Autonomous Resource Management in Microservice Systems via Reinforcement Learning

Ming Gong; Nia Qi; Wuyang Zhang; Yingnan Deng; Yujun Zou; Zhihao Xue

REVIEW 4 major objections 4 minor 1 cited by

Autonomous Resource Management in Microservice Systems via Reinforcement Learning

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a reinforcement-learning agent can schedule microservice resources and data flows in real time, outperforming static rules and four baselines on response time, throughput, and resource utilization.

desk verdict Standard RL recipe with an unverifiable Table 1: the comparison baselines come from different problem settings, so the main claim has no support. read the letter →

arxiv 2507.12879 v1 pith:MA4AM4II submitted 2025-07-17 cs.DC cs.LG

Yujun Zou , Nia Qi , Yingnan Deng , Zhihao Xue , Ming Gong , Wuyang Zhang This is my paper · ORCID

classification cs.DCcs.LG

keywords reinforcementlearningmicroservicearchitectureresourceschedulingallocationQ-learningdeepQ-networkthroughputcostefficiency

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

This paper tries to establish that a reinforcement-learning agent can make microservice resource scheduling and data-flow allocation adaptive enough to beat static, rule-based methods in dynamic cloud environments. The agent observes each service's workload and resource usage, picks a discrete allocation and scheduling action, and receives a reward that weights lower response time against higher resource utilization. Using Q-learning or a deep Q-network with experience replay and target networks, the policy is updated continuously as load and network conditions change. The reported experiments claim the method reaches 120 ms response time, 950 requests per second throughput, 85% resource utilization, and 92% cost efficiency, outperforming four comparison methods. If these numbers hold, the practical payoff is that microservice systems could self-adjust in real time instead of relying on manual configuration or static load-balancing rules.

What carries the argument

The load-bearing object is the reinforcement-learning update loop over a discrete action space. State $S_t$ collects per-service workload and resource usage; action $a_t$ selects a resource allocation and data-flow path; the reward is a weighted sum $R_t = \sum_i (\lambda_i(-R_i(t)) + \alpha_i U_i(t))$, where $R_i(t)$ is service $i$'s response time and $U_i(t)$ its utilization. The agent estimates the state-action value $Q(s_t,a_t)$ with Q-learning or with a deep Q-network that uses experience replay and target networks, updating $Q(s_t,a_t) \leftarrow Q(s_t,a_t) + \alpha[r_t + \gamma \max_{a'} Q(s_{t+1},a') - Q(s_t,a_t)]$. This mechanism is what converts live performance feedback into scheduling decisions, and it is the component the experiments are testing.

What would settle it

A reader could settle the claim by running the proposed RL scheduler and the four baselines on the same workload trace and instance pool, measuring response time, throughput, and utilization with identical instrumentation. If the proposed model's 120 ms, 950 req/s, and 85% figures do not reproduce, or if any baseline matches or beats it under those shared conditions, the claimed superiority is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that its reinforcement-learning-based scheduling method outperforms existing approaches across the metrics that matter for microservice operation. In the comparison, the proposed model records 120 ms response time versus 130–180 ms for the baselines, 950 requests per second versus 870–910, 85% resource utilization versus 75–83%, and 92% cost efficiency with 150 joules of energy consumption. The paper also claims the method stays effective as conditions worsen: scheduling efficiency falls from 92% at low load to 72% at ultra-high load, and from 95% at 10 ms network latency to 75% at 50 ms. Under multi-dimensional resource constraints, it reports 94% efficiency for CPU-bound scheduling, 88% for memory, 82% for storage, and 78% for network. The author's conclusion is that reinforcement learning is a viable basis for autonomous, real-time resource management in microservice systems.

Load-bearing premise

The paper's headline numbers stand or fall on the assumption that the four baseline results it quotes were measured under the same workload, hardware, and metric definitions as the proposed model, since no shared experimental setup is shown.

Editorial extensions

If this is right

If the reported 120 ms response time is reproducible, RL scheduling removes a meaningful fixed latency component in high-concurrency microservice calls.
Sustaining 950 requests per second implies the learned data-flow paths do not become the bottleneck under the tested load.
The efficiency curve—92% at low load, 72% at ultra-high load—means the method pays off most when the system still has slack to redistribute.
The reported 92% cost efficiency connects the scheduling policy to energy and operating cost, so performance gains and cost gains move together.

