acceptodds
Under review as a conference paper at ICLR 2027

OptRoute: Evolving Trees that Couple Skill Organization and Retrieval for LLM-Based Optimization Modeling

Abstract

Large language models (LLMs) have shown strong potential for automating optimization modeling from natural-language problem descriptions. Prompt-based methods are particularly attractive because they can reuse external modeling knowledge at inference time without updating model parameters. However, as reusable modeling knowledge grows, it becomes increasingly challenging to organize related skills while ensuring that the right knowledge can be reliably retrieved for new problems. Existing approaches typically organize skills using flat catalogs or predefined hierarchies, leaving skill organization disconnected from how the stored knowledge is subsequently retrieved. We propose OptRoute, a framework that couples skill organization and retrieval through an evolving tree. Within the tree, paradigm nodes organize related formulation families and provide hierarchical routing, while leaf nodes preserve complete, solver-ready skills. As new reference-grounded skills are added, OptRoute jointly updates skill content and tree structure, refining existing skills and evolving paradigms that distinguish formulation families relevant to future retrieval. At inference time, a new problem is routed through the resulting hierarchy to retrieve relevant skills for solver-code generation. Experiments on eight OR benchmarks show that OptRoute achieves 79.2% macro-average Pass@1 accuracy with DeepSeek-V4, outperforming the evaluated baselines. These results demonstrate the effectiveness of coupling skill organization and retrieval for scalable reuse of optimization modeling knowledge.

Primary area:
optimization
Keywords:
LLM-based optimization modeling, operations research, skill tree evolution, retrieval-augmented generation

Reviews

Among:
Shape:
1234this paper, Bottom 27%
Figure 1. Mean review rating of the reviewed papers at ICLR 2027. This paper averages 2.00 over 3 reviews, above 27% of the others.
Reviewer_HwdROpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

The paper proposes OptRoute, a framework for organizing and retrieving reusable formulation skills for optimization modeling. Skills are extracted from reference solutions and organized into an evolving tree, where paradigm nodes guide retrieval and leaf nodes store formulation knowledge. Retrieved skills are synthesized into a modeling plan for solver-code generation. Experiments on eight benchmarks and two backbones compare overall performance, memory organization, retrieval strategies, and memory scaling.

Critical strengths

The paper presents a clear framework for organizing and retrieving reusable optimization-modeling skills. Table 3 provides a useful controlled comparison by fixing the leaf skills and downstream generation pipeline while varying retrieval strategies. Improvements across eight benchmarks and two backbones further suggest that the approach is useful beyond a single model or problem setting.

Critical weaknesses
  1. The contribution of structural evolution is not sufficiently separated from other changes. Table 2 jointly changes skill refinement, consolidation, and tree organization. Consequently, the improvement does not establish how much comes from evolving the hierarchy rather than improving the skill content. Table 3 controls the leaf skills, but does not compare against a stronger retrieval pipeline such as dense retrieval followed by LLM reranking. Given the existing work on hierarchical retrieval and dynamic tree memory discussed in Section 4.2, the paper needs clearer evidence of what the proposed organization–retrieval coupling adds.

  2. The analysis does not yet establish the robustness or scope of the advantage. The macro-average improvement over the strongest agentic baseline is 3.0 percentage points with DeepSeek but only 0.9 with DouBao, and the main tables provide no uncertainty estimates. The scaling study increases both available knowledge and the evolved structure, so it does not isolate the benefit of tree evolution as memory grows. Beyond one illustrative case, there is little systematic analysis of retrieval failures, routing errors, or which formulation distinctions benefit most from the hierarchy. This limits the insight beyond showing an aggregate performance improvement.

Questions for authors
  1. Could you separate the effects of structural evolution and skill-content updates? For example, compare an evolving tree with frozen leaf content against a fixed tree with enriched skills.

  2. How does OptRoute compare with dense retrieval followed by LLM reranking under comparable inference budgets? What additional benefit does it provide over existing dynamic tree-memory approaches?

  3. How stable are the gains across different construction samples and insertion orders? Uncertainty estimates would help assess the smaller improvements, particularly with DouBao.

  4. Could you provide systematic retrieval-error analysis and comparisons at each memory scale using the same skills? This would clarify when hierarchical routing helps and whether the scaling gains come from better organization or additional knowledge.

I would consider raising my rating if the responses and supporting evidence address these concerns, particularly by isolating the benefit of structural evolution and demonstrating that the gains are robust.

3 comments

Response to Reviewer HwdR

We particularly appreciate the reviewer’s concrete suggestions and the indication that stronger evidence could change the assessment. We have prioritized directly isolating the proposed contribution, testing stronger retrieval baselines, and reporting robustness limitations transparently.

1. W1/Q1 — Isolating structural evolution from skill-content updates

W1: The contribution of structural evolution is not sufficiently separated from other changes. Table 2 jointly changes skill refinement, consolidation, and tree organization. Consequently, the improvement does not establish how much comes from evolving the hierarchy rather than improving the skill content. Q1: Could you separate the effects of structural evolution and skill-content updates? For example, compare an evolving tree with frozen leaf content against a fixed tree with enriched skills.

Thank you for this constructive suggestion to disentangle structural evolution from skill-content updates.

  • Isolating structural evolution from skill-content updates

We followed the suggested comparison and added a factor-isolating ablation in revised Section 3.3 (Table 3). The configurations use the same validated construction skills and downstream modeling pipeline:

ConfigurationMacro Pass@1
Structure-only (frozen skill content)74.8 ± 0.9
Content-only (fixed hierarchy)75.6 ± 1.0
Joint content + structure (OptRoute)79.3 ± 0.6

The full model improves by 4.5 and 3.7 points over the two restricted variants, respectively. The particularly large difference on MAMO-Complex (88.0 versus 65.5 and 76.6) suggests that the joint design is valuable when closely related formulations must be distinguished. We agree this is evidence of complementarity, rather than a precise additive attribution to each component.

2. W1/Q2 — Stronger retrieval baselines and dynamic-memory novelty

W1: ...Table 3 controls the leaf skills, but does not compare against a stronger retrieval pipeline such as dense retrieval followed by LLM reranking. Given the existing work on hierarchical retrieval and dynamic tree memory discussed in Section 4.2, the paper needs clearer evidence of what the proposed organization–retrieval coupling adds. Q2: How does OptRoute compare with dense retrieval followed by LLM reranking under comparable inference budgets? What additional benefit does it provide over existing dynamic tree-memory approaches.

Thank you for suggesting stronger retrieval baselines and asking us to clarify the contribution beyond existing dynamic memories.

  • Comparison with LLM-reranked retrieval.

We strengthened the controlled retrieval comparison in revised Section 3.3 (Table 4). BM25 and dense retrieval are each followed by LLM-based selection under a matched budget (the LLM selects at most three skills from the top 30 retrieved candidates). The leaf-skill memory and downstream generation pipeline are held fixed:

Retrieval strategyMacro Pass@1
BM25 + LLM selection75.3 ± 0.7
Dense + LLM selection76.8 ± 0.5
Flat LLM selection76.8 ± 0.2
Static clustering72.8 ± 0.8
OptRoute79.3 ± 0.6

OptRoute exceeds Dense + LLM by 2.5 points and BM25 + LLM by 4.0 points. On IndustryOR, the gap to Dense + LLM is 5.5 points (73.0 versus 67.5). Since the skill set and downstream generator are held fixed, these comparisons more directly demonstrate the value of paradigm-guided routing rather than gains from access to additional modeling knowledge.

  • Differences from existing tree-memory approaches.

We also revised Section 4.2 and Appendix A.2 (Table 5) to avoid equating novelty with “dynamic trees.” Unlike OptiTree’s subproblem hierarchy or AlphaOPT’s insight taxonomy, OptRoute uses formulation-family paradigms as both the evolving organization units and the routing decisions for retrieving complete skill leaves. We recognize that prior work also updates hierarchical memories; the novelty claim is specifically about this formulation-aware organization–retrieval coupling.

3. W2/Q3 — Repeated runs, construction samples, and insertion order

W2: The analysis does not yet establish the robustness or scope of the advantage. The macro-average improvement over the strongest agentic baseline is 3.0 percentage points with DeepSeek but only 0.9 with DouBao, and the main tables provide no uncertainty estimates. Q3: How stable are the gains across different construction samples and insertion orders? Uncertainty estimates would help assess the smaller improvements, particularly with DouBao.

Thank you for emphasizing the importance of robustness across runs, construction samples, and insertion orders.

