Pith. sign in

REVIEW 4 major objections 6 minor 3 cited by

Brownian Bridge Augmented Surrogate Simulation and Injection Planning for Geological CO$_2$ Storage

T0 review · 4 major / 6 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 A Brownian bridge augmentation for surrogate simulation and injection planning consistently improves reservoir-state prediction accuracy and storage performance in geological CO2 storage while keeping computational overhead low.

desk verdict Interesting Brownian-bridge regularization for surrogate simulation, but the planning evaluation is too self-referential and Eq. (7) has a unit mismatch; the paper deserves review but needs fixes. read the letter →

arxiv 2505.18204 v1 pith:ASU3CKYU submitted 2025-05-21 cs.RO

Haoyue Bai , Guodong Chen , Wangyang Ying , Xinyuan Wang , Nanxu Gong , Sixun Dong , Giulia Pedrielli , Haoyu Wang
show 2 more authors
Haifeng Chen Yanjie Fu
This is my paper
classification cs.RO
keywords geologicalCO2storageBrownianbridgesurrogatesimulationinjectionplanninggoal-conditionedtemporalsmoothnessperformanceindexreservoirengineering
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 tackles adaptive injection planning for geological CO2 storage, where operators must choose injection rates and well pressures over a long lifecycle to store CO2 safely and efficiently. The authors claim that deep Brownian bridges, stochastic processes pinned at a start and an end point, capture two properties GCS needs: smooth state transitions and goal-directed, time-constrained trajectories. They use one learned Brownian bridge to interpolate next reservoir states as auxiliary supervision for a surrogate simulator, and another to generate utility trajectories from the current utility toward a target utility that guide the injection planner. In experiments on reservoir datasets built with a high-fidelity simulator, the framework improves surrogate prediction accuracy across five baseline architectures and raises a storage performance index above reinforcement learning, POMDP, and evolutionary baselines, while keeping training time comparable. If the claim holds, it offers a low-overhead way to make learned surrogates and planners respect temporal continuity and goal targeting in physical systems.

What carries the argument

The load-bearing object is the deep Brownian bridge in latent space: an encoder maps a reservoir state or utility value to an embedding, a generator interpolates between start and end embeddings by $((1 - t/T)z_0 + (t/T)z_T)$, and a decoder reconstructs the sequence, with contrastive and reconstruction losses training the bridges. The state bridge supplies the surrogate simulator with a smooth next-state embedding target, and the utility bridge supplies the injection planner with goal-time-conditioned guidance, so the same mathematical object carries both the smoothness regularization and the goal-directed planning.

What would settle it

Run the injection plans produced by this framework and by a strong baseline such as POMDP through a high-fidelity numerical simulator on held-out reservoir models that were not used in training, and compare the resulting Storage Performance Index and pressure variance; if the bridge-guided plans do not beat the baselines on those unseen physics, the central claim fails. A more direct check is to test whether plans targeting the historical maximum $r^*$ are consistently dominated by plans targeting a higher physically feasible target when one exists.

Watch

Extended reading notes

Core claim

The central claim is that a Brownian bridge, a stochastic process conditioned to start at one point and end at another, can be lifted into a learned latent space and used twice in a CO2 storage pipeline. First, the bridge's linear interpolation between the latent embeddings of consecutive reservoir states acts as an auxiliary supervision signal: training the surrogate simulator on the Brownian embedding of the next state rather than only on the raw observation smooths the predicted transition and improves utility estimation. Second, a utility bridge generates a goal-time-conditioned trajectory from the recent utility toward a predefined target $r^*$, and the injection planning model conditions on that trajectory when producing an injection plan, with the objective of bringing the surrogate-estimated utility in line with the bridge's desired utility. The paper reports that these two bridges consistently reduce surrogate prediction error and improve a storage performance index, which rewards injection, penalizes production, and penalizes pressure variance, across homogeneous and heterogeneous reservoir datasets.

Load-bearing premise

The planning target is the best storage utility found in the historical training data, and the quality of a plan is judged mostly by the same surrogate used to train the planner; if the surrogate rewards plans that do not transfer to real reservoir physics, the reported performance gains could be artifacts of surrogate exploitation.

Editorial extensions

If this is right

  • Surrogate simulators trained with Brownian next-state interpolation reduce storage-utility prediction error across five baseline architectures, with the largest reported drop being 62.96% in the heterogeneous H-COM_1 scenario.
  • Injection planners conditioned on Brownian utility trajectories achieve a higher Storage Performance Index than random, reinforcement learning, POMDP, and evolutionary baselines on all ten datasets tested.
  • The added Brownian supervision costs little extra training time because interpolation happens in a low-dimensional latent space, so the framework remains practical for real-time planning.
  • Plans produced by the framework yield smoother reservoir pressure evolution and a more uniform CO2 distribution when rolled out in a high-fidelity numerical simulator, consistent with safer storage.

