Graduate School and Research Center in Digital Sciences

Team deep mixture of Eexperts for distributed power control

Zecchin, Matteo; Gesbert, David; Kountouris, Marios

SPAWC 2020, 21st IEEE International Workshop on Signal Processing Advances in Wireless Communications, 26-29 May 2020, Atlanta, GA, USA

In the context of wireless networking, it was recently shown that multiple DNNs can be jointly trained to offer a desired collaborative behaviour capable of coping with a broad range of sensing uncertainties. In particular, it was established that DNNs can be used to derive policies that are robust with respect to the information noise statistic affecting the local information (e.g. CSI in a wireless network) used by each agent (e.g. transmitter) to make its decision. While promising, a major challenge in the implementation of such method is that information noise statistics may  differ from agent to agent and, more importantly, that such statistics may not be available at the time of training or may evolve over time, making burdensome retraining necessary. This situation makes it desirable to devise a “universal” machine learning model, which can be trained once for all so as to allow for decentralized cooperation in any future feedback noise environment. With this goal in mind, we propose an architecture inspired from the well-known Mixture of Experts (MoE) model, which was previously used for non-linear regression and classification tasks in various contexts, such as computer vision and speech recognition. We consider the decentralized power control problem as an example to showcase the validity of the proposed model and to compare it against other power control algorithms. We show the ability of the so called Team-DMoE model to efficiently track time-varying statistical scenarios.

Document Bibtex

Title:Team deep mixture of Eexperts for distributed power control
Keywords:team decision theory, machine learning, distributed power control, interference channel, wireless network
Type:Conference
Language:English
City:Atlanta
Country:UNITED STATES
Date:
Department:Communication systems
Eurecom ref:6199
Copyright: © 2020 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.
Bibtex: @inproceedings{EURECOM+6199, year = {2020}, title = {{T}eam deep mixture of {E}experts for distributed power control}, author = {{Z}ecchin, {M}atteo and {G}esbert, {D}avid and {K}ountouris, {M}arios}, booktitle = {{SPAWC} 2020, 21st {IEEE} {I}nternational {W}orkshop on {S}ignal {P}rocessing {A}dvances in {W}ireless {C}ommunications, 26-29 {M}ay 2020, {A}tlanta, {GA}, {USA}}, address = {{A}tlanta, {UNITED} {STATES}}, month = {05}, url = {http://www.eurecom.fr/publication/6199} }
See also: