Agile Reinforcement Learning for Real-Time Task Scheduling in Edge Computing

Akramul Azim; Amin Avan; Qusay Mahmoud

REVIEW 3 major objections 3 minor 22 references

Agile Reinforcement Learning for Real-Time Task Scheduling in Edge Computing

T0 review · 3 major / 3 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read By replacing random exploration with an earliest-deadline heuristic and masking irrelevant actions, a deep Q-network scheduler for edge computing converges faster and schedules more soft real-time tasks on time than vanilla RL or standard…

desk verdict A sensible incremental idea on DQN for edge scheduling, but the headline speedup is confounded by the convergence criterion. read the letter →

arxiv 2506.08850 v1 pith:RRRI57ZF submitted 2025-06-10 cs.LG

Amin Avan , Akramul Azim , Qusay Mahmoud This is my paper

classification cs.LG

keywords reinforcementlearningactionmaskinginformedexplorationedgecomputingtaskschedulingsoftreal-timedeepQ-networkhit-ratio

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes aRL, a reinforcement-learning scheduler for soft real-time tasks offloaded onto heterogeneous edge servers. Its central claim is that a deep Q-network can learn to produce deadline-satisfying schedules quickly if two mechanisms curb the action space: informed exploration that always picks the earliest-deadline unassigned task, and action masking that forbids reassigning a task and limits each episode to one action per task. In an EdgeSimPy video-surveillance scenario, aRL reports a higher hit-ratio than EDF, BestFit, and a vanilla DQN baseline, while converging 46% faster and using 65.5% less RAM and 60.6% less power than the vanilla agent. The paper's intended significance is that slow training, the main obstacle to RL-based real-time schedulers, can be removed by borrowing structure from classical scheduling heuristics.

What carries the argument

The named machinery is aRL's dual mechanism. Action masking uses a decision matrix $G$ whose entries mark whether task $\tau_j$ has been assigned to server $s_k$; the single-assignment constraint restricts every action to the set of unassigned tasks $\Gamma_{ua}$, and the action bound $|A| \le |T|$ caps each episode at one action per task, so rollouts are short and every episode ends within a predictable horizon. Informed exploration replaces uniform random action selection with a modified Earliest Deadline First rule: during exploration the agent picks the unassigned task with the earliest absolute deadline and assigns it to an available server subject to capacity and utilization thresholds. The reward function accumulates positive rewards for satisfying the deadline, capacity, utilization, and single-assignment constraints and negative rewards for violations, so the DQN's value estimates are shaped by exactly the objectives the scheduler is meant to optimize.

What would settle it

Run aRL with uniform random exploration, identical action masking and reward function, and measure episodes to the same 98% threshold; if the runtime gap over vRL disappears, the claimed 46% saving comes from EDF-guided exploration rather than from faster learning.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that random exploration is what makes RL schedulers slow, and that replacing it with an EDF-guided exploration policy while masking irrelevant actions concentrates learning on decisions that matter. The policy network still picks actions once the exploration threshold is passed, but the agent's exploration behavior is already deadline-aware. In the evaluated workload, aRL attains the highest hit-ratio among EDF, BestFit, and vRL, exceeding the heuristics by 24% and vRL by 1.9%, while reaching convergence 46% faster than vRL. The authors present these results as evidence that informed exploration and action masking improve the predictability of the agent, which in turn shortens the learning phase and makes RL practical for soft real-time edge scheduling.

Load-bearing premise

The load-bearing premise is that crossing the 98%-for-100-episodes hit-ratio threshold measures the quality of the learned policy, even though the agent explores by following EDF, so the threshold could be met largely by the exploration heuristic alone.