Reading between the lines

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

  • Because the planning target $r^*$ is set to the empirical maximum in historical data, the framework is implicitly bounded by the best past performance; targeting a physically feasible but unobserved optimum could lead to further gains, a testable extension.
  • The Brownian bridge is used only as a supervision signal, not as a calibrated uncertainty model; using the bridge's variance $t(T-t)/T$ to output confidence intervals for next states could support risk-aware injection decisions, which the paper does not claim.
  • The smoothness regularization likely helps most under data scarcity, since a strong prior over transition shape compensates for few trajectories; ablations across the 50-trajectory datasets would test this directly.
  • The single numerical-simulator check is limited to one geological setting, so a blind multi-scenario transfer test would tell whether the surrogate gains translate to genuinely better field operations.
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.

Referee Report

4 major / 6 minor

Summary. The manuscript proposes a three-stage Brownian bridge–augmented framework for geological CO2 storage: (i) learning deep Brownian bridge embeddings of reservoir state and storage utility trajectories using contrastive and reconstruction losses; (ii) regularizing a surrogate simulator by supervising its predicted next-state Brownian embedding against a Brownian interpolation; and (iii) training an injection planner with a Brownian utility trajectory from the recent utility to a target utility r*. The surrogate predicts both storage utility and a latent next reservoir state, while the planner is evaluated by a surrogate-computed Storage Performance Index and, in one scenario, by qualitative ECLIPSE comparisons. Experiments on synthetic ECLIPSE datasets in two settings, H-WAG and H-COM, report consistent MSE improvements for five surrogate baselines and higher SPI than random, SAC, POMDP, and NSGA-II baselines.

Significance. Conditional on the planning evaluation being made independent, the paper offers a thoughtful way to inject temporal smoothness and goal conditioning into surrogate-based GCS optimization. Its strengths include a clearly described three-stage architecture, broad baseline coverage, an ablation study, and computational cost comparisons. The Brownian-bridge auxiliary supervision idea is simple and potentially transferable to other physics-constrained surrogate modeling tasks. However, the reported planning gains are not yet supported by a closed-loop-free assessment, and the objective in Eq. (7) is underspecified; these issues are central to the paper's second claimed contribution.

major comments (4)
  1. [Section 3.4, Eq. (7)] The planning loss compares the surrogate's estimated storage utility S(o_t, s_hat_t) with the raw Brownian generator output Br_g(r_{t-1}, r*, t'), which lives in the utility latent space, while S outputs a physical utility value. The text states that the utility Brownian decoder can decode the embedding into a desired utility, but Eq. (7) omits Br_d. As written, the objective trains the planner to match an interpolated latent point rather than a decoded utility, and the relation between matching that point and achieving high storage utility is unverified. Please either apply the decoder in Eq. (7) or define a utility-space Brownian generator, and report the utility decoder's reconstruction error to show that interpolated latents correspond to meaningful utility values.
  2. [Section 4.3, Table 2] The SPI values in Table 2 are computed from the same surrogate S that is used to train the planner, so the reported improvements may partly reflect surrogate exploitation rather than better physical injection plans. Only one scenario is checked in ECLIPSE (Figure 3), and that check is qualitative (CDF, pressure plots, and 3D CO2 distributions), with no ECLIPSE-evaluated SPI numbers for any method. Please report ECLIPSE-evaluated SPI for all methods on at least one scenario, or otherwise demonstrate that the surrogate reward transfers to the numerical simulator.
  3. [Section 3.4 and Section 4.3, Table 2] The near-constant SPI values for Ours across all ten datasets (1.333 to 1.398), despite widely varying baseline performance, suggest that the planner may be dominated by the fixed target r* rather than adapting to each test condition. Since r* is defined as the empirical maximum from the historical data, and the Brownian bridge is anchored at that value, the objective may encode an over-optimistic or infeasible goal for new lifecycles. Please analyze the feasibility of the target, report sensitivity to r*, and describe how r* is chosen (e.g., validation-based selection) rather than using the maximum of the same training data.
  4. [Section 4.1 and Tables 1-2] All results are reported as averages over five runs, but no standard deviations, confidence intervals, or significance tests are provided for the MSE or SPI comparisons. Several improvements in Table 1 are small (e.g., 2.5% to 6%) and could easily fall within run-to-run variation. Please add error bars or confidence intervals and, where appropriate, paired statistical tests across repeated training runs for both the simulation-fidelity and planning comparisons.
minor comments (6)
  1. [Section 3.2, Eq. (3)] The notation for the Brownian decoder is inconsistent (B_d versus B^o_d), and the contrastive objective is printed in a way that is hard to parse; please rewrite Eq. (3) with clean indexing and a single notation scheme.
  2. [Section 4.1, Eq. (8)] The SPI formula mixes rates, cumulative volumes, and a pressure standard deviation without normalization; please clarify the units and justify the additive combination, or use a dimensionless/normalized variant.
  3. [Figure 1] Several labels in Figure 1 appear as garbled symbols (e.g., the placeholder markers in panel (a)); please regenerate the figure for clarity.
  4. [Figure 3] The pressure units in Figure 3 are labeled in Pa while Section 4.1 states initial reservoir pressure in psi; please make the units consistent across the text and figures.
  5. [Section 4.1] The hardware description '1 way RTX A6000 and 48GB of RAM' is unclear; please specify the CPU model, memory configuration, and GPU setup precisely.
  6. [Section 3.2 and 4.4] Please provide a table of hyperparameter values (alpha, eta, tau, latent dimensions, and network architectures) for reproducibility; only partial sensitivity plots are currently given.

