Real-time, Robust, and Reliable (R³) Machine Learning over Wireless Networks

Team: Joshua Menezes, Nihal Abdul Muneer, Ayaan Qayyum,

Madhav Subramaniyam, Hasan Ali

​

Prof. Anand Sarwate, Prof. Waheed Bajwa

L2H

PCA

Joshua ‘Machine-Learner’ Menezes

Rutgers ECE (GR)

Madhav

‘Goals and Aspirations’

Subramaniyam

Rutgers SAS (UG)

Nihal ‘Abdul’ Muneer

Rutgers ECE (GR)

Ayaan ‘AirDrop the Code’ Qayyum

Columbia EE (GR)

Hasan ‘Bayes Classifier’ Ali

Rutgers SAS (UG)

The Team

Learning to Help (L2H)

Principal Component Analysis (PCA)

• Server helps client model
• Rejector chooses model

• Reduce data dimensionality using distributed Krasulina’s method in high-rate streaming settings

Project Overview

Progress – L2H

• PTP set up for delay recording
• Switched to Monotonic Clock
• Updated testing/training code for latency
• Tested accuracy/latency at different thresholds

Progress – PCA

• Analyzed 5G cell-tower usage dataset and suitability
• Performed Krasulina’s PCA algorithm using synthetic dataset
• Started implementing central node network PCA

​

Fig. 1: Compared accuracy of most important features using different learning rates

L2H

PCA

• Ask PI for next step/direction
• Compare L2H to using just Server and just Client
• Use new Loss function

• Make code more robust
• Integrate PCA code with 5G beams dataset
• Simulate streaming data node

Next Steps

Thank you!

Any Questions?

Thank you?

Any Questions!

Click to add subtitle