Editorial extensions

If this is right

If aRL's claim holds, deadline-guided exploration plus action masking offers a general recipe for making DQN schedulers trainable in edge environments where the state-action space is large and timing constraints are tight.
The measured runtime reduction to convergence implies that RL-based schedulers can be retrained or adapted when the edge environment changes, which is the setting that usually makes RL impractical for real-time workloads.
The higher hit-ratio over both EDF and BestFit indicates that a learned value function can improve on the heuristic that guides exploration rather than merely reproducing it.
The reductions in RAM and power consumption during learning mean the training process itself is cheaper, which matters on resource-constrained edge servers.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper does not isolate the contribution of informed exploration from that of action masking; a natural extension is ablating each mechanism separately to see which one drives the runtime saving.
Because the exploration policy is itself EDF, the 98% convergence criterion may be partially satisfied by the heuristic rather than by the learned component; a fair test would compare against a random-exploration agent with identical masking to isolate the learning effect.
The same template could apply to other scheduling objectives by swapping EDF for a different prior, such as least-loaded server, best-fit resource match, or energy-aware ordering, suggesting the result may transfer beyond deadline-hit-ratio metrics.
Action masking shortens rollouts and stabilizes the number of actions per episode, which should reduce variance in value estimates; this could matter even more at the larger task scales the paper does not test.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

First, the useful part: this is a clear, well-structured paper on a real problem. The MDP formulation includes hit-ratio, deadlines, and resource constraints, which is more real-time aware than most edge-scheduling RL papers. The vRL baseline is a reasonable control that shares the DQN architecture, and the authors promise code. The combination of action masking with a constructive EDF-based exploration is a sensible engineering idea for cutting learning time.

The problem: the headline number—46% faster than vRL—is not isolated to learning. Convergence is declared when hit-ratio exceeds 98% for 100 episodes, and aRL explores using a feasible-server EDF heuristic. If that heuristic alone clears 98%, the timer stops before the Q-network has learned anything useful. The paper gives no ablation with exploration-only (no Q-learning) or with masking-only, and no random-exploration-plus-masking condition. So the speedup could come entirely from the heuristic exploration, not from the RL loop. The fact that a plain EDF baseline gets only ~76% doesn't settle it, because aRL's exploration is a modified EDF that selects feasible servers.

Also, the reported averages over 31 runs have no confidence intervals or significance tests, Eq. (3.5) has a unit mismatch (cycles times RAM over frequency times cores), and the while condition in Algorithm 1 is typo'd (the OR should likely be an AND). These are fixable.

Bottom line: the paper is worth a serious referee—the idea is plausible and the review will help the authors design the right ablations—but the empirical claims should not be taken at face value. I'd ask for major revision before acceptance.

Referee Report

3 major / 3 minor

Summary. The paper proposes Agile Reinforcement Learning (aRL), a DQN-based scheduler for soft real-time tasks in edge computing. The method augments DQN with two mechanisms: action masking, which restricts assignments to unassigned tasks and feasible servers, and informed exploration, which replaces random exploration with an EDF-guided heuristic that selects the earliest-deadline unassigned task. The authors claim that this combination yields faster convergence and higher hit-ratio than vanilla DQN (vRL), EDF, and BestFit, with reported reductions of 46% in runtime, 65.5% in RAM usage, and 60.6% in power consumption relative to vRL. The evaluation is carried out in the EdgeSimPy simulator on a video surveillance scenario with 52 edge users and 4 heterogeneous edge servers, and the paper provides a link to source code.

Significance. If the claims are robust, the paper makes a useful practical contribution: it combines a simple, interpretable exploration heuristic with action masking to reduce the training time of a DQN scheduler, which is a known bottleneck for RL-based edge task scheduling. The problem formulation explicitly includes real-time constraints (deadline, response time, hit-ratio), which is appropriate for soft real-time applications. The availability of source code and 31-run logs is a strength and supports reproducibility. However, the significance is currently limited by the evaluation design: the convergence-based runtime metric is confounded with the exploration heuristic, and the absence of error bars or significance tests weakens the quantitative claims. The central idea is worth pursuing, but the evidence as presented is not yet convincing.

major comments (3)