Circularity Check

2 steps flagged · score 6.0 of 10

The planning objective Eq. (7) trains the planner to match a Brownian latent interpolation (decoder omitted) from the same historical data, and Table 2's SPI is computed from the same surrogate used for training, so the central planning gains are a closed loop rather than an independently validated prediction.

  1. fitted input called prediction [Section 3.4, Eq. (7)]
    "the utility-related Brownian decoder can decode the embedding of the desired utility at the t-th timestep in the latent utility embedding trajectory, into a desired storage utility. ... The injection planning model is optimized by minimizing the gap between the estimated storage utility and the desired storage utility, given by: L_D = ||S(o_t, \hat{s}_t) - B^r_g(r_{t-1}, r^*, t')||^2_2."

    The loss compares S's output, a storage-utility value, with B^r_g(...), a point in the latent space of the utility autoencoder. The paper states the decoder B^r_d should convert the latent target into a utility, but Eq. (7) does not apply it. The Brownian generator and its latent space were themselves learned from the same historical utility trajectories, and r* is 'the empirical maximum from historical data.' Hence the planner is trained to match a latent interpolation that is an artifact of the fitted Brownian autoencoder, not an independently defined high-utility target. The 'goal guidance' is therefore defined by the same fitted representation that generates it, closing the loop without grounding in true storage physics.

  2. fitted input called prediction [Section 4.1, Eq. (8), Table 2]
    "we define a Storage Performance Index (SPI) based on simulator outputs to assess CO2 storage performance of a whole lifecycle. ... These parameters can be obtained from the storage utility. We compute the mean SPI across all lifecycles in the test dataset as the final metric. We compare the SPI achieved by the injection plan when executed in both the surrogate simulator and the numerical simulator."

    The planner is trained by minimizing (7), which uses S(o_t, \hat{s}_t) as the estimated utility; the reported SPI is defined 'based on simulator outputs' and its parameters are 'obtained from the storage utility,' i.e., from S's predictions. Thus the quantitative planning comparison in Table 2 is the output of the very function the planner was optimized against. A planner that matches S's utility estimate is, by construction, likely to score well on S-derived SPI, regardless of true reservoir performance. The only numerical-simulator validation is a single qualitative ECLIPSE visualization (Fig. 3), so the central quantitative 'planning effectiveness' claim is not independently confirmed; it reduces to self-consistency between the planner and the surrogate.

full rationale

The simulation-fidelity leg (Eq. 5) is not circular: it is supervised by ground-truth utilities r_t and by the Brownian encoder's embedding of the actual next state z^o_{t+1}, and test-set MSE is a held-out measure. The planning leg, however, closes the loop. Eq. (7) trains the planner to make S(o_t, \hat{s}_t) equal to B^r_g(r_{t-1}, r^*, t'), a latent-space Brownian interpolation whose endpoint r* is the historical empirical maximum. The text says the decoder should convert the latent target into a desired utility, but the loss omits the decoder; even with the decoder, the target is an interpolation inside a fitted autoencoder rather than an independent physical optimum. The reported SPI (Eq. 8) is defined 'based on simulator outputs' with parameters 'obtained from the storage utility'—i.e., from the same surrogate S that generates the training signal. A planner trained against S will, by construction, score well on S-derived SPI; the single qualitative ECLIPSE check on one scenario does not break the loop. No load-bearing self-citation is involved; the issue is the self-referential planning and evaluation design.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The framework rests on modeling choices rather than new physical entities. The main cost is that the Brownian bridge is used as a linear interpolator without its stochastic variance, and the performance depends on hand-tuned hyperparameters and a data-derived utility target.

free parameters (5)
  • alpha (data augmentation noise scaling) = not specified
    Controls how much Gaussian noise is added to sampled subsequences in Section 3.2; Figure 4a shows sensitivity to it.
  • eta (auxiliary loss weight) = around 1e-3 chosen per Figure 4b
    Balances utility loss and Brownian next-state embedding loss in Eq. (5); the paper reports a U-shaped sensitivity curve and picks the minimum.
  • tau (contrastive temperature) = not specified
    Temperature in contrastive loss Eq. (3), chosen by hand.
  • storage utility target r* = empirical maximum of historical training utilities
    Anchors the planning trajectory in Eq. (6)-(7); the choice of goal is data-derived and affects plan quality.
  • latent dimension of Brownian embedding = not specified
    Dimensionality of the learned encoder/decoder latent space, set by the authors.
assumptions (4)
  • ad hoc to paper GCS reservoir state and storage utility trajectories can be represented in a latent space where Brownian bridge (linear interpolation) is a meaningful and useful prior.
    Section 3.2, Eq. (2) replaces the stochastic bridge by its mean; no evidence that linear interpolation in this latent space reflects physical transitions.
  • domain assumption The surrogate simulator is differentiable and accurate enough that training the injection planner against it improves real injection plans.
    Section 3.4 and 4.3 use S(o_t, s_hat_t) as the training signal and SPI evaluation; this assumes no simulator exploitation.
  • domain assumption ECLIPSE 2016 output constitutes ground truth for reservoir states and storage utilities.
    Section 4.1 uses ECLIPSE-generated trajectories as the only source of data.
  • domain assumption Adding Gaussian noise to sampled subsequences preserves the physics-relevant dynamics of GCS trajectories.
    Section 3.2, data augmentation step; no physical justification beyond increasing diversity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Brownian Bridge Augmented Surrogate Simulation and Injection Planning for Geological CO$_2$ Storage." pith.science (2026). https://pith.science/paper/ASU3CKYU