  • Repeated runs

We added three-run means and standard deviations for the main comparisons (revised Table 2). OptRoute achieves 79.3 ± 0.6 versus OptiTree 75.2 ± 0.7 on DeepSeek, and 79.2 ± 1.0 versus 77.5 ± 0.5 on DouBao. The DouBao advantage is smaller (The previously reported 0.9 was obtained from a single run. Across multiple independent runs, our method consistently outperformed OptiTree, with a substantial improvement in one run) .

  • construction samples and insertion order

We also added construction sensitivity in Appendix C.4 (Table 8):

VariationMacro Pass@1 across the three settings
Construction sample seed79.2 / 81.3 / 78.4
Skill insertion order79.2 / 79.8 / 75.0

The construction-sample results show a 2.9-point range. More importantly, we acknowledge a 4.8-point insertion-order range, including one weaker order. Inspection of that run suggests that some formulation-distinguishing skills were skipped as redundant, leading to retrieval of continuous-variable guidance where integer formulations were needed; MAMO-Complex fell to 59.5. This is a real limitation of sequential, LLM-guided memory construction. At the same time, the other tested order reaches 79.8. The revised Appendix C.4 reports these outcomes rather than hiding the adverse case.

4. W2/Q4 — Matched-skill scaling and retrieval-error analysis

W2: Beyond one illustrative case, there is little systematic analysis of retrieval failures, routing errors, or which formulation distinctions benefit most from the hierarchy. This limits the insight beyond showing an aggregate performance improvement. Q4: Could you provide systematic retrieval-error analysis and comparisons at each memory scale using the same skills This would clarify when hierarchical routing helps and whether the scaling gains come from better organization or additional knowledge.

Thank you for requesting a more controlled scaling comparison and clearer analysis of retrieval failures.

  • Comparisons at each memory scale using the same skills

We expanded the memory-scaling study in revised Section 3.4 (Figure 3). At each construction scale, the retrieval methods have access to the same leaf-skill set and use the same downstream generator:

Construction casesOptRouteFlat LLMDense
10076.374.776.8
20078.176.275.9
40078.373.872.4
80079.276.976.6

OptRoute is not best at the smallest scale; its advantage emerges at the larger tested scales, but OptRoute improves from 76.3% to 79.2% macro accuracy as memory grows. In contrast, Flat LLM and Dense show no consistent improvement despite access to the same growing skill sets. This suggests that the gains are not simply due to more knowledge; effective skill organization and retrieval also play an important role.

  • Error analysis

We also expanded Appendix C.5 (Error Analysis and Limitations) to distinguish three observed failure sources: missing formulation-family coverage, confusion between similar skills with different applicability conditions, and incorrect synthesis/code generation even after relevant retrieval.

The benchmarks provide neither gold retrieval labels nor standardized intermediate modeling traces, making these failure sources difficult to disentangle. For example, the LLM may sometimes recover missing fine-grained formulation details on its own, whereas similar omissions may lead to incorrect or unbounded models. We therefore provide a qualitative analysis rather than a fully quantified error decomposition.

Besides, revised Appendix C.3 (Figures 6–8) gives concrete examples of content refinement, restructuring, and cross-branch synthesis. We acknowledge that a fully quantified retrieval-versus-generation error decomposition remains open.

In summary, the revision directly addresses the reviewer’s two most actionable requests: (i) separate content from structure, and (ii) compare against LLM-reranked retrieval with matched skills and budget. We have also added repeated runs, sample/order sensitivity, matched-skill scaling, and a candid discussion of remaining failure modes.

We sincerely thank the reviewer for these valuable suggestions. If our responses have addressed your concerns, we would greatly appreciate your reconsideration of the score.

Reviewer_nJpLOpenReview ↗
Rating
2
Confidence
2
Informativeness
2
Presentation
3
Soundness
2
Read review
Paper summary

OptRoute organizes reusable optimization-modeling skills into an evolving tree, using the same hierarchy to guide retrieval. It updates both skill content and tree structure as new examples arrive. Across eight benchmarks and two backbones, it improves modeling accuracy, reaching 79.2% macro-average Pass@1 with DeepSeek-V4.

Critical strengths
  • The motivation is practically relevant. Similar problem descriptions can require different formulations, so organizing skills around modeling assumptions could make retrieval more useful than surface-level similarity alone.

