{"id":"a7127f04-6356-4df2-90b3-c8db48da8cba","arxiv_id":"2412.01839","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":1.0,"correctness_risk":"high","formal_verification":"none","parameter_count":6,"one_line_summary":"A replication study of PPO and ACER for O-RAN resource allocation that qualitatively claims DRL beats greedy, but contains an internal contradiction and no quantitative results.","lead":"This paper re-runs an existing deep reinforcement learning experiment for resource allocation in Open Radio Access Networks (O-RAN), comparing PPO and ACER against a greedy baseline. It reports that both DRL methods beat greedy, but the evidence is qualitative and includes a direct contradiction about which algorithm converges faster.","discovery_kind":"replication","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The paper's own results section says PPO converges faster while the abstract and conclusion say ACER; this unresolved contradiction undermines the central replication claim.","rationale":"The reader's weakest assumption focuses on circularity and unspecified training conditions. That is a serious concern, but the most load-bearing problem is internal: the paper directly contradicts itself on which algorithm converges faster. Even if the original code were sound and the training setup were fully specified, a replication claiming to confirm the original observations must at least be consistent about what was observed. The results section's statement that PPO converges faster versus the abstract's and conclusion's claim that ACER converges faster cannot both describe the same experiment. No convergence metric, quantitative result, or raw training curve is provided to resolve the contradiction. The paper also provides no code or data and reuses the original authors' model, so independent verification is impossible from the manuscript alone. I am not alleging misconduct; the text as written simply does not support the central claim in a checkable way. The reader's REJECT verdict should stand, and the concrete re-running test above would settle which convergence statement is correct.","tokens_in":7168,"tokens_out":3942,"duration_ms":39296,"concrete_test":"Run the original code from ref [16] (or, if unavailable, a clean reimplementation from the paper's MDP description) with the Table II settings, for at least 10 random seeds per algorithm. Define convergence as the first episode (or step) at which the rolling mean reward reaches a pre-specified threshold, e.g., 95% of the final converged reward or of the MIP optimum. Record both episodes-to-threshold and wall-clock time, and report medians with interquartile ranges. If PPO's median is lower, Section V is correct and the abstract/conclusion need correction; if ACER's is lower, the results section needs correction. Without such numbers, the contradiction cannot be resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section V states: 'we observed a faster convergence for the PPO model compared to ACER.' The abstract and Section VI assert the opposite: 'the off-policy model (ACER) shows a faster convergence' and 'ACER achieves faster convergence.' These cannot both be true. The central claim is that this replication corroborates the original observations; one of the two observed findings is internally contradicted. The paper defines no convergence metric (episodes to threshold, wall-clock time, sample complexity), reports no quantitative convergence values, no seeds, and no error bars for Figures 3-4. Therefore the reader cannot tell which algorithm is claimed to converge faster, and the repeated conclusion that ACER converges faster is load-bearing but unsupported by the only results sentence in the evaluation. This is not a style issue: if PPO converged faster in the actual runs, the abstract and conclusion are wrong; if ACER converged faster, the results section contradicts the paper's own summary. Either way, the replication evidence as presented cannot validate the original claim, so the central conclusion is not checkable.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper reports a replication study of Hammami and Nguyen's work on on-policy (PPO) and off-policy (ACER) deep reinforcement learning for resource allocation in sliced O-RAN systems. The authors re-use the original model and code (referenced only through a GitHub link), train both DRL agents for three hours, and compare their energy consumption and latency trade-off against a greedy baseline and a MIP-derived optimum. The abstract and conclusion claim that the results confirm the original findings: both DRL models outperform a greedy algorithm, PPO gives a favorable balance between energy consumption and user latency, and ACER converges faster. The evaluation section, however, states the opposite convergence result, reporting that PPO converged faster than ACER. The paper provides no quantitative metrics, error bars, seeds, or statistical tests to support any of these qualitative claims, and it does not numerically report the MIP or greedy baseline results to which the DRL models are compared.","tokens_in":7378,"tokens_out":3435,"duration_ms":35153,"significance":"A faithful, independent replication of a published O-RAN resource-allocation study would be valuable to the community, and the authors deserve credit for attempting to reproduce prior results on the Alibaba cluster-trace dataset and for applying two DRL paradigms in the same experimental framework. However, the manuscript as written provides no useful checkable