@misc{pith2026250518204,
  author       = {Pith},
  title        = {Pith review of: Brownian Bridge Augmented Surrogate Simulation and Injection Planning for Geological CO$_2$ Storage},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ASU3CKYU}},
  note         = {Machine review of arXiv:2505.18204}
}
read the original abstract

Geological CO2 storage (GCS) involves injecting captured CO2 into deep subsurface formations to support climate goals. The effective management of GCS relies on adaptive injection planning to dynamically control injection rates and well pressures to balance both storage safety and efficiency. Prior literature, including numerical optimization methods and surrogate-optimization methods, is limited by real-world GCS requirements of smooth state transitions and goal-directed planning within limited time. To address these limitations, we propose a Brownian Bridge-augmented framework for surrogate simulation and injection planning in GCS and develop two insights: (i) Brownian bridge as a smooth state regularizer for better surrogate simulation; (ii) Brownian bridge as goal-time-conditioned planning guidance for improved injection planning. Our method has three stages: (i) learning deep Brownian bridge representations with contrastive and reconstructive losses from historical reservoir and utility trajectories, (ii) incorporating Brownian bridge-based next state interpolation for simulator regularization, and (iii) guiding injection planning with Brownian utility-conditioned trajectories to generate high-quality injection plans. Experimental results across multiple datasets collected from diverse GCS settings demonstrate that our framework consistently improves simulation fidelity and planning effectiveness while maintaining low computational overhead.

Figures

Figures reproduced from arXiv: 2505.18204 by the authors.

Figure 1
Figure 1. Problem and Technique Background 2 [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Framework 2 incorporates the smooth interpolation of the next reservoir state in Brownian space to regularize surrogate simulator learning. Step 3 leverages the goal-conditioned pursuing ability of Brownian bridges to guide long-term and forward-thinking injection planning. 3.2 Deep Brownian Bridge: GCS Lifecycle State and Utility Dynamics Modeling as Interpolation in Brownian Embedding Space Given a start point and… view at source ↗
Figure 3
Figure 3. Pressure changes and underground CO2 storage distribution This experiment evaluates the effectiveness of our injection planning model in improving CO2 storage per￾formance. We use the SPI to quantify the injection plan across the whole lifecycle and further validate the results by visualizing the cumulative distribution function (CDF) of CO2, pressure evolution, and spatial CO2 distribution using high-fidelity numer… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Investigation of Proposed Method decision-phase ablation results. Across both all H-WAG scenarios, adding bridge-guided trajectory planning significantly improves SPI scores, confirming that trajectory-aware decision guidance is critical for achieving a high-quality in…

Discussion (0). Sign in to comment.

Forward citations

Cited by 3 Pith papers

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

  1. Supply Chain Optimization via Generative Simulation and Iterative Decision Policies

    cs.AI 2025-07 reject novelty 5.0 of 10

    Sim-to-Dec, an autoregressive simulator paired with a dual-loss decision policy, improves simulated timely delivery and profit, but the reported gains are measured inside the paper's own simulator.

  2. Teaching Time Series to See and Speak: Forecasting with Aligned Visual and Textual Perspectives

    cs.LG 2025-06 reject novelty 5.0 of 10

    TimesCLIP aligns image-based and text-based views of the same time series via contrastive learning to improve forecasting accuracy on several benchmarks, but the full multimodal model is not used on two of the six lon...

  3. LLM-ML Teaming: Integrated Symbolic Decoding and Gradient Search for Valid and Stable Generative Feature Transformation

    cs.LG 2025-06 conditional novelty 4.0 of 10

    A product-of-experts decoder that blends a fine-tuned LLM's token probabilities with a gradient-searched sequence decoder produces more valid and stable feature transformations than either alone.