  • The fixed-skill retrieval comparisons provide meaningful evidence. Using identical leaf skills and downstream generation, OptRoute outperforms several retrieval alternatives. Results across two backbones further support its practical usefulness.

Critical weaknesses
  • The paper convincingly explains why semantically similar problems may require different formulations. However, this motivates retrieving skills according to their modeling assumptions; the need to continually evolve both skill content and tree structure is less clear. Structural updates rely on LLM judgments about how skills should be organized, without directly using evidence of retrieval failures. A more coherent hierarchy may help, but the paper should better explain why its evolution decisions are aligned with downstream modeling success. This matters because the evolving organization is the central contribution.

  • Optimization problems often combine requirements such as scheduling, inventory balance, and resource capacity, which cut across a single hierarchy. OptRoute supports multiple retrieved skills and subsequent synthesis, but each skill still occupies one location, and routing determines which combinations become available. The design therefore needs a clearer explanation of how it preserves useful connections across branches. Beyond retrieving a suitable existing formulation, an important question is whether the memory helps construct new formulations by combining knowledge in unfamiliar ways. This distinction affects the strength of the claim about generalizable skill reuse.

Questions for authors

See weekness above

1 comment

Response to Reviewer nJpL

1. W1 — Why evolve both content and structure, and why should LLM-guided evolution help downstream modeling?

W1: ...the need to continually evolve both skill content and tree structure is less clear. Structural updates rely on LLM judgments about how skills should be organized, without directly using evidence of retrieval failures. A more coherent hierarchy may help, but the paper should better explain why its evolution decisions are aligned with downstream modeling success. This matters because the evolving organization is the central contribution.

Thank you for this insightful question about the motivation and evidence for joint evolution.

  • Why evolve both content and structure.

We agree with the distinction in the review: formulation-aware retrieval is well motivated, but the benefit of continuously evolving the memory needs separate evidence. Our mechanism addresses two different sources of error. Content updates retain or enrich transferable formulation details, whereas structural updates separate formulation families so that routing can distinguish otherwise similar skills.

To address this concern, we added an ablation experiment. It separately evaluates the contributions of content and structural updates to memory evolution. Specifically, in Section 3.3 and Table 3, we isolate their respective effects using the same set of validated construction skills and the same downstream pipeline.

ConfigurationMacro Pass@1
Structure-only (frozen skill content)74.8 ± 0.9
Content-only (fixed hierarchy)75.6 ± 1.0
Joint content + structure (OptRoute)79.3 ± 0.6

Neither component alone reproduces the full result. In particular, on MAMO-Complex, the full method reaches 88.0, versus 65.5 (structure-only) and 76.6 (content-only). This provides downstream evidence for the joint mechanism under the tested setup.

  • Why evolution decisions are aligned with modeling success.

To further illustrate these effects, the revised Appendix C.3, Figures 6–7 illustrate, respectively, skill refinement and structural reorganization on concrete modeling cases.

  • Why not directly using evidence of retrieval failures

We acknowledge that our updates are formulation-driven, not retrieval-failure-driven: the construction procedure uses validated source formulations and LLM judgments about skill distinctions, rather than a repeated retrieve–solve–diagnose loop.

Failure-driven optimization could provide a more direct signal but would require additional solver calls and iterative memory revision. We now present this as a scope limitation. The paper’s claim is that formulation-aware joint evolution is useful without such a feedback loop, as supported by the controlled ablation.

2. W2 — Cross-branch connections and composition of unfamiliar formulations

W2: ...but each skill still occupies one location, and routing determines which combinations become available. The design therefore needs a clearer explanation of how it preserves useful connections across branches. Beyond retrieving a suitable existing formulation, an important question is whether the memory helps construct new formulations by combining knowledge in unfamiliar ways. This distinction affects the strength of the claim about generalizable skill reuse.

Thank you for highlighting the important issue of cross-branch skill composition.

  • Cross-branch retrieval mechanism.

A skill has a single storage location, but a query is not restricted to one branch. As clarified in Section 2.3, routing may traverse multiple relevant paradigm branches; the union of their terminal-region skills forms the candidate pool, and selected complementary skills are synthesized into one modeling plan. This preserves access to cross-branch combinations at retrieval time without duplicating each skill in the tree.

  • Case study of cross-branch composition.