evidence: the only explicit statement about convergence in the evaluation contradicts the paper's own abstract and conclusion; no convergence metric is defined; no numerical values, confidence intervals, or statistical tests are reported; and the study is not an independent implementation but a re-run of the original authors' model. The central replication claims are therefore unsupported and internally inconsistent, and the paper's potential significance cannot be assessed from the presented material.","major_comments":[{"comment":"The paper contains a direct and load-bearing contradiction about which algorithm converges faster. Section V(a) states \"we observed a faster convergence for the PPO model compared to ACER,\" while the abstract states \"the off-policy model (ACER) shows a faster convergence\" and Section VI concludes \"ACER achieves faster convergence.\" These statements cannot both be true. The paper defines no convergence metric (episodes to threshold, wall-clock time, sample complexity), reports no quantitative convergence values, and provides no error bars or seed statistics for Figures 3-4. Since the replication's stated purpose is to validate the original observations, this unresolved contradiction makes the central finding uncheckable and unsupported.","section":"Section V(a), Abstract, Section VI"},{"comment":"No quantitative results are reported anywhere in the evaluation. The claims that both DRL models \"showed superior performance compared to the greedy baseline,\" that \"PPO showed the most efficient energy consumption, which resembles the optimal solution,\" and that PPO \"showed better optimization in balancing latency and energy efficiency\" are made without any numbers, such as mean power consumption, average latency, percentage improvement, or confidence intervals. Figures 3-8 are qualitative line plots with no error bands or multiple-seed statistics. The absence of quantitative evidence means the reader cannot independently verify the main comparative claims, and the paper's stated aim of \"proving the efficacy\" of the models is not met.","section":"Section V"},{"comment":"The replication is not independent. The text states \"we used the available model proposed in the original research\" and cites only a GitHub repository (reference [16]), with no description of the original model, no list of modifications made by the authors, no hyperparameters for PPO or ACER beyond Table II (which omits e.g. clipping ratio, entropy coefficient, and replay buffer size), and only the statement that training lasted three hours. A replication that reuses the same implementation and experimental setup cannot by itself validate the original findings unless it provides an independent implementation or at least a detailed comparison of training configurations. As presented, the findings are generated by the same artifacts under test, so the claimed \"confirmation\" risks circularity rather than providing independent evidence.","section":"Section V, Ref [16]"},{"comment":"The comparison against the MIP optimal solution and the greedy baseline is not verifiable. Section V mentions that the setup includes \"the optimal solution derived from the MIP model which uses the Gurobi solver, and the same for the greedy baseline approach,\" but no table or figure reports the MIP and greedy numerical values, and the paper does not state over how many instances or under what conditions these baselines were computed. Without those data, statements such as PPO's energy consumption \"resembles the optimal solution\" are unsupported. The paper should provide at least one table with the mean and standard deviation of power consumption and latency for MIP, greedy, PPO, and ACER across multiple seeds and across the different network load levels shown in Figures 7-8.","section":"Section V, Section III-B"}],"minor_comments":[{"comment":"The paper misnames ACER in the text: it introduces \"Advantage Actor-Critic (ACER)\" but ACER stands for Actor-Critic with Experience Replay, as correctly stated in the abstract and in reference [14]. The naming should be corrected throughout.","section":"Section IV, paragraph before 'ACER'"},{"comment":"Equation (3) contains a typo in the subscript: \"zt,m\" should be \"z_t^m\" to match the notation used in the power consumption formula and elsewhere.","section":"Equation (3)"},{"comment":"The paper states \"we used the available model proposed in the original research and trained them for a duration of three hours\" without specifying the hardware, the number of episodes or steps, or the number of independent runs. These details are necessary for any reproducibility claim.","section":"Section V"},{"comment":"Reference [16] is only a GitHub URL; the manuscript should cite the original paper by Hammami and Nguyen with full publication details, and should state explicitly what version of the code was used and whether any modifications were made.","section":"References"},{"comment":"The conclusion states that the study \"proves\" the efficacy of the DRL models; given the qualitative nature of the results and the absence of statistical analysis, a claim of proof is too strong and should be tempered to \"supports\" or \"is consistent with.