[Section 5 (Convergence criterion) and Section 4 (Informed Exploration; Algorithm 1, lines 4–8)] The headline claim that aRL converges 46% faster than vRL is confounded by the definition of convergence. Section 5 states that convergence is achieved when the hit-ratio of the generated schedule exceeds 98% for 100 consecutive episodes. In Algorithm 1, the exploration branch returns mapAction(GetEDFtask()), i.e., it schedules the earliest-deadline unassigned task to an available server satisfying resource constraints. This exploration policy is itself a constructive deadline-aware heuristic, so it may by itself produce schedules that meet the 98% threshold. Consequently, the reported runtime reduction could reflect the contribution of the EDF-guided exploration and action masking rather than the learned Q-network. The paper does not include an ablation with EDF exploration plus action masking but without Q-learning updates, nor an ablation with random exploration plus masking. Without such controls, the causal attribution of the speed-up to the learning component is not established. I recommend adding these ablations and/or reporting hit-ratio as a function of learning episodes separately from the exploration heuristic's standalone performance.
[Section 3.1, Eq. (3.5)] Equation (3.5) defines execution time as eτj = (cτj × mτj)/(fsk × Nsk), where cτj is required processor cycles and mτj is required RAM. The preceding sentence states that execution time is obtained by dividing the total required number of processor cycles by the processing capacity, which would yield cτj/(fsk × Nsk). The product cτj × mτj is dimensionally inconsistent and appears to be a typo. This is load-bearing because eτj enters the response-time formula (3.4), the deadline constraint C1.1, and the reward function, and hence directly affects the measured hit-ratio. Please correct the formula or provide a justification for the multiplication.
[Section 5 (Experimental results, Figures 3–5)] The experiments are repeated 31 times, but the paper reports only average hit-ratio, runtime, RAM usage, and power consumption, with no confidence intervals, standard deviations, or significance tests. The central quantitative claims (46% runtime reduction, 65.5% RAM reduction, 60.6% power reduction) are therefore not supported with evidence that these differences are stable or statistically meaningful. I request error bars or variance information for the RL-based methods, and at least a statement of the test used to compare aRL and vRL.

minor comments (3)

[Section 4, Algorithm 1] The loop condition 'while (t ≤ |T |) or (hitT asks== |T |)' appears to use OR where the text in Section 4 says the episode terminates when either the number of actions equals the number of tasks or all tasks are assigned. With OR, the loop continues even after all tasks are assigned, so the episode is not actually bounded by |T| actions. Also, the exploration branch 'Return mapAction(GetEDFtask())' does not specify how the edge server is chosen; please clarify.
[Section 3.1, Eq. (3.6) and Eq. (3.7)] The utilization constraint is written as (UPsk + upτj) < 1, but the definition of upτj as eτj/dτj is not standard for a multi-server system with heterogeneous servers and may assume a single-server capacity of 1.0. Please clarify how utilization is normalized across different server speeds.
[General] There are several notation inconsistencies: Eq. (3.15) minimizes f2(Θ) while Eq. (3.14) defines f2(Φ); Algorithm 1 line 18 has malformed conditions; and the abstract contains a typo ('andlarge-scaleproblems'). Please proofread.

Circularity Check

1 steps flagged · score 5.0 of 10

Headline runtime/convergence claim is confounded: convergence is defined by the same hit-ratio that aRL's EDF-based exploration can itself achieve.

self definitional [Section 5 (runtime/convergence definition); Section 4 'aRL - Informed Exploration' and Algorithm 1, lines 4-8.]
"the runtime for the RL-based task scheduling algorithm is defined as the total time required to complete its learning phase and achieve convergence ... Convergence is considered to be achieved if the hit-ratio of the generated task schedule exceeds 98% for 100 consecutive episodes and remains stable above this threshold [22]. ... In exploration mode, aRL explores the action-state space using a modified version of the Earliest Deadline First (EDF) algorithm, adapted for EC."

The paper measures RL runtime as time until the generated schedule's hit-ratio exceeds 98% for 100 episodes. The exploration branch of aRL (Algorithm 1, lines 4-8) returns mapAction(GetEDFtask()), i.e., it schedules the earliest-deadline unassigned task to a feasible server; this is a constructive deadline-aware heuristic, not random exploration. Because the stopping criterion is defined by hit-ratio, and EDF-based exploration directly targets deadline satisfaction, the learning timer can stop before the Q-network has contributed substantial policy improvement. Hence the reported 46% runtime reduction over vRL (and the ensuing RAM/power reductions) may quantify the substitution of EDF-informed exploration for random exploration rather than faster convergence of the learned policy.