We added a concrete example in Appendix C.3, Figure 8: a truck-allocation problem combines demand/resource bounds with pairwise balance requirements. OptRoute retrieves complementary skills from different branches and synthesizes a formulation satisfying the combined requirements, yielding the reference optimal cost of 450. Therefore, the mechanism supports multi-skill composition, while systematic evaluation on deliberately unseen combinations remains an important future direction.

We thank the reviewer for identifying where the causal motivation and the scope of our generalization claim needed to be more precise. We also reviewed and updated the relevant discussions. If our responses have addressed your concerns, we would greatly appreciate your reconsideration of the score.

Reviewer_PShkOpenReview ↗
Rating
2
Confidence
3
Informativeness
2
Presentation
2
Soundness
2
Read review
Paper summary

This paper studies how to organize a growing skill memory for LLM-based optimization modeling. It proposes OptRoute, an evolving tree framework that couples skill organization and retrieval. Within a shared tree, paradigm nodes organize related formulation families and support hierarchical routing, while leaf nodes store complete, reusable formulation skills.

Critical strengths
  1. The paper treats skill organization itself as the research object, rather than proposing yet another agentic workflow. The observation that effective reuse depends not only on what knowledge is stored but on how it is organized for retrieval is well motivated for OR modeling, where semantically similar problems often require different formulations. TThe paradigm/skill separation gives a clean design answer: paradigm nodes route, leaves store complete skills, and the retrieval process has a good design.
  2. The ablation design is the strongest part of the paper. Table 2 and Table 3 substantiates the claim that a hierarchy alone does not account for the gains, which is the paper's central argument and the main basis for my assessment.
Critical weaknesses
  1. Little improvement. On DouBao-Seed-2.0, OptRoute scores 78.9 vs OptiTree 78.0, a 0.9-point gap, and OptiTree is better on MAMO-Easy and NLP4LP; OptSkills also ends up at 72.3 (DouBao base model at 72.4), I don't know whether external knowledge helps.
  2. Prior work already includes hierarchical knowledge trees and incrementally updated tree memories. Besides the engineering design, could the authors summarize the key novelty of the paradigm and tree-structure design itself, and how it differs from these existing hierarchical-memory approaches beyond the coupling of organization and retrieval?
Questions for authors
  1. On DouBao-Seed-2.0, OptRoute scores 78.9 vs OptiTree 78.0, a 0.9-point gap, and OptiTree is better on MAMO-Easy and NLP4LP; OptSkills also ends up at 72.3 (DouBao base model at 72.4), I don't know whether external knowledge helps.
  2. Could you summarize the key innovations of the paradigm and tree-structure design, specifically what distinguishes them from some similar works like AlphaOPT, OptiTree, ...
  3. I would like to understand how some of the hyperparameters were chosen, e.g., K = 3 skills selected for synthesis, b = 4 child paradigms per visited parent, 0.05 in Eq. (10), a child should usually contain no more than about 30 skills. How were these values determined, and what evidence shows they are not cherry-picked
2 comments

Response to Reviewer PShk

1. W1/Q1 — Magnitude and consistency of improvements; usefulness of external knowledge

W1: Little improvement. On DouBao-Seed-2.0, OptRoute scores 78.9 vs OptiTree 78.0, a 0.9-point gap, and OptiTree is better on MAMO-Easy and NLP4LP; OptSkills also ends up at 72.3 (DouBao base model at 72.4), I don't know whether external knowledge helps. Q1: On DouBao-Seed-2.0, OptRoute scores 78.9 vs OptiTree 78.0, a 0.9-point gap, and OptiTree is better on MAMO-Easy and NLP4LP; OptSkills also ends up at 72.3 (DouBao base model at 72.4), I don't know whether external knowledge helps.

Thank you for raising this important concern about the magnitude and consistency of the improvements.

  • Magnitude and consistency of improvements.

We agree that the original 0.9-point DouBao gap alone was not compelling evidence. We therefore repeated the evaluations for the prompt-based methods and now report means over three runs (revised Table 2). The updated macro-average results are:

BackboneZero-shotOptSkillsOptiTreeOptRoute
DeepSeek-V471.2 ± 1.874.8 ± 2.075.2 ± 0.779.3 ± 0.6
DouBao-Seed-2.071.3 ± 1.070.9 ± 1.877.5 ± 0.579.2 ± 1.0

Thus, the updated gains over OptiTree are 4.1 points (DeepSeek) and 1.7 points (DouBao). (The previously reported 0.9 was obtained from a single run. Across multiple independent runs, our method consistently outperformed OptiTree, with a substantial improvement in one run.)