Reference graph

Works this paper leans on

56 extracted references · 46 canonical work pages · cited by 3 Pith papers

  1. [1]

    Gorec: a generative cold-start recommendation framework

    Haoyue Bai, Min Hou, Le Wu, Yonghui Yang, Kun Zhang, Richang Hong, and Meng Wang. Gorec: a generative cold-start recommendation framework. In Proceedings of the 31st ACM international conference on multimedia, pages 1004--1012, 2023

    work page 2023

  2. [2]

    Multimodality invariant learning for multimedia-based new item recommendation

    Haoyue Bai, Le Wu, Min Hou, Miaomiao Cai, Zhuangzhuang He, Yuyang Zhou, Richang Hong, and Meng Wang. Multimodality invariant learning for multimedia-based new item recommendation. In Proceedings of the 47th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 677--686, 2024

    work page 2024

  3. [3]

    Optimization and monitoring of geological carbon storage operations

    David A Cameron. Optimization and monitoring of geological carbon storage operations. Stanford University, 2013

    work page 2013

  4. [4]

    Optimization of pressure management strategies for geological co2 storage using surrogate model-based reinforcement learning

    Jungang Chen, Eduardo Gildin, and Georgy Kompantsev. Optimization of pressure management strategies for geological co2 storage using surrogate model-based reinforcement learning. International Journal of Greenhouse Gas Control, 138: 0 104262, 2024 a

    work page 2024

  5. [5]

    SynTraC: A Synthetic Dataset for Traffic Signal Control from Traffic Monitoring Cameras

    Tiejin Chen, Prithvi Shirke, Bharatesh Chakravarthi, Arpitsinh Vaghela, Longchao Da, Duo Lu, Yezhou Yang, and Hua Wei. Syntrac: A synthetic dataset for traffic signal control from traffic monitoring cameras. arXiv preprint arXiv:2408.09588, 2024 b

    work page Pith review arXiv 2024

  6. [6]

    Computational methods for multiphase flows in porous media

    Zhangxin Chen, Guanren Huan, and Yuanle Ma. Computational methods for multiphase flows in porous media. SIAM, 2006

    2006

  7. [7]

    A POMDP Model for Safe Geological Carbon Sequestration

    Anthony Corso, Yizheng Wang, Markus Zechner, Jef K. Caers, and Mykel J. Kochenderfer. A pomdp model for safe geological carbon sequestration. ArXiv, abs/2212.00669, 2022. URL https://api.semanticscholar.org/CorpusID:254125236

    work page Pith review arXiv 2022

  8. [8]

    Cityflower: An efficient and realistic traffic simulator with embedded machine learning models

    Longchao Da, Chen Chu, Weinan Zhang, and Hua Wei. Cityflower: An efficient and realistic traffic simulator with embedded machine learning models. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pages 368--373. Springer, 2024

    work page 2024

Show all 56 references
  1. [9]

    An encoder-decoder convlstm surrogate model for simulating geological co2 sequestration with dynamic well controls

    Zhao Feng, Zeeshan Tariq, Xianda Shen, Bicheng Yan, Xuhai Tang, and Fengshou Zhang. An encoder-decoder convlstm surrogate model for simulating geological co2 sequestration with dynamic well controls. Gas Science and Engineering, 125: 0 205314, 2024

    2024

  2. [10]

    Evolutionary large language model for automated feature transformation

    Nanxu Gong, Chandan K Reddy, Wangyang Ying, Haifeng Chen, and Yanjie Fu. Evolutionary large language model for automated feature transformation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 16844--16852, 2025 a

    2025

  3. [11]

    Unsupervised feature transformation via in-context generation, generator-critic llm agents, and duet-play teaming

    Nanxu Gong, Xinyuan Wang, Wangyang Ying, Haoyue Bai, Sixun Dong, Haifeng Chen, and Yanjie Fu. Unsupervised feature transformation via in-context generation, generator-critic llm agents, and duet-play teaming. arXiv preprint arXiv:2504.21304, 2025 b

    2025 arXiv

  4. [12]

    Neuro-symbolic embedding for short and effective feature selection via autoregressive generation

    Nanxu Gong, Wangyang Ying, Dongjie Wang, and Yanjie Fu. Neuro-symbolic embedding for short and effective feature selection via autoregressive generation. ACM Transactions on Intelligent Systems and Technology, 16 0 (2): 0 1--21, 2025 c

    2025

  5. [13]

    Neeraj Gupta, Mark Kelley, Matt Place, Lydia Cumming, Sanjay Mawalkar, Mishra Srikanta, Autumn Haagsma, Robert Mannes, and Rick Pardini. Lessons learned from co2 injection, monitoring, and modeling across a diverse portfolio of depleted closed carbonate reef oil fields--the mi...

    2017

  6. [14]

    Double correction framework for denoising recommendation

    Zhuangzhuang He, Yifan Wang, Yonghui Yang, Peijie Sun, Le Wu, Haoyue Bai, Jinqi Gong, Richang Hong, and Min Zhang. Double correction framework for denoising recommendation. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 1062--1072, 2024

    2024

  7. [15]

    Carbon capture, utilization, and storage in saline aquifers: subsurface policies, development plans, well control strategies and optimization approaches—a review

    Ismail Ismail and Vassilis Gaganis. Carbon capture, utilization, and storage in saline aquifers: subsurface policies, development plans, well control strategies and optimization approaches—a review. Clean Technologies, 5 0 (2): 0 609--637, 2023

    2023

  8. [16]

    Impact of relative permeability hysteresis on geological co2 storage

    Ruben Juanes, EJ Spiteri, FM Orr Jr, and MJ Blunt. Impact of relative permeability hysteresis on geological co2 storage. Water resources research, 42 0 (12), 2006