Reading between the lines

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

The paper leaves implicit that the reward weights $\lambda_i$ and $\alpha_i$ are tunable; a direct extension would be multi-objective RL that learns these weights from an SLO or energy budget rather than setting them by hand.
Since the four baselines in Table 1 come from separate studies, the cleanest next test is a head-to-head rerun on one shared trace and instance pool—that experiment is not in the paper.
The latency sweep, from 95% efficiency at 10 ms down to 75% at 50 ms, suggests that training the agent under a distribution of network latencies rather than one condition could raise worst-case efficiency.
The discrete action space limits the granularity of allocation changes; extending the same reward to a continuous-action actor-critic would test whether finer-grained scheduling improves the reported utilization numbers.

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

Quick take: this is a standard Q-learning/DQN scheduling write-up, and the only evidence for its headline claim is a single table with bare numbers. The method section is a textbook exposition, the related work is on-topic, and the writing is clear. So the paper is not nonsense; it just has no experimental backbone.

What is actually new? Nothing algorithmic. The reward is a weighted sum of response time and utilization, the update is textbook Q-learning/DQN, and the paper itself cites several RL-based microservice schedulers it resembles. The only candidate new artifact is the experimental table, and that does not hold up.

The problems are concentrated in Section IV. The dataset is described as 'publicly available open-source' but never named, so no one can reproduce the environment. Table 1 reports response time, throughput, and utilization as single points with no error bars, no trial counts, and no measurement procedure. The baselines are not from a shared testbed: [24] is a Kubernetes multi-cluster deployment system, [25] is a review, [26] is multi-workflow scheduling in product design, and [27] is monolith-extraction. They were never run under the same workload. So the 120 ms vs 130-180 ms comparison is meaningless. The energy (150 joules) and cost efficiency (92%) numbers appear only in prose, not in the table. Figures 2-4 have no axis labels or workload definitions.

The reward function in Section III explicitly rewards low response time and high utilization, the same metrics that later appear as improvements. That does not prove fraud, but it means you cannot tell whether the reported gains come from the scheduling logic or from the chosen weights, and without a controlled comparison the claim is just the objective function coming back.

The math is standard and not wrong. The citation pattern is generally on-topic, though the experimental baselines are mismatched. There is no shipped code or data.

Who this is for: someone looking for a quick summary of how RL can be framed for microservice scheduling, not someone who needs a reliable result. A serious referee would spend an hour and find no testable claim.

Recommendation: desk reject in current form. If the authors rerun a controlled experiment with a named public workload, actual baselines from the same testbed, and variance statistics, it could become a workshop-level empirical paper. As is, the central claim is unverifiable.

Referee Report

4 major / 4 minor

Summary. The paper proposes a reinforcement learning-based approach to microservice resource scheduling, using Q-learning or DQN to choose actions from a discrete action space based on a reward function that penalizes response time and rewards resource utilization. The authors report experiments claiming the proposed method outperforms four baselines on response time, throughput, resource utilization, energy consumption, and cost efficiency, and they conclude that RL offers strong adaptability for dynamic cloud environments.

Significance. The problem of adaptive resource management in microservices is timely and relevant, and the proposed formulation is a standard application of RL to scheduling. If the reported improvements were obtained through a rigorous, reproducible experimental comparison, the paper would provide useful evidence for RL-based scheduling in cloud-native systems. However, the current manuscript does not supply a controlled evaluation: the dataset is not identified, the baselines are drawn from unrelated problem settings, and the reported numbers lack any statistical grounding. As presented, the central claim of superiority is unsupported and cannot be independently assessed.

major comments (4)

