1 of 8

Resilient Edge-Cloud Autonomous Learning with Timely inferences - Week 7

James Chang, Tanushree Mehta, Shreya Venugopal, Lakshya Gour, Haider Abdelrahman

​

​

2 of 8

Project Overview

The Problem:

  • Models are getting more complex
  • Running these models on less powerful (mobile) devices is becoming increasingly difficult (latency issues)

​

Low latency in MEC systems is essential for systems that

need to process large amounts of data in a short time

​

  • A viable solution to this problem is edge computing

​

3 of 8

Project Overview

Objective:

  • To use ORBIT in to design/run experiments to analyze the training/prediction of ML models across multiple edge devices.

​

  • Develop a latency profiling framework for MEC-assisted machine learning using split computing and early exiting.

​

  • Analyze these models for latency and accuracy tradeoff analysis, along with measuring network delays.

4 of 8

This Week

  • Explored different research questions with the data collected
  • Limited CPU power in terminal to imitate mobile devices
  • Created a script to Implement different threshold values based on confidence for sending the data to the edge and server for inference
  • Generated graphs for threshold vs latency and accuracy

5 of 8

6 of 8

7 of 8

8 of 8

Next Week

  • Latency vs. Accuracy; Number of Help Requests vs. Latency
  • Test and analyze different types of thresholds for latency and accuracy
  • Try other neural networks and see how that can affect the result
  • Do out of distribution analysis - confusion matrix/create a new class
  • Implement plots using a model with 99% accuracy rather than 83% accuracy
    • latency vs accuracy
  • Replace edge with oracle - already knows the answer