Learning-Based Context-Aware Resource Allocation for Edge-Computing-Empowered Industrial IoT
read more
Citations
6G Wireless Systems: A Vision, Architectural Elements, and Future Directions
Energy-Efficient Random Access for LEO Satellite-Assisted 6G Internet of Remote Things
A Secured Framework for SDN-Based Edge Computing in IoT-Enabled Healthcare System
Secure and Energy Efficient-Based E-Health Care Framework for Green Internet of Things
Machine Learning Techniques for 5G and Beyond
References
Reinforcement Learning: An Introduction
Finite-time Analysis of the Multiarmed Bandit Problem
Mobile Edge Computing: A Survey on Architecture and Computation Offloading
Stochastic Network Optimization with Application to Communication and Queueing Systems
Mobile Edge Computing: A Survey on Architecture and Computation Offloading
Related Papers (5)
Dynamic Computation Offloading for Mobile-Edge Computing With Energy Harvesting Devices
Human-level control through deep reinforcement learning
Frequently Asked Questions (14)
Q2. What have the authors stated for future works in "Learning-based context-aware resource allocation for edge computing-empowered industrial iot" ?
Their future work will focus on the online cross-layer resource optimization including local computation, rate control, channel selection, and resource allocation in the edge server under information uncertainty.
Q3. How does the proposed SEB-UCB improve throughput?
Simulation results demonstrate that the proposed SEB-UCB can improve throughput by 30% and 36% compared with UCB and random selection.
Q4. Why do the authors only consider task offloading?
Due to the limited computational capability and battery capacity of MTDs, the authors only consider the scenario of task offloading, while local13computing is ignored.
Q5. what is the energy consumption of mk in the t-th slot?
In the t-th slot, the energy consumption of mk for data transmission is the transmission power multiplied by the transmission delay, i.e.,Ek,j,t = { PTX min{Qk(t)Rk,j,t , τ}, j = 1, 2, · · · , J . 0, j = J + 1. (7)The limited battery capacity exerts a direct impact on the total energy budget of mk over T slots, which is denoted by4 Ek,max.
Q6. How many CPU cycles are available for mk in the t-th slot?
The available computational resource for mk in the t-th slot ξk,t is randomly distributed within the interval [0.9ξ̄k, 1.1ξ̄k] CPU cycles, where ξ̄k = 18× 109 CPU cycles represents the timeaverage amount of computational resource.
Q7. Why does the service reliability deficit of UCB increase after t = 700?
The service reliability deficit of EBC-MUCB increases dramatically after t = 700 due to the negligence of service reliability awareness.
Q8. What is the BER for mk in the t-th slot?
The throughput of mk in the t-th slot is given byzk,j,t = min{Qk(t), τRk,j,t}. (3)The amount of data transmitted to the edge server can beUk(t) = J+1∑ j=1 xk,j,tzk,j,t. (4)Denote the bit error rate (BER) for mk transmitting data through subchannel cj in the t-th slot as P ek,j,t.
Q9. What is the way to estimate k,j,t?
Instead of directly calculating θk,j,t in SEB-GSI, SEB-UCB estimates θk,j,t based on historical observations while simultaneously taking into account the uncertainty of estimation via confidence bound.
Q10. What is the procedure for resolving a match?
Mt transmits it to the edge server for resolving matching conflicts based on the following procedures:3) Iterative Matching: Step 1: Initialization • Initialize φ = ∅ and Ω = ∅.
Q11. What is the preference list of mk towards all the options?
Denote the preference list of mk towards all the J + 1 options as Fk, which is obtained by sorting all the Lk,j,t, j = 1, 2, · · · , J + 1, in a descending order.
Q12. What is the information required to solve P2?
the information required to solve P2 can be classified into two categories, i.e., • Local Information: information that can be possessedby mk without additional information exchange, e.g., the queue backlog Qk(t), the transmission power PTX , the total energy budget Ek,max, the computational intensity of task data λk,t, the task delay requirement dk,t, and the service reliability requirement ηk.
Q13. What is the transmission rate of subchannel sj in each slot?
The achievable transmission rate of subchannel sj in each slot follows a uniform distribution within the range [0.8R̄j , 1.2R̄j ], where R̄j represents the average transmission rate.
Q14. What is the definition of learning regret?
When ω = 1, the learning regret of the SEBCMUCB is upper bounded asR ≤ 8(J + 1) K∑ k=1 (∆θk,j̈,j̆) 3 ln(T ) +K(J + 1)∆θk,j̈,j̆+ (J + 1) K∑ k=1 +∞∑ t=1 [2t−4K+2∆θk,j̈,j̆ ] (28)Proof: See Appendix B. Based on the definition of learning regret, the cumulative throughput achieved by SEBC-MUCB can be derived as the cumulative throughput achieved by SEB-MGSI minus learning regret.