[Section IV-B, Table 1] The comparative evaluation is not controlled. The four baselines ([24]–[27]) address fundamentally different tasks: [24] is a Kubernetes multi-cluster deployment RL system, [25] is a review article on load balancing and auto-scaling, [26] targets multi-workflow scheduling in product design via meta-RL, and [27] extracts microservices from monolithic systems. None of these is shown to have been evaluated on the same microservice application, workload trace, cluster capacity, or measurement window as the proposed method. Without a shared experimental setup, the numerical comparisons in Table 1 cannot support the claim that the proposed model 'significantly improves' over these baselines.
[Section IV-A, Section IV-B, Figures 2–4] The experimental setup is not reproducible. The 'publicly available open-source Cloud Resource Management dataset' is never identified by name, version, or URL, and no details are given about the microservice application, workload generation, hardware, or measurement methodology. Figures 1–4 are referenced in the text but are not actually present in the manuscript, and the prose descriptions of Figures 2–4 report 'efficiency' percentages without axis labels, units, or a definition of efficiency. In addition, the energy consumption (150 joules) and cost efficiency (92%) values appear only in prose and are absent from Table 1, so they cannot be checked against any baseline.
[Section IV-B, Abstract] There is no statistical support for the claimed improvements. Table 1 reports single point estimates with no error bars, confidence intervals, number of trials, or significance tests, and the word 'significantly' in the Abstract is not backed by any statistical evidence. The paper also does not report variance across runs or different workload seeds, which is essential for stochastic RL-based schedulers.
[Section III, Eq. (1) and Q-learning update] The reward function and hyperparameters are underspecified. The reward is a weighted combination of response time and resource utilization, with weights λ_i and α_i, but the paper does not state how these weights were chosen or whether the results are sensitive to them. Similarly, the Q-learning learning rate α, discount factor γ, and DQN architecture are not given. Because the reported metrics are exactly the components of the reward, the reported gains on response time and resource utilization are at least partly by construction; without a sensitivity analysis or a description of how the weights were set, the comparison is difficult to interpret as a fair evaluation of the scheduling policy.

minor comments (4)

[Section III, equations] The equations are heavily garbled by formatting artifacts, with missing subscripts and misplaced symbols; for example, the Q-learning update is unreadable and the text refers to 'E value' where 'Q value' is intended. Clean versions of all equations are needed.
[Section IV-B] The energy and cost-efficiency metrics mentioned in prose should be added to Table 1 with corresponding baseline values, or removed from the claims if they were not measured for all methods.
[References] Several references are incomplete: [7] and [22] lack venue information, and [4], [5], and [23] lack page numbers or article numbers where appropriate. The bibliography should be checked for completeness.
[General] The paper does not include a limitations section. Given that the experimental results degrade under high load and high latency (as described in Section IV-B), a discussion of these limitations and their implications for real-world deployment would improve the manuscript.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reward function is an optimization objective, not a restated prediction, and no load-bearing derivation reduces to the paper's own inputs.

full rationale

The only candidate for a circular step is the reward function in Section III, which is defined as a weighted sum of response time and resource utilization, while Section IV-B reports improved response time and utilization. But this is not a circular derivation: optimizing a reward that includes those metrics is the standard RL objective, and the reported absolute values in Table 1 are empirical outcomes of training rather than a restatement of equation (1). The reward does not include throughput, energy, or cost efficiency, so the broader claims cannot reduce to the reward definition either. The paper does cite a work by a co-author, reference [7], but only in a generic applications list and not as load-bearing evidence for the central claim. The substantial weaknesses are experimental: Table 1 compares against heterogeneous baselines [24]-[27] without a described shared setup, the dataset is unnamed, and Figures 2-4 lack axis labels and measurement procedures. These are validity and evidence problems, not circularity, and per the review rules they do not raise the circularity score.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

No new physical or formal entities are introduced. The paper relies on standard MDP components and an unnamed dataset. The main unaccounted-for elements are free reward weights and unverified comparability assumptions, not invented entities.

free parameters (2)

Reward weights λ_i and α_i = Not specified
The reward in Section III is defined as a weighted sum of response time and resource utilization for each service, but the paper never states the weight values or how they were chosen. The reported behavior depends on them.
Q-learning learning rate and discount factor = Not specified
The update rule in Section III uses standard Q-learning hyperparameters, but no values or schedules are reported, so any reproduction would require arbitrary choices.

assumptions (4)

domain assumption Microservice resource scheduling can be modeled as a Markov decision process with a discrete action space.
Stated in Section III: the state S and action space A are assumed. No validation that the real system satisfies Markov property or discrete actions.
domain assumption A publicly available Cloud Resource Management dataset exists and captures realistic microservice behavior.
Section IV-A describes the dataset but never names it or provides a URL, so its existence and suitability are assumed.
ad hoc to paper The baseline values in Table 1 were produced under conditions equivalent to the proposed model's.
Section IV-B presents baseline numbers from four papers with different scopes; the paper assumes comparability without a shared experimental setup.
standard math Q-learning with function approximation converges to a good policy in this setting.
The paper relies on standard convergence properties of Q-learning/DQN in Section III, without verifying the conditions (e.g., finite state, exploration schedule) hold for its problem.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Autonomous Resource Management in Microservice Systems via Reinforcement Learning." pith.science (2026). https://pith.science/paper/MA4AM4II