    2006

  9. [17]

    Sehf: A summary-enhanced hierarchical framework for financial report sentiment analysis

    Haozhou Li, Qinke Peng, Xinyuan Wang, Xu Mou, and Yonghao Wang. Sehf: A summary-enhanced hierarchical framework for financial report sentiment analysis. IEEE Transactions on Computational Social Systems, 11 0 (3): 0 4087--4101, 2023

    2023

  10. [18]

    Sade: A speaker-aware dual encoding model based on diagbert for medical triage and pre-diagnosis

    Haozhou Li, Xinyuan Wang, Hongkai Du, Wentong Sun, and Qinke Peng. Sade: A speaker-aware dual encoding model based on diagbert for medical triage and pre-diagnosis. In ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 127...

    2024

  11. [19]

    Multi-objective optimization of co2 ejector by combined significant variables recognition, ann surrogate model and multi-objective genetic algorithm

    Guangdi Liu, Liang Pu, Hongxia Zhao, Zhuang Chen, and Guangpeng Li. Multi-objective optimization of co2 ejector by combined significant variables recognition, ann surrogate model and multi-objective genetic algorithm. Energy, 295: 0 131010, 2024 a

    2024

  12. [20]

    Multi-objective optimization of co2 ejector by combined significant variables recognition, ann surrogate model and multi-objective genetic algorithm

    Guangdi Liu, Liang Pu, Hongxia Zhao, Zhuang Chen, and Guangpeng Li. Multi-objective optimization of co2 ejector by combined significant variables recognition, ann surrogate model and multi-objective genetic algorithm. Energy, 295: 0 131010, 2024 b

    2024

  13. [21]

    Pth and the regulation of mesenchymal cells within the bone marrow niche

    Hanghang Liu, Linyi Liu, and Clifford J Rosen. Pth and the regulation of mesenchymal cells within the bone marrow niche. Cells, 13 0 (5): 0 406, 2024 c

    2024

  14. [22]

    Edta enhances stromal cell--derived factor 1 --induced migration of dental pulp cells by up-regulating chemokine receptor 4 expression

    Linyi Liu, Sha Leng, Junli Yue, Qian Lu, Weizhe Xu, Xiaowei Yi, Dingming Huang, and Lan Zhang. Edta enhances stromal cell--derived factor 1 --induced migration of dental pulp cells by up-regulating chemokine receptor 4 expression. Journal of Endodontics, 45 0 (5): 0 599--605, 2019

    2019

  15. [23]

    Calorie restriction in mice impairs cortical but not trabecular peak bone mass by suppressing bone remodeling

    Linyi Liu, Phuong T Le, J Patrizia Stohn, Hanghang Liu, Wangyang Ying, Roland Baron, and Clifford J Rosen. Calorie restriction in mice impairs cortical but not trabecular peak bone mass by suppressing bone remodeling. Journal of Bone and Mineral Research, 39 0 (8): 0 1188--119...

    2024

  16. [24]

    Numerical simulation and optimization of injection rates and wells placement for carbon dioxide enhanced gas recovery using a genetic algorithm

    Shuyang Liu, Ramesh Agarwal, Baojiang Sun, Bin Wang, Hangyu Li, Jianchun Xu, and Guangming Fu. Numerical simulation and optimization of injection rates and wells placement for carbon dioxide enhanced gas recovery using a genetic algorithm. Journal of Cleaner Production, 280: 0...

    2021

  17. [25]

    A review of intelligent decision-making strategy for geological co2 storage: Insights from reservoir engineering

    Jinjie Mao and Ashkan Jahanbani Ghahfarokhi. A review of intelligent decision-making strategy for geological co2 storage: Insights from reservoir engineering. Geoenergy Science and Engineering, 240: 0 212951, 2024

    2024

  18. [26]

    Proxy model development for the optimization of water alternating co2 gas for enhanced oil recovery

    D Aqnan Marusaha Matthew, Ashkan Jahanbani Ghahfarokhi, Cuthbert Shang Wui Ng, and Menad Nait Amar. Proxy model development for the optimization of water alternating co2 gas for enhanced oil recovery. Energies, 16 0 (8): 0 3337, 2023

    2023

  19. [27]

    Deep convolutional encoder-decoder networks for uncertainty quantification of dynamic multiphase flow in heterogeneous media

    Shaoxing Mo, Yinhao Zhu, Nicholas Zabaras, Xiaoqing Shi, and Jichun Wu. Deep convolutional encoder-decoder networks for uncertainty quantification of dynamic multiphase flow in heterogeneous media. Water Resources Research, 55 0 (1): 0 703--728, 2019

    2019

  20. [28]

    Practical closed-loop reservoir management using deep reinforcement learning

    Yusuf Nasir and Louis J Durlofsky. Practical closed-loop reservoir management using deep reinforcement learning. SPE Journal, 28 0 (03): 0 1135--1148, 2023

    2023

  21. [29]

    Fast evaluation of well placements in heterogeneous reservoir models using machine learning