full rationale

The paper's central derivation is not mathematically circular: aRL is an algorithm, and the results against baselines are empirical. However, the headline 'converges faster' claim is evaluated with a metric that is defined in terms of the same objective (hit-ratio) that its EDF-based exploration heuristic directly optimizes. Under this definition, the exploration policy itself can satisfy the convergence criterion, making the runtime comparison confounded. The paper is otherwise self-contained; self-citations ([1], [7], [20]) are not load-bearing for the main claim. Because the learned component may still contribute after convergence, this is a partial evaluative circularity, not an equivalence-by-definition of the whole derivation.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim depends on domain assumptions about the timing model and simulator, plus a convergence criterion that may be partly satisfied by the EDF exploration heuristic. No new physical entities are introduced. Key free parameters (epsilon, rewards, utilization threshold, DQN hyperparameters) are not reported.

free parameters (5)

epsilon_threshold = not reported
Algorithm 1 line 4 computes an epsilon threshold but the value and decay schedule are not given, so the exploration-exploitation trade-off is unspecified.
positiveReward and negativeReward = not reported
Reward magnitudes appear in Algorithm 1 lines 17-24 but values are never stated; they can strongly affect DQN convergence.
utilization threshold Uth = not reported
Eq. (3.17) imposes Uth on each resource; the paper does not state the value used in experiments.
convergence threshold (98% hit-ratio for 100 episodes) = 98% / 100 episodes
Section 5 uses this to define runtime for RL methods; the headline 46% speedup is measured against this criterion.
DQN hyperparameters (learning rate, batch size, replay buffer, hidden sizes) = not reported
Only a three-layer feedforward network is specified; without these, reproduction requires guessing.

assumptions (5)

standard math The task scheduling problem is a finite-horizon MDP and DQN will converge to a good policy with enough episodes.
Section 4 formulates the problem as an MDP with actions a=[tau_j, s_k] and uses a DQN; this assumes the usual RL convergence preconditions.
domain assumption Execution time on a server is accurately modeled by e = (c_j * m_j) / (f_sk * N_sk) and response time is the sum in Eq. (3.4).
Eq. (3.5) mixes required processor cycles and required RAM into a single product with no unit reconciliation, and no hardware validation is provided.
domain assumption EdgeSimPy faithfully represents edge computing timing, power, and memory behavior for comparative evaluation.
Section 5 uses EdgeSimPy as the sole evaluation platform; no real testbed or calibration is reported.
ad hoc to paper Convergence to a hit-ratio above 98% for 100 consecutive episodes is a meaningful measure of learning completion.
Section 5 defines convergence this way; because exploration uses EDF, this threshold may be met by the exploration heuristic rather than by a learned policy.
ad hoc to paper Restricting each task to one assignment and bounding episode length to the number of tasks does not sacrifice schedule quality.
Section 4 introduces the action bound and single-assignment constraint as mechanisms to shrink the action space, without proof that optimal schedules remain representable.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Agile Reinforcement Learning for Real-Time Task Scheduling in Edge Computing." pith.science (2026). https://pith.science/paper/RRRI57ZF

@misc{pith2026250608850,
  author       = {Pith},
  title        = {Pith review of: Agile Reinforcement Learning for Real-Time Task Scheduling in Edge Computing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RRRI57ZF}},
  note         = {Machine review of arXiv:2506.08850}
}

read the original abstract

Soft real-time applications are becoming increasingly complex, posing significant challenges for scheduling offloaded tasks in edge computing environments while meeting task timing constraints. Moreover, the exponential growth of the search space, presence of multiple objectives and parameters, and highly dynamic nature of edge computing environments further exacerbate the complexity of task scheduling. As a result, schedulers based on heuristic and metaheuristic algorithms frequently encounter difficulties in generating optimal or near-optimal task schedules due to their constrained ability to adapt to the dynamic conditions and complex environmental characteristics of edge computing. Accordingly, reinforcement learning algorithms have been incorporated into schedulers to address the complexity and dynamic conditions inherent in task scheduling in edge computing. However, a significant limitation of reinforcement learning algorithms is the prolonged learning time required to adapt to new environments and to address medium- and large-scale problems. This challenge arises from the extensive global action space and frequent random exploration of irrelevant actions. Therefore, this study proposes Agile Reinforcement learning (aRL), in which the RL-agent performs informed exploration and executes only relevant actions. Consequently, the predictability of the RL-agent is enhanced, leading to rapid adaptation and convergence, which positions aRL as a suitable candidate for scheduling the tasks of soft real-time applications in edge computing. The experiments demonstrate that the combination of informed exploration and action-masking methods enables aRL to achieve a higher hit-ratio and converge faster than the baseline approaches.