@misc{pith2026250712879,
  author       = {Pith},
  title        = {Pith review of: Autonomous Resource Management in Microservice Systems via Reinforcement Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MA4AM4II}},
  note         = {Machine review of arXiv:2507.12879}
}

read the original abstract

This paper proposes a reinforcement learning-based method for microservice resource scheduling and optimization, aiming to address issues such as uneven resource allocation, high latency, and insufficient throughput in traditional microservice architectures. In microservice systems, as the number of services and the load increase, efficiently scheduling and allocating resources such as computing power, memory, and storage becomes a critical research challenge. To address this, the paper employs an intelligent scheduling algorithm based on reinforcement learning. Through the interaction between the agent and the environment, the resource allocation strategy is continuously optimized. In the experiments, the paper considers different resource conditions and load scenarios, evaluating the proposed method across multiple dimensions, including response time, throughput, resource utilization, and cost efficiency. The experimental results show that the reinforcement learning-based scheduling method significantly improves system response speed and throughput under low load and high concurrency conditions, while also optimizing resource utilization and reducing energy consumption. Under multi-dimensional resource conditions, the proposed method can consider multiple objectives and achieve optimized resource scheduling. Compared to traditional static resource allocation methods, the reinforcement learning model demonstrates stronger adaptability and optimization capability. It can adjust resource allocation strategies in real time, thereby maintaining good system performance in dynamically changing load and resource environments.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Artificial Intelligence-Based Multiscale Temporal Modeling for Anomaly Detection in Cloud Services
cs.LG 2025-08 reject novelty 3.0 of 10

A Transformer plus multiscale attention-weighted fusion is claimed to improve cloud anomaly detection metrics by 2-3 points, but the missing label definition and artifacts block verification.

Reference graph

Works this paper leans on

27 extracted references · 22 canonical work pages · cited by 1 Pith paper

[24]

Efficient microservice deployment in Kubernetes multi-clusters through reinforcement learning,

J. Santos, M. Zaccarini, F. Poltronieri, et al., "Efficient microservice deployment in Kubernetes multi-clusters through reinforcement learning," Proceedings of the NOMS 2024 IEEE Network Operations and Management Symposium, pp. 1–9, 2024

work page 2024
[25]

A cloud-based container microservices: a review on load-balancing and auto-scaling issues,

S. Rabiu, C. H. Yong, and S. M. S. Mohamad, "A cloud-based container microservices: a review on load-balancing and auto-scaling issues," International Journal of Data Science, vol. 3, no. 2, pp. 80–92, 2022

work page 2022
[26]

Multi-workflow dynamic scheduling in product design: a generalizable approach based on meta-reinforcement learning,

Z. Chen, L. Zhang, W. Cai, et al., "Multi-workflow dynamic scheduling in product design: a generalizable approach based on meta-reinforcement learning," Journal of Manufacturing Systems, vol. 79, pp. 334–346, 2025

work page 2025
[27]

Extracting microservices from monolithic systems using deep reinforcement learning,

K. Sellami and M. A. Saied, "Extracting microservices from monolithic systems using deep reinforcement learning," Empirical Software Engineering, vol. 30, no. 1, p. 1, 2025

work page 2025
[1]

Optimizing microservice orchestration using reinforcement learning for enhanced system efficiency,

S. R. Peddinti, B. K. Pandey, A. Tanikonda, et al., "Optimizing microservice orchestration using reinforcement learning for enhanced system efficiency," Distributed Learning and Broad Applications in Scientific Research| Annual, vol. 7, 2021

work page 2021
[2]

M. U. Hassan, A. A. Al-Awady, A. Ali, et al., "Smart resource allocation in mobile cloud next-generation network (NGN) orchestration with context-aware data and machine learning for the cost optimization of microservice applications," Sensors, vol. 24, no. 3, p. 865, 2024

work page 2024
[3]

Delay-aware optimization of fine-grained microservice deployment and routing in edge via reinforcement learning,

K. Peng, J. He, J. Guo, et al., "Delay-aware optimization of fine-grained microservice deployment and routing in edge via reinforcement learning," IEEE Transactions on Network Science and Engineering, 2024

work page 2024
[4]

