Pith. sign in

REVIEW 5 major objections 5 minor 62 references

Physics-Guided Fair Graph Sampling for Water Temperature Prediction in River Networks

T0 review · 5 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The paper argues that in graph neural networks for stream networks, the influence of an upstream neighbor on a node's prediction can be estimated from physical streamflow mixing ratios, and that rebalancing graph aggregation against this…

desk verdict A novel and useful physics-guided fairness mechanism, but the influence derivation has an unsupported constant-gradient assumption and the results lack error bars. read the letter →

arxiv 2412.16523 v1 pith:M3M2CC7E submitted 2024-12-21 cs.LG cs.CYphysics.soc-phstat.ML

Erhu He , Declan Kutscher , Yiqun Xie , Jacob Zwart , Zhe Jiang , Huaxiu Yao , Xiaowei Jia This is my paper · ORCID
classification cs.LGcs.CYphysics.soc-phstat.ML
keywords watertemperaturepredictiongraphneuralnetworksfairnessinmachinelearningriverphysics-guidededgesamplingspatialbiasDelawareBasin
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 claims that the spatial bias a graph neural network introduces when predicting stream water temperature—bias that falls hardest on low-income and low-education communities—can be corrected by making graph aggregation aware of the physical strength of upstream influence rather than only the number or weight of edges. It defines the influence of an upstream stream segment on a downstream one through the streamflow mixing ratio from the heat-transfer equations, turns that influence into an edge modifier that rebalances each node's sampled neighborhood across sensitive groups, and shows on 456 Delaware River Basin segments over 40 years that this lowers fairness disparities and worst-case errors while keeping overall RMSE at the base model's level. A sympathetic reader would take from this that physics-based influence is a usable, precomputable proxy for GNN aggregation influence, and that rebalancing against it is a practical route to environmental justice in learned river models.

What carries the argument

The central object is the physics-based influence estimate $\mathrm{Influence}_t(j,i) \approx \frac{q_{j,t}}{q_{j,t}+q_{i,t}} \|\partial \hat{y}_{j,t}/\partial z_{j,t}\|$, with the gradient-norm factor treated as constant across nodes, and with multi-hop influence formed as the product of successive pairwise ratios along the upstream-to-downstream stream path (Eq. 8). This precomputable quantity replaces gradient-based influence from the GNN itself, avoiding instability from sparse observations and immature aggregation parameters. The edge modifier converts the influence estimates into graph edits: for discrete sensitive groups it rescales edge weights so each group's summed influence matches the largest (Eq. 9), and for continuous attributes it adds candidate neighbors with the lowest influence density, spreading each node's received influence across the attribute range.

What would settle it

Measure the per-node gradient norms $\|\partial \hat{y}_j/\partial z_j\|$ in the trained base GraphSAGE model on the Delaware River Basin; if these norms vary by more than a small factor across nodes, Eq. 7's influence estimates are not the true GNN influences. A direct check is to replace the physics-based influence in the edge modifier with the empirically computed gradients: if fairness outcomes are unchanged, the physical assumption is not doing the work, and if they improve further, the constant-gradient simplification is the bottleneck.

Watch

Extended reading notes

Core claim

The central claim is that the influence of a neighbor node $j$ on node $i$'s prediction is measured by the gradient norm $\|\partial \hat{y}_i/\partial z_j\|$, and that this can be computed from physics instead of backpropagation: the heat-transfer relation for stream temperature gives $\partial y_i/\partial y_j = q_j/(q_j+q_i)$, and the remaining factor $\|\partial \hat{y}_j/\partial z_j\|$ is declared constant across nodes because all nodes share the same output transformation layers. Multi-hop influence is then the product of these pairwise mixing ratios along the stream path, computed from observed streamflow supplemented by PRMS-SNTemp simulations. The paper's edge modifier uses this influence, rather than edge counts or raw adjacency weights, to add or remove sampled neighbors and rescale edge weights so each node receives balanced total influence from each sensitive group, or—for continuous attributes—a balanced influence density over the sensitive-value range. On the Delaware River Basin, this yields the lowest fairness disparity among all tested models ($0.134$ vs $0.206$ for GraphSAGE on income; $0.189$ vs $0.253$ on education) and the best worst-case RMSE across continuous sliding windows, with overall RMSE of $1.774$ matching the base model.

Load-bearing premise

The whole reweighting depends on treating the gradient term $\|\partial \hat{y}_j/\partial z_j\|$ as the same for every node, so that neighbor influence is fully captured by the physical mixing ratio; if that per-node gradient actually varies, the physics-based influence can mis-rank neighbors and the fairness gains may not come from the physics the paper credits.