Figures

Figures reproduced from arXiv: 2506.08850 by the authors.

**Figure 1.** Overview of the edge computing (EC) system model. In this study, an EC environment is composed of multiple base stations (BSs), each providing wireless connectivity to nearby edge users. These BSs are interconnected through wired links and network switches, forming a partial-mesh topology, as shown in [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗

**Figure 2.** Overview of the edge computing (EC) architecture. services, respectively. The development of ML models encompasses re-training, hyperparameter optimization, and fine-tuning. For each edge user, the service workload comprises tasks represented by τj ∈ Γδi . Each task is uniquely defined by a set of parameters, which include the arrival time (aj ), period (pj ), absolute deadline (dj ), required processor cycles (cj … view at source ↗

**Figure 3.** Hit-ratio versus runtime analysis [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Hit-ratio versus RAM usage analysis [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]

**Figure 5.** Figure 5: Hit-ratio versus power consumption analysis. 6. Conclusion Complex soft real-time applications can be executed by edge users with limited resources using edge computing. However, scheduling offloaded tasks across edge servers in EC is challenging due to heterogeneous e…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

[1]

A Robust Scheduling Algorithm for Overload- Tolerant Real-Time Systems

A. Avan, A. Azim, and Q. H. Mahmoud. “A Robust Scheduling Algorithm for Overload- Tolerant Real-Time Systems”. In:2023 IEEE 26th International Symposium on Real-Time Distributed Computing (ISORC) . 2023, pp. 1–10.doi: 10.1109/ISORC58943.2023.00013

work page arXiv 2023
[2]

Soft real-time scheduling

J. P. Erickson and J. H. Anderson. “Soft real-time scheduling”. In:Handbook of Real-Time Computing. Springer, 2022, pp. 233–267

work page 2022
[3]

Scheduling IoT applications in edge and fog computing environments: a taxonomy and future directions

M. Goudarzi, M. Palaniswami, and R. Buyya. “Scheduling IoT applications in edge and fog computing environments: a taxonomy and future directions”. In:ACM Computing Surveys 55.7 (2022), pp. 1–41

work page 2022
[4]

TF-DDRL: ATransformer-enhanced DistributedDRL Technique for Scheduling IoT Applications in Edge and Cloud Computing Environments

Z. Wang, M.Goudarzi, andR. Buyya. “TF-DDRL: ATransformer-enhanced DistributedDRL Technique for Scheduling IoT Applications in Edge and Cloud Computing Environments”. In: IEEE Transactions on Services Computing (2025). 12

work page 2025
[5]

Resource scheduling in edge computing: A survey

Q. Luo, S. Hu, C. Li, G. Li, and W. Shi. “Resource scheduling in edge computing: A survey”. In: IEEE Communications Surveys & Tutorials 23.4 (2021), pp. 2131–2165

work page 2021
[6]

A comprehensive survey on reinforcement- learning-based computation offloading techniques in edge computing systems

D. Hortelano, I. de Miguel, R. J. D. Barroso, J. C. Aguado, N. Merayo, L. Ruiz, A. Asensio, X. Masip-Bruin, P. Fernández, R. M. Lorenzo, et al. “A comprehensive survey on reinforcement- learning-based computation offloading techniques in edge computing systems”. In:Journal of Network and Computer Applications 216 (2023), p. 103669

work page 2023
[7]

A state-of-the-art review of task scheduling for edge computing: A delay-sensitive application perspective

A. Avan, A. Azim, and Q. H. Mahmoud. “A state-of-the-art review of task scheduling for edge computing: A delay-sensitive application perspective”. In:Electronics 12.12 (2023), p. 2599

work page 2023
[8]

Deep Reinforcement Learning for Online Computation Offloading in Wireless Powered Mobile-Edge Computing Networks

L. Huang, S. Bi, and Y.-J. A. Zhang. “Deep Reinforcement Learning for Online Computation Offloading in Wireless Powered Mobile-Edge Computing Networks”. In:IEEE Transactions on Mobile Computing 19.11 (2020), pp. 2581–2593.doi: 10.1109/TMC.2019.2928811

work page arXiv 2020

Show all 22 references

[9]

Age-Based Scheduling for Mobile Edge Computing: A Deep Reinforcement Learning Approach

