail due to time-varying link reliability properties in a wireless network of smart grid, which not only slows down the model convergence rate but also wastes resources, such as energy consumption for invalid local training. This paper studies a dynamic federated learning problem in a power grid mobile edge computing (GMEC) environment, considering the high dynamic of link reliability. We design a delay deadline constrained federated learning framework to avoid extremely long training delay, and then formulate a dynamic client selection problem for computing utility maximization in such learning framework. Two online client selection algorithms, including climax greedy and uti-positive guarantee, are proposed to address the problem. The theoretical analysis and simulation results are conducted to illustrate the efficiency of the proposal. INDEX TERMS Mobile edge computing, machine learning, federated learning, smart grid, link reliability.

show abstract

Section: Energy Consumption and Learning Delaymentioning

confidence: 99%

“…This is because, under cli-max greedy, the number of clients successfully finish a round of training is the highest. Then, according to [26], [30], the more number of clients joining a training, the fast to converge. Therefore, the number of rounds given by cli-max greedy is the lowest.…”

Section: Energy Consumption and Learning Delaymentioning

confidence: 99%

Dynamic Federated Learning for GMEC With Time-Varying Wireless Link

et al. 2021

View full text Add to dashboard Cite

show abstract

“…In the first category, the convergence time in FL with optimal bandwidth allocation and scheduling strategies was studied in [22]. Similarly, the delay in FL with optimal scheduling strategies was studied in [23]. The minimized delay in FL with optimal bandwidth allocation, scheduling strategies, and transmit power was studied in [24].…”

Section: Introductionmentioning

confidence: 99%

“…The minimized delay in FL with optimal bandwidth allocation, scheduling strategies, and transmit power was studied in [24]. However, the optimal solutions were only evaluated numerically in [22], [23], and there are no closed-form expressions to facilitate the analyses. In addition, the energy cost was not involved in the above works.…”

Section: Introductionmentioning

confidence: 99%

Revisiting Communication-Efficient Federated Learning with Balanced Global and Local Updates

Yan¹,

Liu²,

Zhang³

et al. 2022

Preprint

View full text Add to dashboard Cite

In federated learning (FL), a number of devices train their local models and upload the corresponding parameters or gradients to the base station (BS) to update the global model while protecting their data privacy. However, due to the limited computation and communication resources, the number of local trainings (a.k.a. local update) and that of aggregations (a.k.a. global update) need to be carefully chosen. In this paper, we investigate and analyze the optimal trade-off between the number of local trainings and that of global aggregations to speed up the convergence and enhance the prediction accuracy over the existing works. Our goal is to minimize the global loss function under both the delay and the energy consumption constraints. In order to make the optimization problem tractable, we derive a new and tight upper bound on the loss function, which allows us to obtain closed-form expressions for the number of local trainings and that of global aggregations. Simulation results show that our proposed scheme can achieve a better performance in terms of the prediction accuracy, and converge much faster than the baseline schemes.

show abstract

“…Mobile edge computing (MEC) is a promising paradigm for delay-sensitive machine learning [2], [3]. By deploying machine learning servers at the network edge closer to data sources (e.g., IoT devices), the network transmission delay for uploading data from data sources to a remote server could be explicitly reduced.…”

Section: Introductionmentioning

confidence: 99%

Collaborative Edge Learning in MIMO-NOMA Uplink Transmission Environment

Guo

Shan

Mukherjee

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

Multiple-input multiple-output non-orthogonal multiple access (MIMO-NOMA) cellular network is promising for supporting massive connectivity. This paper exploits low-latency machine learning in the MIMO-NOMA uplink transmission environment, where a substantial amount of data must be uploaded from multiple data sources to a one-hop away edge server for machine learning. A delay-aware edge learning framework with the collaboration of data sources, the edge server, and the base station, referred to as DACEL, is proposed. Based on the delay analysis of DACEL, a NOMA channel allocation algorithm is further designed to minimize the learning delay. The simulation results show that the proposed algorithm outperforms the baseline schemes in terms of learning delay reduction.

show abstract

Computation Offloading for Machine Learning in Industrial Environments

Cited by 8 publications

References 9 publications

Dynamic Federated Learning for GMEC With Time-Varying Wireless Link

Dynamic Federated Learning for GMEC With Time-Varying Wireless Link

Revisiting Communication-Efficient Federated Learning with Balanced Global and Local Updates

Collaborative Edge Learning in MIMO-NOMA Uplink Transmission Environment

Contact Info

Product

Resources

About