\"","section":"Section VI"}],"recommendation":"reject","confidential_remarks":"The manuscript is more of a technical note than a rigorous replication study. The internal contradiction on convergence, the complete absence of quantitative evaluation, and the reliance on the original authors' code without independent implementation are all load-bearing issues that cannot be fixed by minor revision. Rejection seems appropriate; even a major revision would require a new experimental section with statistical results and an independent implementation or, at minimum, a detailed description of the original model and a resolution of the contradiction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: this is an explicit replication of Hammami and Nguyen's O-RAN resource-allocation DRL work. The paper's own results section says PPO converged faster, while the abstract and conclusion say ACER did. No convergence metric, numerical values, seeds, or error bars are provided, so the central claim is not checkable.\n\nWhat the paper does well: it is honest about being a replication, and it doesn't claim more than a validation attempt. The system model and MIP formulation in Section III are clearly laid out, and Table II lists some key hyperparameters. Naming the original codebase in ref [16] is the right thing to do.\n\nThe problems are load-bearing, not cosmetic. The contradiction between Section V ('we observed a faster convergence for the PPO model compared to ACER') and the abstract/conclusion ('ACER achieves faster convergence') is exactly the kind of thing that decides which algorithm is the takeaway. On top of that, all empirical claims rest on figures that are not present in the manuscript text, so the reader cannot see the actual curves. There is no definition of convergence, no seeds, no error bars, and no numerical comparisons against greedy or the optimal MIP solution. The study reuses the original authors' model and code, so it is a re-run of the same artifacts; that is not an independent replication of the result. Key training details like neural-network sizes, number of episodes, and the mentioned 'variations in learning rates' are unspecified, which makes the three-hour training run impossible to reproduce from the paper alone. The conclusion's assertion of robustness and reproducibility is unsupported by the evidence provided.\n\nA reader who wants a clean statement of the O-RAN slicing problem might get some value from Section III, but anyone looking for a trustworthy comparison of PPO and ACER in O-RAN will not. This paper should be desk-rejected; a serious referee would spend the whole time asking for the omitted numbers and the contradiction resolution. If the authors add the actual learning curves, seeds, error bars, a clear definition of convergence, and settle which algorithm converged faster, this could become a modest replication note, but not in its current form.","headline":"A replication effort that undermines itself: the results section and the abstract say opposite things about which DRL algorithm converges faster, and no quantitative evidence is provided.","tokens_in":7882,"tokens_out":3721,"would_cite":false,"duration_ms":32161,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A replication rerun finds two DRL models beat greedy in O-RAN resource allocation, with PPO balancing energy and latency best.","keywords":["O-RAN","resource allocation","deep reinforcement learning","Proximal Policy Optimization","ACER","network slicing","latency","replication study"],"falsifier":"Use the original code to retrain both agents with published hyperparameters and fixed random seeds, then check per-episode rewards against the reported curves: if either DRL agent fails to beat the greedy baseline, or if ACER converges no faster than PPO, the replication claim is false.","tokens_in":6942,"feed_emoji":"📡","tokens_out":8212,"duration_ms":71470,"temperature":0.7,"pith_summary":"This paper is a replication of an earlier study that compared two deep reinforcement learning (DRL) approaches for allocating computing resources in an Open Radio Access Network (O-RAN). The question is whether an on-policy agent (Proximal Policy Optimization, PPO) or an off-policy agent (Sample Efficient Actor-Critic with Experience Replay, ACER) better handles a radio network with latency-sensitive surveillance video users and latency-tolerant users. Training both agents for three hours on the same simulation setup and workload, the authors find that both outperform a greedy baseline and that PPO gives a better balance between energy consumption and user latency while ACER converges faster. The intended contribution is confirmation that the original findings reproduce, so that operators have dependable guidance on which DRL family to use for dynamic resource orchestration.","feed_headline":"Study replicates: deep RL beats greedy in Open RAN slicing","feed_subtitle":"A rerun of prior work finds PPO best balances energy and latency; ACER converges faster.","key_machinery":"The load-bearing object is the O-RAN cross-slice resource orchestrator cast as a Markov decision process: at each slicing window the agent observes the demands of video cameras and latency-tolerant users plus the current vO-DU loads, chooses which vO-DU serves the incoming request, and receives a reward equal to the negative weighted sum of vO-DU power consumption $-\\alpha \\sum_m P_t^m$. The power model $P_t^m = P(0\\%) + (P(100\\%)-P(0\\%))(2z_t^m-(z_t^m)^{1.4})$, taken from the cited literature, is the same objective used in the mixed-integer programming optimum and the greedy baseline. PPO carries the argument through a clipped surrogate objective that keeps policy updates stable, while ACER exploits a replay buffer to reuse past transitions; the two mechanisms are what produce the claimed stability-versus-sample-efficiency