X. He, C. You, and T. Q. Quek. “Age-Based Scheduling for Mobile Edge Computing: A Deep Reinforcement Learning Approach”. In:IEEE Transactions on Mobile Computing (2024)

2024
[10]

Decentralized Scheduling for Concurrent Tasks in Mobile Edge Computing via Deep Reinforcement Learning

Y. Fan, J. Ge, S. Zhang, J. Wu, and B. Luo. “Decentralized Scheduling for Concurrent Tasks in Mobile Edge Computing via Deep Reinforcement Learning”. In:IEEE Transactions on Mobile Computing (2023)

2023
[11]

Deep Reinforcement Learn- ing Based Distributed Computation Offloading in Vehicular Edge Computing Networks

L. Geng, H. Zhao, J. Wang, A. Kaushik, S. Yuan, and W. Feng. “Deep Reinforcement Learn- ing Based Distributed Computation Offloading in Vehicular Edge Computing Networks”. In: IEEE Internet of Things Journal (2023)

2023
[12]

GA-DRL: Graph Neural Network-Augmented Deep Reinforcement Learning for DAG Task Scheduling over Dynamic Vehicular Clouds

Z. Liu, L. Huang, Z. Gao, M. Luo, S. Hosseinalipour, and H. Dai. “GA-DRL: Graph Neural Network-Augmented Deep Reinforcement Learning for DAG Task Scheduling over Dynamic Vehicular Clouds”. In:IEEE Transactions on Network and Service Management (2024)

2024
[13]

Asynchronous Deep Reinforcement Learning for Collaborative Task Computing and On-Demand Resource Allocation in Vehic- ular Edge Computing

L. Liu, J. Feng, X. Mu, Q. Pei, D. Lan, and M. Xiao. “Asynchronous Deep Reinforcement Learning for Collaborative Task Computing and On-Demand Resource Allocation in Vehic- ular Edge Computing”. In:IEEE Transactions on Intelligent Transportation Systems (2023)

2023
[14]

Deep reinforcement learning based ap- proach for online service placement and computation resource allocation in edge computing

T. Liu, S. Ni, X. Li, Y. Zhu, L. Kong, and Y. Yang. “Deep reinforcement learning based ap- proach for online service placement and computation resource allocation in edge computing”. In: IEEE Transactions on Mobile Computing (2022)

2022
[15]

Urbanenqosplace: A deep reinforcement learning model for service placement of real-time smart city iot applications

M. Bansal, I. Chana, and S. Clarke. “Urbanenqosplace: A deep reinforcement learning model for service placement of real-time smart city iot applications”. In: IEEE Transactions on Services Computing 16.4 (2022), pp. 3043–3060

2022
[16]

Deep reinforcement learning-based online re- source management for uav-assisted edge computing with dual connectivity

L. T. Hoang, C. T. Nguyen, and A. T. Pham. “Deep reinforcement learning-based online re- source management for uav-assisted edge computing with dual connectivity”. In:IEEE/ACM Transactions on Networking 31.6 (2023), pp. 2761–2776

2023
[17]

MESON: A mobility-aware dependent task offloading scheme for urban vehicular edge com- puting

L. Zhao, E. Zhang, S. Wan, A. Hawbani, A. Y. Al-Dubai, G. Min, and A. Y. Zomaya. “MESON: A mobility-aware dependent task offloading scheme for urban vehicular edge com- puting”. In: IEEE Transactions on Mobile Computing 23.5 (2023), pp. 4259–4272

2023
[18]

Deep reinforcement learning-based task assign- ment for cooperative mobile edge computing

L.-T. Hsieh, H. Liu, Y. Guo, and R. Gazda. “Deep reinforcement learning-based task assign- ment for cooperative mobile edge computing”. In:IEEE Transactions on Mobile Computing 23.4 (2023), pp. 3156–3171

2023
[19]

Smart Resource Allocation for Mobile Edge Comput- ing: A Deep Reinforcement Learning Approach

J. Wang, L. Zhao, J. Liu, and N. Kato. “Smart Resource Allocation for Mobile Edge Comput- ing: A Deep Reinforcement Learning Approach”. In:IEEE Transactions on Emerging Topics in Computing 9.3 (2021), pp. 1529–1541.doi: 10.1109/TETC.2019.2902661