    Azor Nwachukwu, Hoonyoung Jeong, Michael Pyrcz, and Larry W Lake. Fast evaluation of well placements in heterogeneous reservoir models using machine learning. Journal of Petroleum Science and Engineering, 163: 0 463--475, 2018

    2018

  22. [30]

    Artificial neural network based surrogate modelling for multi-objective optimisation of geological co2 storage operations

    Indranil Pan, Masoud Babaei, Anna Korre, and Sevket Durucan. Artificial neural network based surrogate modelling for multi-objective optimisation of geological co2 storage operations. Energy Procedia, 63: 0 3483--3491, 2014

    2014

  23. [31]

    Geomechanical modeling of co2 geological storage: A review

    Pengzhi Pan, Zhenhua Wu, Xiating Feng, and Fei Yan. Geomechanical modeling of co2 geological storage: A review. Journal of Rock Mechanics and Geotechnical Engineering, 8 0 (6): 0 936--947, 2016

    2016

  24. [32]

    Continuous martingales and Brownian motion, volume 293

    Daniel Revuz and Marc Yor. Continuous martingales and Brownian motion, volume 293. Springer Science & Business Media, 2013

    2013

  25. [33]

    Co-optimization of carbon dioxide storage and enhanced oil recovery in oil reservoirs using a multi-objective genetic algorithm (nsga-ii)

    Mohammad Amin Safarzadeh and Seyyed Mahdia Motahhari. Co-optimization of carbon dioxide storage and enhanced oil recovery in oil reservoirs using a multi-objective genetic algorithm (nsga-ii). Petroleum Science, 11: 0 460--468, 2014

    2014

  26. [34]

    ECLIPSE Reservoir Simulator, 2016

    Schlumberger . ECLIPSE Reservoir Simulator, 2016. https://www.software.slb.com/products/eclipse

    2016

  27. [35]

    Geomechanical challenges during geological co2 storage: A review

    Youngsoo Song, Sungjun Jun, Yoonsu Na, Kyuhyun Kim, Youngho Jang, and Jihoon Wang. Geomechanical challenges during geological co2 storage: A review. Chemical Engineering Journal, 456: 0 140968, 2023

    2023

  28. [36]

    Optimization of subsurface co2 injection based on neural network surrogate modeling

    Zhuang Sun, Jianping Xu, D Nicolas Espinoza, and Matthew T Balhoff. Optimization of subsurface co2 injection based on neural network surrogate modeling. Computational Geosciences, 25 0 (6): 0 1887--1898, 2021

    2021

  29. [37]

    Graph network surrogate model for subsurface flow optimization

    Haoyu Tang and Louis J Durlofsky. Graph network surrogate model for subsurface flow optimization. Journal of Computational Physics, 512: 0 113132, 2024

    2024

  30. [38]

    Towards data-centric ai: A comprehensive survey of traditional, reinforcement, and generative approaches for tabular data transformation

    Dongjie Wang, Yanyong Huang, Wangyang Ying, Haoyue Bai, Nanxu Gong, Xinyuan Wang, Sixun Dong, Tao Zhe, Kunpeng Liu, Meng Xiao, et al. Towards data-centric ai: A comprehensive survey of traditional, reinforcement, and generative approaches for tabular data transformation. arXiv...

    2025 arXiv

  31. [39]

    Dialogue planning via brownian bridge stochastic process for goal-directed proactive dialogue

    Jian Wang, Dongding Lin, and Wenjie Li. Dialogue planning via brownian bridge stochastic process for goal-directed proactive dialogue. In Findings of the Association for Computational Linguistics: ACL 2023, pages 370--387, Toronto, Canada, July 2023. Association for Computatio...

    2023

  32. [40]

    Wang, Esin Durmus, Noah D

    Rose E. Wang, Esin Durmus, Noah D. Goodman, and Tatsunori Hashimoto. Language modeling via stochastic processes. ArXiv, abs/2203.11370, 2022 a . URL https://api.semanticscholar.org/CorpusID:247597138

    2022 arXiv

  33. [41]

    A deep learning based surrogate model for reservoir dynamic performance prediction

    Sen Wang, Jie Xiang, Xiao Wang, Qihong Feng, Yong Yang, Xiaopeng Cao, and Lei Hou. A deep learning based surrogate model for reservoir dynamic performance prediction. Geoenergy Science and Engineering, 233: 0 212516, 2024 a

    2024

  34. [42]

    Lcmdc: Large-scale chinese medical dialogue corpora for automatic triage and medical consultation

    Xinyuan Wang, Haozhou Li, Dingfang Zheng, and Qinke Peng. Lcmdc: Large-scale chinese medical dialogue corpora for automatic triage and medical consultation. arXiv preprint arXiv:2410.03521, 2024 b

    2024 arXiv

  35. [43]

    Knockoff-guided feature selection via a single pre-trained reinforced agent

    Xinyuan Wang, Dongjie Wang, Wangyang Ying, Rui Xie, Haifeng Chen, and Yanjie Fu. Knockoff-guided feature selection via a single pre-trained reinforced agent. arXiv preprint arXiv:2403.04015, 2024 c