contrast.","core_discovery":"The central claim is that the replication confirms the original study: a PPO-based on-policy orchestrator and an ACER-based off-policy orchestrator both reduce the power consumption of the vO-DU pool compared with a greedy heuristic, while respecting the latency constraint of real-time video surveillance. In head-to-head comparison, PPO approaches the optimal energy-latency trade-off more closely than ACER, and ACER reaches a stable policy faster in training. The paper also reports that PPO's training performance is stable across different dense-layer sizes, whereas ACER's is not, which matters when an operator must choose a model under hardware or time constraints.","pith_inferences":["The paper's own evidence stops at aggregate training curves; a sharper test would publish the trained weights and per-run random seeds so another team can quantify how much of the gap between PPO and ACER is algorithmic rather than noise.","The same MDP could be extended to mixed service classes with different latency budgets, but the paper does not test whether PPO's balance holds when the reward weights change.","A practical implication the authors leave implicit is that the faster-converging ACER might be preferable for online retraining in a non-real-time RIC, where traffic drift forces periodic model updates."],"forward_implications":["O-RAN operators can realistically choose PPO when the goal is to minimize energy without exceeding the latency budget of real-time services.","ACER becomes attractive when training time is scarce, since it converges faster even if it is more sensitive to network architecture.","The comparison template of DRL agent versus MIP optimum versus greedy baseline gives a reusable yardstick for evaluating any future resource-allocation policy in O-RAN slicing.","If the replication is truthful, the original result is robust to a different three-hour training run and, for PPO, to variations in dense-layer size."],"supporting_citations":[{"why":"Original study's shared code and experimental configuration that this paper replicates; without it, the setup is underspecified.","marker":"[16]"},{"why":"Defines the Proximal Policy Optimization algorithm used as the on-policy DRL agent.","marker":"[13]"},{"why":"Defines the Sample Efficient Actor-Critic with Experience Replay algorithm used as the off-policy DRL agent.","marker":"[14]"},{"why":"Public production cluster trace dataset used as the workload for training and evaluation.","marker":"[15]"},{"why":"Supplies the vO-DU power consumption model that defines the objective and the reward.","marker":"[11]"},{"why":"Establishes the MDP formulation for DRL-based resource management in network slicing.","marker":"[12]"},{"why":"Provides the on-policy versus off-policy update comparison that motivates the paper's central question.","marker":"[8]"}],"fun_headline_variants":["PPO vs ACER in O-RAN: replication confirms DRL wins","Replication: PPO balances energy and latency, ACER converges faster","O-RAN: PPO best trade-off, ACER fastest convergence","Deep RL for O-RAN: replication validates PPO and ACER over greedy"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The conclusions stand only if the original study's code and experimental configuration, which this paper references rather than re-specifies, are correct and if a three-hour training run with unreported hyperparameters truly matches that setup.","fun_headline_variants_meta":{"raw":{"variants":["PPO vs ACER in O-RAN: replication confirms DRL wins","Replication: PPO balances energy and latency, ACER converges faster","O-RAN: PPO best trade-off, ACER fastest convergence","Deep RL for O-RAN: replication validates PPO and ACER over greedy"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000447,"raw_usage":{"total_tokens":2286,"prompt_tokens":1003,"completion_tokens":1283,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":619,"completion_tokens_details":{"reasoning_tokens":1201}},"tokens_in":619,"tokens_out":1283,"duration_ms":41161,"temperature":1.0,"reasoning_tokens":1201,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T18:51:31.450748+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Use the original code to retrain both agents with published hyperparameters and fixed random seeds, then check per-episode rewards against the reported curves: if either DRL agent fails to beat the greedy baseline, or if ACER converges no faster than PPO, the replication claim is false.","supporting_citations":[{"cited_title":"”On-policy vs","cited_arxiv_id":null,"evidence_quote":"Original study's shared code and experimental configuration that this paper replicates; without it, the setup is underspecified."},{"cited_title":"[online] Available: https: //github.com/alibaba/clusterdata/tree/master/cluster-trace-v2018/","cited_arxiv_id":null,"evidence_quote":"Public production cluster trace dataset used as the workload for training and evaluation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the vO-DU power consumption model that defines the objective and the reward."},{"cited_title":"”Deep reinforcement learning for resource management in network slicing.” IEEE Access , 6:74429–74441, 2018","cited_arxiv_id":null,"evidence_quote":"Establishes the MDP formulation for DRL-based resource management in network slicing."},{"cited_title":"”On-Policy vs","cited_arxiv_id":null,"evidence_quote":"Provides the on-policy versus off-policy update comparison that motivates the paper's central question."}],"review_version":1}