2021
[20]

A Task Scheduler for Mobile Edge Computing Using Priority-based Reinforcement Learning

A. Avan, F. Kheiri, Q. H. Mahmoud, A. Azim, M. Makrehchi, and S. Rahnamayan. “A Task Scheduler for Mobile Edge Computing Using Priority-based Reinforcement Learning”. In: 2023 IEEE Symposium Series on Computational Intelligence (SSCI) . IEEE. 2023, pp. 539– 546

2023
[21]

EdgeSimPy: Python-Based Modeling and Sim- ulation of Edge Computing Resource Management Policies

P. S. Souza, T. Ferreto, and R. N. Calheiros. “EdgeSimPy: Python-Based Modeling and Sim- ulation of Edge Computing Resource Management Policies”. In:Future Generation Computer Systems 148 (2023), pp. 446–459. issn: 0167-739X. doi: https : / / doi . org / 10 . 1016 / j . future...

2023
[22]

Convergent policy optimization for safe reinforce- ment learning

M. Yu, Z. Yang, M. Kolar, and Z. Wang. “Convergent policy optimization for safe reinforce- ment learning”. In:Advances in Neural Information Processing Systems 32 (2019)

2019

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

A Robust Scheduling Algorithm for Overload- Tolerant Real-Time Systems

A. Avan, A. Azim, and Q. H. Mahmoud. “A Robust Scheduling Algorithm for Overload- Tolerant Real-Time Systems”. In:2023 IEEE 26th International Symposium on Real-Time Distributed Computing (ISORC) . 2023, pp. 1–10.doi: 10.1109/ISORC58943.2023.00013

work page arXiv 2023

[2] [2]

Soft real-time scheduling

J. P. Erickson and J. H. Anderson. “Soft real-time scheduling”. In:Handbook of Real-Time Computing. Springer, 2022, pp. 233–267

work page 2022

[3] [3]

Scheduling IoT applications in edge and fog computing environments: a taxonomy and future directions

M. Goudarzi, M. Palaniswami, and R. Buyya. “Scheduling IoT applications in edge and fog computing environments: a taxonomy and future directions”. In:ACM Computing Surveys 55.7 (2022), pp. 1–41

work page 2022

[4] [4]

TF-DDRL: ATransformer-enhanced DistributedDRL Technique for Scheduling IoT Applications in Edge and Cloud Computing Environments

Z. Wang, M.Goudarzi, andR. Buyya. “TF-DDRL: ATransformer-enhanced DistributedDRL Technique for Scheduling IoT Applications in Edge and Cloud Computing Environments”. In: IEEE Transactions on Services Computing (2025). 12

work page 2025

[5] [5]

Resource scheduling in edge computing: A survey

Q. Luo, S. Hu, C. Li, G. Li, and W. Shi. “Resource scheduling in edge computing: A survey”. In: IEEE Communications Surveys & Tutorials 23.4 (2021), pp. 2131–2165

work page 2021

[6] [6]

A comprehensive survey on reinforcement- learning-based computation offloading techniques in edge computing systems

D. Hortelano, I. de Miguel, R. J. D. Barroso, J. C. Aguado, N. Merayo, L. Ruiz, A. Asensio, X. Masip-Bruin, P. Fernández, R. M. Lorenzo, et al. “A comprehensive survey on reinforcement- learning-based computation offloading techniques in edge computing systems”. In:Journal of Network and Computer Applications 216 (2023), p. 103669

work page 2023

[7] [7]

A state-of-the-art review of task scheduling for edge computing: A delay-sensitive application perspective

A. Avan, A. Azim, and Q. H. Mahmoud. “A state-of-the-art review of task scheduling for edge computing: A delay-sensitive application perspective”. In:Electronics 12.12 (2023), p. 2599

work page 2023

[8] [8]

Deep Reinforcement Learning for Online Computation Offloading in Wireless Powered Mobile-Edge Computing Networks

L. Huang, S. Bi, and Y.-J. A. Zhang. “Deep Reinforcement Learning for Online Computation Offloading in Wireless Powered Mobile-Edge Computing Networks”. In:IEEE Transactions on Mobile Computing 19.11 (2020), pp. 2581–2593.doi: 10.1109/TMC.2019.2928811

work page arXiv 2020