    2024 arXiv

  36. [44]

    Llm-enhanced user-item interactions: Leveraging edge information for optimized recommendations

    Xinyuan Wang, Liang Wu, Liangjie Hong, Hao Liu, and Yanjie Fu. Llm-enhanced user-item interactions: Leveraging edge information for optimized recommendations. arXiv preprint arXiv:2402.09617, 2024 d

    2024 arXiv

  37. [45]

    A successful hybrid deep learning model aiming at promoter identification

    Ying Wang, Qinke Peng, Xu Mou, Xinyuan Wang, Haozhou Li, Tian Han, Zhao Sun, and Xiao Wang. A successful hybrid deep learning model aiming at promoter identification. BMC bioinformatics, 23 0 (Suppl 1): 0 206, 2022 b

    2022

  38. [46]

    Industry-scale co2 flow simulations with model-parallel fourier neural operators

    Philipp A Witte, Russell Hewett, and Ranveer Chandra. Industry-scale co2 flow simulations with model-parallel fourier neural operators. In NeurIPS 2022 workshop tackling climate change with machine learning, 2022

    2022

  39. [47]

    Numerical simulation of co2 disposal by mineral trapping in deep aquifers

    Tianfu Xu, John A Apps, and Karsten Pruess. Numerical simulation of co2 disposal by mineral trapping in deep aquifers. Applied geochemistry, 19 0 (6): 0 917--936, 2004

    2004

  40. [48]

    Self-optimizing feature generation via categorical hashing representation and hierarchical reinforcement crossing

    Wangyang Ying, Dongjie Wang, Kunpeng Liu, Leilei Sun, and Yanjie Fu. Self-optimizing feature generation via categorical hashing representation and hierarchical reinforcement crossing. In 2023 IEEE International Conference on Data Mining (ICDM), pages 748--757. IEEE, 2023

    2023

  41. [49]

    Topology-aware reinforcement feature space reconstruction for graph data

    Wangyang Ying, Haoyue Bai, Kunpeng Liu, and Yanjie Fu. Topology-aware reinforcement feature space reconstruction for graph data. arXiv preprint arXiv:2411.05742, 2024 a

    2024 arXiv

  42. [50]

    Feature selection as deep sequential generative learning

    Wangyang Ying, Dongjie Wang, Haifeng Chen, and Yanjie Fu. Feature selection as deep sequential generative learning. ACM Transactions on Knowledge Discovery from Data, 18 0 (9): 0 1--21, 2024 b

    2024

  43. [51]

    Revolutionizing biomarker discovery: Leveraging generative ai for bio-knowledge-embedded continuous space exploration

    Wangyang Ying, Dongjie Wang, Xuanming Hu, Ji Qiu, Jin Park, and Yanjie Fu. Revolutionizing biomarker discovery: Leveraging generative ai for bio-knowledge-embedded continuous space exploration. In Proceedings of the 33rd ACM International Conference on Information and Knowledg...

    2024

  44. [52]

    Unsupervised generative feature transformation via graph contrastive pre-training and multi-objective fine-tuning

    Wangyang Ying, Dongjie Wang, Xuanming Hu, Yuanchun Zhou, Charu C Aggarwal, and Yanjie Fu. Unsupervised generative feature transformation via graph contrastive pre-training and multi-objective fine-tuning. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery ...

    2024

  45. [53]

    A survey on data-centric ai: Tabular learning from reinforcement learning and generative ai perspective

    Wangyang Ying, Cong Wei, Nanxu Gong, Xinyuan Wang, Haoyue Bai, Arun Vignesh Malarkkan, Sixun Dong, Dongjie Wang, Denghui Zhang, and Yanjie Fu. A survey on data-centric ai: Tabular learning from reinforcement learning and generative ai perspective. arXiv preprint arXiv:2502.08828, 2025

    2025 arXiv

  46. [54]

    Cityflow: A multi-agent reinforcement learning environment for large scale city traffic scenario

    Huichu Zhang, Siyuan Feng, Chang Liu, Yaoyao Ding, Yichen Zhu, Zihan Zhou, Weinan Zhang, Yong Yu, Haiming Jin, and Zhenhui Li. Cityflow: A multi-agent reinforcement learning environment for large scale city traffic scenario. In The world wide web conference, pages 3620--3624, 2019

    2019

  47. [55]

    Training effective deep reinforcement learning agents for real-time life-cycle production optimization

    Kai Zhang, Zhongzheng Wang, Guodong Chen, Liming Zhang, Yongfei Yang, Chuanjin Yao, Jian Wang, and Jun Yao. Training effective deep reinforcement learning agents for real-time life-cycle production optimization. Journal of Petroleum Science and Engineering, 208: 0 109766, 2022

    2022

  48. [56]

    Numerical simulation and optimization of co2 sequestration in saline aquifers

    Zheming Zhang and Ramesh Agarwal. Numerical simulation and optimization of co2 sequestration in saline aquifers. Computers & Fluids, 80: 0 79--87, 2013

    2013

Pith tools

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