Sequential recommendation via time-aware and multi-channel convolutional user modeling,

Y. Xing, Y. Wang, and L. Zhu, "Sequential recommendation via time-aware and multi-channel convolutional user modeling," Transactions on Computational and Scientific Methods, vol. 5, no. 5, 2025

work page 2025

Show all 27 references

[5]

Collaborative optimization in federated recommendation: integrating user interests and differential privacy,

L. Zhu, W. Cui, Y. Xing, and Y. Wang, "Collaborative optimization in federated recommendation: integrating user interests and differential privacy," Journal of Computer Technology and Software, vol. 3, no. 8, 2024

2024
[6]

State-aware IoT scheduling using deep Q-networks and edge-based coordination,

Q. He, C. Liu, J. Zhan, W. Huang, and R. Hao, "State-aware IoT scheduling using deep Q-networks and edge-based coordination," arXiv preprint, arXiv:2504.15577, 2025

2025 arXiv
[7]

Unified instruction encoding and gradient coordination for multi-task language models,

W. Zhang, Z. Xu, Y. Tian, Y. Wu, M. Wang, and X. Meng, "Unified instruction encoding and gradient coordination for multi-task language models," 2025

2025
[8]

Integrating cognition cost with reliability QoS for dynamic workflow scheduling using reinforcement learning,

X. Yu, W. Wu, and Y. Wang, "Integrating cognition cost with reliability QoS for dynamic workflow scheduling using reinforcement learning," IEEE Transactions on Services Computing, vol. 16, no. 4, pp. 2713– 2726, 2023

2023
[9]

DRS: A deep reinforcement learning enhanced Kubernetes scheduler for microservice based system,

Z. Jian, X. Xie, Y. Fang, et al., "DRS: A deep reinforcement learning enhanced Kubernetes scheduler for microservice based system," Software: Practice and Experience, vol. 54, no. 10, pp. 2102–2126, 2024

2024
[10]

Continuous control-based load balancing for distributed systems using TD3 reinforcement learning,

Y. Duan, "Continuous control-based load balancing for distributed systems using TD3 reinforcement learning," Journal of Computer Technology and Software, vol. 3, no. 6, 2024

2024
[11]

Topology-aware decision making in distributed scheduling via multi-agent reinforcement learning,

B. Wang, "Topology-aware decision making in distributed scheduling via multi-agent reinforcement learning," Transactions on Computational and Scientific Methods, vol. 5, no. 4, 2025

2025
[12]

A scalable modified deep reinforcement learning algorithm for serverless IoT microservice composition infrastructure in fog layer,

M. E. Khansari and S. Sharifian, "A scalable modified deep reinforcement learning algorithm for serverless IoT microservice composition infrastructure in fog layer," Future Generation Computer Systems, vol. 153, pp. 206–221, 2024

2024
[13]

Sinan: ML based and QoS aware resource management for cloud microservices,

Y. Zhang, W. Hua, Z. Zhou, et al., "Sinan: ML based and QoS aware resource management for cloud microservices," Proceedings of the 26th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, pp. 167–181, 2021

2021
[14]

AI enhanced performance optimization for microservice based systems,

V. Ramamoorthi, "AI enhanced performance optimization for microservice based systems," Journal of Advanced Computing Systems, vol. 4, no. 9, pp. 1–7, 2024

2024
[15]

Anomaly detection in microservice environments via conditional multiscale GANs and adaptive temporal autoencoders,

Y. Ma, "Anomaly detection in microservice environments via conditional multiscale GANs and adaptive temporal autoencoders," Transactions on Computational and Scientific Methods, vol. 4, no. 10, 2024

2024
[16]

A deep learning-based predictive framework for backend latency using AI-augmented structured modeling,

Z. Fang, "A deep learning-based predictive framework for backend latency using AI-augmented structured modeling," Journal of Computer Technology and Software, vol. 3, no. 7, 2024

2024
[17]

Deep graph modeling for performance risk detection in structured data queries,

D. Gao, "Deep graph modeling for performance risk detection in structured data queries," Journal of Computer Technology and Software, vol. 4, no. 5, 2025

2025
[18]

Graph neural network-based collaborative perception for adaptive scheduling in distributed systems,

