### Abstract

The ultimate product of distributed sensing is normally a model that describes the data or a set of processes for which the data is an observation. The sensor network itself affects the collected data, often due to the need to conserve deployment costs, including energy provisioning. These effects include data compression and transmission censoring in addition to the typical noise and distortion of signal transduction. This paper describes a framework for performing data and model inference that incorporates network effects into the sophisticated data/model inference techniques used in environmental and ecological field studies. Both the monitored processes and the effects of network data gathering tend to be strongly non-linear and, with uncertainty arising at different levels, hierarchical. Hence closed-form solutions for estimation are beyond reach. Hierarchical Bayesian modeling can jointly capture models and the effects of compression/censoring, and Markov chain Monte Carol (MCMC) simulation at the data center can form posterior estimates. Integral to the inference are estimates of the uncertainty of data and parameters. As the model evolves at the data center, continuously updated estimates of uncertainty can drive adaptation of source/channel coding and data transmission policies at the sensor nodes. As an example application, this paper considers strongly asymmetric data processing to minimize computational complexity and energy cost at the sensor nodes and exploit abundant data center resources. Using an example of compression via simple requantization of the data, we show that MCMC-based inference can yield good performance even at substantial compression rates.

Original language | English (US) |
---|---|

Title of host publication | Proceedings of IEEE Sensors |

Pages | 1843-1847 |

Number of pages | 5 |

DOIs | |

State | Published - 2010 |

Event | 9th IEEE Sensors Conference 2010, SENSORS 2010 - Waikoloa, HI, United States Duration: Nov 1 2010 → Nov 4 2010 |

### Other

Other | 9th IEEE Sensors Conference 2010, SENSORS 2010 |
---|---|

Country | United States |

City | Waikoloa, HI |

Period | 11/1/10 → 11/4/10 |

### Fingerprint

### ASJC Scopus subject areas

- Electrical and Electronic Engineering

### Cite this

*Proceedings of IEEE Sensors*(pp. 1843-1847). [5690013] https://doi.org/10.1109/ICSENS.2010.5690013

**Energy-efficient model inference in wireless sensing : Asymmetric data processing.** / Flikkema, Paul G.

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

*Proceedings of IEEE Sensors.*, 5690013, pp. 1843-1847, 9th IEEE Sensors Conference 2010, SENSORS 2010, Waikoloa, HI, United States, 11/1/10. https://doi.org/10.1109/ICSENS.2010.5690013

}

TY - GEN

T1 - Energy-efficient model inference in wireless sensing

T2 - Asymmetric data processing

AU - Flikkema, Paul G

PY - 2010

Y1 - 2010

N2 - The ultimate product of distributed sensing is normally a model that describes the data or a set of processes for which the data is an observation. The sensor network itself affects the collected data, often due to the need to conserve deployment costs, including energy provisioning. These effects include data compression and transmission censoring in addition to the typical noise and distortion of signal transduction. This paper describes a framework for performing data and model inference that incorporates network effects into the sophisticated data/model inference techniques used in environmental and ecological field studies. Both the monitored processes and the effects of network data gathering tend to be strongly non-linear and, with uncertainty arising at different levels, hierarchical. Hence closed-form solutions for estimation are beyond reach. Hierarchical Bayesian modeling can jointly capture models and the effects of compression/censoring, and Markov chain Monte Carol (MCMC) simulation at the data center can form posterior estimates. Integral to the inference are estimates of the uncertainty of data and parameters. As the model evolves at the data center, continuously updated estimates of uncertainty can drive adaptation of source/channel coding and data transmission policies at the sensor nodes. As an example application, this paper considers strongly asymmetric data processing to minimize computational complexity and energy cost at the sensor nodes and exploit abundant data center resources. Using an example of compression via simple requantization of the data, we show that MCMC-based inference can yield good performance even at substantial compression rates.

AB - The ultimate product of distributed sensing is normally a model that describes the data or a set of processes for which the data is an observation. The sensor network itself affects the collected data, often due to the need to conserve deployment costs, including energy provisioning. These effects include data compression and transmission censoring in addition to the typical noise and distortion of signal transduction. This paper describes a framework for performing data and model inference that incorporates network effects into the sophisticated data/model inference techniques used in environmental and ecological field studies. Both the monitored processes and the effects of network data gathering tend to be strongly non-linear and, with uncertainty arising at different levels, hierarchical. Hence closed-form solutions for estimation are beyond reach. Hierarchical Bayesian modeling can jointly capture models and the effects of compression/censoring, and Markov chain Monte Carol (MCMC) simulation at the data center can form posterior estimates. Integral to the inference are estimates of the uncertainty of data and parameters. As the model evolves at the data center, continuously updated estimates of uncertainty can drive adaptation of source/channel coding and data transmission policies at the sensor nodes. As an example application, this paper considers strongly asymmetric data processing to minimize computational complexity and energy cost at the sensor nodes and exploit abundant data center resources. Using an example of compression via simple requantization of the data, we show that MCMC-based inference can yield good performance even at substantial compression rates.

UR - http://www.scopus.com/inward/record.url?scp=79951901871&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=79951901871&partnerID=8YFLogxK

U2 - 10.1109/ICSENS.2010.5690013

DO - 10.1109/ICSENS.2010.5690013

M3 - Conference contribution

AN - SCOPUS:79951901871

SN - 9781424481682

SP - 1843

EP - 1847

BT - Proceedings of IEEE Sensors

ER -