We do not claim uniform superiority on every benchmark: for example, OptiTree remains stronger on DouBao MAMO-Easy and NLP4LP. The overall advantage is more pronounced on formulation-intensive tasks: on DouBao, OptRoute obtains 88.0 versus 82.0 on MAMO-Complex and 50.8 versus 46.8 on OptMATH. These patterns are consistent with the intended benefit of distinguishing closely related formulations.

  • usefulness of external knowledge.

Providing additional knowledge can improve the modeling performance of base LLMs, as both OptiTree and AlphaOpt outperform zero-shot prompting on both backbones. However, external knowledge is not automatically beneficial: OptSkills performs close to, or even below, zero-shot prompting on DouBao. This also highlights the importance of how knowledge is organized and retrieved, which is a key advantage of OptRoute.

We revised Section 3.2 and Table 2 to report the updated results and provide a more balanced interpretation.

2. W2/Q2 — Novelty relative to OptiTree, AlphaOPT, and hierarchical memory

W2: ...could the authors summarize the key novelty of the paradigm and tree-structure design itself, and how it differs from these existing hierarchical-memory approaches beyond the coupling of organization and retrieval? Q2: Could you summarize the key innovations of the paradigm and tree-structure design, specifically what distinguishes them from some similar works like AlphaOPT, OptiTree, ..

Thank you for asking us to clarify the novelty of OptRoute relative to existing hierarchical-memory methods.

We agree that using a tree or updating a hierarchy is not itself new, and we have sharpened the claim accordingly. The contribution is the formulation-aware coupling of memory construction and retrieval for OR modeling:

  1. Representation: paradigm nodes express distinctions between formulation families (e.g., variable domains and constraint structures), while leaf nodes retain complete, reusable solver-ready skills.
  2. Evolution: incoming validated skills can trigger local insertion, enrichment, or consolidation and, where formulation distinctions warrant it, restructuring of the paradigm regions.
  3. Retrieval: the same evolved paradigm distinctions guide multi-branch routing to candidate skills, which are synthesized for solver-code generation.

These choices differ in emphasis from OptiTree's subproblem-oriented hierarchy and AlphaOPT's taxonomy of solver-verified insights. The distinction is what the hierarchy represents and how its evolving distinctions are reused as routing decisions. We revised Introduction, Section 2, Section 4.2, and Appendix A.2 (Table 5) to clarify this positioning. We additionally report controlled retrieval comparisons and content/structure ablations (revised Tables 3–4) to support the practical value of this design rather than relying only on a conceptual novelty claim.

3. Q3 — Hyperparameter selection

Q3: I would like to understand how some of the hyperparameters were chosen, e.g., K = 3 skills selected for synthesis, b = 4 child paradigms per visited parent, 0.05 in Eq. (10), a child should usually contain no more than about 30 skills. How were these values determined, and what evidence shows they are not cherry-picked.

Thank you for your careful questions about the hyperparameter choices.

We added sensitivity results in Appendix C.4 (Table 8), keeping other settings fixed:

ParameterTested settings and macro accuracy
Retrieved-skill limit KK=2: 77.5; K=3: 79.2; K=4: 78.2
Branch limit bb=3: 78.6; b=4: 79.2; b=5: 77.1
  • The default K=3 and b=4 perform best among the tested settings, but performance is not insensitive to every choice (notably b=5). Both are upper bounds, not mandatory counts; routing usually selects only one or two child paradigms at a visited node.
  • The 0.05 objective tolerance follows the evaluation convention used in the compared implementations, including OptiTree, OptSkills, and AlphaOPT.
  • the approximately 30 skills per region is a practical review-size guideline (roughly 9,000–10,000 tokens at about 300 tokens per skill, which keeps structural review manageable and reduces errors when parsing the LLM's output), not a theoretically optimized threshold.

We appreciate the reviewer’s consideration of the revised evidence and hope these clarifications address the main concerns. If our responses have addressed your concerns, we would greatly appreciate your reconsideration of the score.

From OpenReview, CC BY 4.0. Authors as signed there; shown unedited.

Related papers

Loading the map…

Discussion (0)

Sign in to comment.