W. Zhu, Q. Wu, T. Tang, R. Meng, S. Chai, and X. Quan, "Graph neural network-based collaborative perception for adaptive scheduling in distributed systems," arXiv preprint, arXiv:2505.16248, 2025

2025 arXiv
[19]

MobileNet compression and edge computing strategy for low-latency monitoring,

J. Zhan, "MobileNet compression and edge computing strategy for low-latency monitoring," Journal of Computer Science and Software Applications, vol. 4, no. 4, 2024

2024
[20]

Distributed network traffic scheduling via trust-constrained policy learning mechanisms,

Y. Ren, M. Wei, H. Xin, T. Yang, and Y. Qi, "Distributed network traffic scheduling via trust-constrained policy learning mechanisms," Transactions on Computational and Scientific Methods, vol. 5, no. 4, 2025

2025
[21]

Time-series learning for proactive fault prediction in distributed systems with deep neural structures,

Y. Wang, W. Zhu, X. Quan, H. Wang, C. Liu, and Q. Wu, "Time-series learning for proactive fault prediction in distributed systems with deep neural structures," arXiv preprint, arXiv:2505.20705, 2025

2025 arXiv
[22]

A deep Q-network approach to intelligent cache management in dynamic backend environments,

Y. Sun, R. Meng, R. Zhang, Q. Wu, and H. Wang, "A deep Q-network approach to intelligent cache management in dynamic backend environments," 2025

2025
[23]

Unsupervised anomaly detection in structured data using structure-aware diffusion mechanisms,

H. Xin and R. Pan, "Unsupervised anomaly detection in structured data using structure-aware diffusion mechanisms," Journal of Computer Science and Software Applications, vol. 5, no. 5, 2025

2025

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [24]

Efficient microservice deployment in Kubernetes multi-clusters through reinforcement learning,

J. Santos, M. Zaccarini, F. Poltronieri, et al., "Efficient microservice deployment in Kubernetes multi-clusters through reinforcement learning," Proceedings of the NOMS 2024 IEEE Network Operations and Management Symposium, pp. 1–9, 2024

work page 2024

[2] [25]

A cloud-based container microservices: a review on load-balancing and auto-scaling issues,

S. Rabiu, C. H. Yong, and S. M. S. Mohamad, "A cloud-based container microservices: a review on load-balancing and auto-scaling issues," International Journal of Data Science, vol. 3, no. 2, pp. 80–92, 2022

work page 2022

[3] [26]

Multi-workflow dynamic scheduling in product design: a generalizable approach based on meta-reinforcement learning,

Z. Chen, L. Zhang, W. Cai, et al., "Multi-workflow dynamic scheduling in product design: a generalizable approach based on meta-reinforcement learning," Journal of Manufacturing Systems, vol. 79, pp. 334–346, 2025

work page 2025

[4] [27]

Extracting microservices from monolithic systems using deep reinforcement learning,

K. Sellami and M. A. Saied, "Extracting microservices from monolithic systems using deep reinforcement learning," Empirical Software Engineering, vol. 30, no. 1, p. 1, 2025

work page 2025

[5] [1]

Optimizing microservice orchestration using reinforcement learning for enhanced system efficiency,

S. R. Peddinti, B. K. Pandey, A. Tanikonda, et al., "Optimizing microservice orchestration using reinforcement learning for enhanced system efficiency," Distributed Learning and Broad Applications in Scientific Research| Annual, vol. 7, 2021

work page 2021

[6] [2]

M. U. Hassan, A. A. Al-Awady, A. Ali, et al., "Smart resource allocation in mobile cloud next-generation network (NGN) orchestration with context-aware data and machine learning for the cost optimization of microservice applications," Sensors, vol. 24, no. 3, p. 865, 2024

work page 2024

[7] [3]

Delay-aware optimization of fine-grained microservice deployment and routing in edge via reinforcement learning,

K. Peng, J. He, J. Guo, et al., "Delay-aware optimization of fine-grained microservice deployment and routing in edge via reinforcement learning," IEEE Transactions on Network Science and Engineering, 2024

work page 2024

[8] [4]

Sequential recommendation via time-aware and multi-channel convolutional user modeling,

Y. Xing, Y. Wang, and L. Zhu, "Sequential recommendation via time-aware and multi-channel convolutional user modeling," Transactions on Computational and Scientific Methods, vol. 5, no. 5, 2025

work page 2025