Adapting Wireless Mesh Network Configuration from Simulation to Reality via Deep Learning
summary
A system trains a teacher-student neural network using large-scale simulator outputs plus a few field measurements to close the simulation-to-reality gap with domain adaptation. Experiments use four simulators, 15 wireless devices, 88 configurable parameter settings and about 6,600 traces; the method raises correct-configuration prediction from 30.10% to 70.24% by transferring knowledge from cheap simulated data to costly physical tests, producing a practical model that substantially improves real-world WMN configuration.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 0:40Wireless Mesh Network
- 2:16Challenges
- 5:06Machine Learning
- 5:51Testing Accuracy
- 8:37Domain Adaptation
- 10:07Experimental Results
- 11:16Conclusion
related talks
21mMirrorNet: High-fidelity and Scalable Network Emulation for Software-defined WAN
06mTeaching RF to Sense without RF Training Measurements
32mMan or Machine?: Developing a "Turing Test" for Radio Intelligence
17mADR-X: ANN-Assisted Wireless Link Rate Adaptation for Compute-Constrained Embedded...
06mDeepRange: Acoustic Ranging via Deep Learning
57mMachine Learning in the Age of IoT
51mTensorFlow Wide & Deep: Advanced Classification the easy way
17mSLNet: A Spectrogram Learning Neural Network for Deep Wireless Sensing