Multi-Objective Resource Scheduling for IoT Systems Using Reinforcement Learning^†

Round 1

Reviewer 1 Report

The manuscript presents a general Multi-objective Reinforcement Learning framework for the multi-objective optimization problem of IoT embedded systems. It employs the general Multi-objective Markov Decision Process formulation and introduces two novel low-compute algorithms. The framework is evaluated on single-task and dual-task Energy Harvesting Wireless Sensor Node through simulations. It is shown that the proposed framework can learn better policies at lower learning costs and successfully tradeoff between multiple objectives at runtime.

The manuscript is very well written, with an in-depth analysis of the existing solutions and a detailed description of the proposed framework and its evaluation.

 The manuscript can be accepted as is, with only several minor spelling, grammar, or typo corrections. (lines 81, 163, 211,...).

Author Response

Dear Reviewer 1,

Thank you for your review. We have checked the manuscript for typos/grammatical mistakes and corrected them.

Reviewer 2 Report

This paper proposed a general multi-objective reinforcement learning (MORL) framework for multi-objective optimization of IoT embedded systems. Instances of the single-task and dual-task energy harvesting wireless sensor nodes systems were evaluated and the experimental results showed good effectiveness of the proposed method. There are several issues that need to be addressed.  

(1) The writing logic of this manuscript needs to be strengthened. For example, the existing paper framework is confusing.

(2) This manuscript needs to be polished. There are still several obvious syntax errors, and many descriptions are redundant.

(3) While this paper proposed a multi-objective reinforcement learning framework (MORL), another state-of-the-art MORL framework, namely, MOA-DRL [1], is suggested to be compared. The difference between them needs to be emphasized.  

[1] Li, K., Zhang, T., & Wang, R. (2020). Deep reinforcement learning for multiobjective optimization. IEEE transactions on cybernetics, 51(6), 3103-3114.

Author Response

Dear Reviewer 2,

Thank you for your comments.

• We have reworded most of the text so that the flow is better.
• We have also included a discussion about the paper you mentioned in the Related Work section.
• We have checked the manuscript for typos/errors and corrected them.

Round 2

Reviewer 2 Report

I agree to accept this manuscript in the present version.