Research

Computer Sciences and Information Technology

Title :

Internet-of-Things Network Scheduling in a Reinforcement Learning-aided Mobile Edge Computing System

Area of research :

Computer Sciences and Information Technology

Principal Investigator :

Dr. Arghyadip Roy, Indian Institute Of Technology (IIT) Guwahati, Assam

Timeline Start Year :

2022

Timeline End Year :

2024

Contact info :

Equipments :

Details

Executive Summary :

The project aims to use the Reinforcement Learning (RL) paradigm to address issues in Fifth Generation (5G)/Sixth Generation (6G) wireless communication systems. RL algorithms are designed to work without knowledge of system parameters and converge to optimal solutions in the long run, which is crucial in next-generation networks like the Internet of Things (IoT) and Mobile Edge Computing (MEC). Tasks can be executed in-device or offloaded to MEC/cloud servers, with in-device processing being good in terms of power consumption but may perform poorly in the face of deadlines. Traditional RL methods can operate in a trial-and-error manner, learning the optimal policy that balances power consumption and task computation delay. However, they fail if the training environment differs from the actual one due to the lack of robustness properties in state-of-the-art RL algorithms. This necessitates the investigation of robust algorithms for task scheduling, which train the system for a range of system parameters rather than specific values. Rubberness is desirable for RL algorithms, but it comes at the cost of conservativeness, leading to performance degradation compared to traditional RL schemes when the training environment is similar to the actual one. The project proposes RL schemes that balance conservativeness and robustness by providing task scheduling policies that optimize against average perturbation instead of worst-case perturbation. The proposed schemes will be evaluated in a real-world IoT simulation environment.

Total Budget (INR):

19,48,470

Organizations involved