Editorial extensions

If this is right

  • On the Delaware River Basin, PGFG lowers the income-group fairness disparity from $0.206$ to $0.134$ and the education-group disparity from $0.253$ to $0.189$, while keeping overall RMSE equal to the base model's $1.774$.
  • For continuous sensitive attributes, the method improves worst-case RMSE across sliding windows (e.g., from $5.057$ to $4.456$ at window size $1000$ for income), meaning the hardest-hit locations improve most.
  • The influence measure is precomputed from streamflow before training, so the fairness mechanism adds no gradient-estimation cost or training instability.
  • The method transfers to other GNN backbones: integrating PGFG into an STGNN reduces the income fairness disparity further to $0.096$ while preserving accuracy.
  • The same physics-guided influence recipe is advertised as applicable to any scientific modeling task where the governing PDE yields spatial interaction derivatives $\partial y_i/\partial y_j$.
  • If the constant-gradient assumption fails in practice, the fairness gains may still hold because the mixing ratios act as a spatially smooth proxy for influence; a cheap way to test this is to rerun PGFG with raw $q_j/(q_j+q_i)$ weights and no gradient term at all.
  • The same influence-balancing recipe should transfer to other advection-dominated environmental predictions, such as dissolved oxygen, nutrient load, or flood stage, where a governing PDE gives $\partial y_i/\partial y_j$ and sensitive populations are spatially segregated.
  • Because low-income segments have sparser temperature observations, coupling this method with active data collection at those sites could compound the physics-based reweighting; the paper's own Fig. 1 shows the bias is partly data-driven.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. This paper proposes PGFG, a physics-guided fair graph sampling method for stream water temperature prediction. The authors define a GNN influence measure between nodes, approximate it using physical mixing sensitivities derived from heat advection in river networks, and use the resulting influence values to modify edge sampling and reweight edges so that each node's aggregated influence is balanced across sensitive groups (income, education). Experiments on the Delaware River Basin with a GraphSAGE backbone report improved fairness metrics (e.g., discrete income fairness 0.134 for PGFG vs. 0.206 for GraphSAGE) while keeping RMSE essentially unchanged (1.774 vs. 1.774). The central claim is that PGFG enforces fairness without compromising predictive accuracy.

Significance. The work addresses a timely and socially important problem: spatial bias introduced by GNN aggregation in environmental monitoring, particularly over low-income and low-education communities. The core idea of using physics-derived influence, rather than degree or adjacency weight, to drive fair neighbor selection and weighting is original and potentially applicable beyond river networks to other scientific graphs (e.g., hydrology, climate). A notable strength is that the influence measure is derived from streamflow physics and is not fitted to the temperature target, so the mechanism is not circular. The paper ships code and reports five-seed runs, although without variance. However, the theoretical derivation has load-bearing gaps—the constant-gradient-norm assumption and the multi-hop confluence formula—so the empirical superiority, while promising, is not yet backed by a sound mechanism or by significance testing.

major comments (5)
  1. [Proposed Method, Physical Knowledge Integration, Eq. (7)] The approximation in Eq. (7) is not a chain-rule decomposition of the quantity defined in Eq. (4). Equation (4) defines Influence(j,i) as ||∂ŷ_i/∂z_j||, whereas Eq. (7) replaces this with ||∂y_i/∂y_j · ∂ŷ_j/∂z_j||. The first factor is a physical sensitivity of true temperature at i to true temperature at j, not the GNN's sensitivity of ŷ_i to the hidden state z_j, and the second factor is the output Jacobian at node j, not at node i. The two expressions are different objects, and the substitution requires a justification that is not provided.
  2. [Proposed Method, Physical Knowledge Integration, Eq. (7)] The claim that ||∂ŷ_j,t/∂z_j,t|| is constant across nodes is incorrect for a nonlinear GNN. The Jacobian of the output layers with respect to z_j depends on z_j through activation derivatives and weights, and z_j differs by node and by training epoch. Dropping this term changes the ranking of neighbor influences, so the edge modifier and weight rescaling in Eq. (9) are not justified as implementing the stated GNN influence. The authors should either compute the actual gradient norms (e.g., from a converged base model), provide empirical evidence that their variation is negligible, or redefine the influence measure so that it does not require this assumption.
  3. [Proposed Method, Physical Knowledge Integration, Eq. (8)] The product formula for multi-hop influence is incorrect at confluences. When two upstream branches merge into a downstream node, the derivative of downstream temperature with respect to an upstream branch temperature is q_branch/(q_downstream), where q_downstream includes all inflows; the pairwise ratio q_{j_{m-1}}/(q_{j_{m-1}} + q_{j_m}) uses the wrong denominator unless there is only one upstream branch. Because the graph includes edges between any upstream and downstream segment, many paths pass through confluences, so the precomputed influence values are miscalibrated. A corrected formula should accumulate the total inflow along the path.
  4. [Proposed Method, Physical Knowledge Integration, Eq. (5)-(6)] The mixing model in Eq. (5) writes the denominator as q_i,t + q_j,t even when the numerator sums over multiple upstream neighbors j ∈ N(i). The correct mixing denominator for multiple upstream inflows is q_i,t + Σ_k q_k,t, and the partial derivative ∂y_i/∂y_j is q_j,t/(q_i,t + Σ_k q_k,t). This matters because many river segments in the dataset have multiple upstream segments; the simplified derivative overestimates the influence of each upstream neighbor. The authors should either justify the single-neighbor approximation or use the multi-input derivative.
  5. [Experiments, Implementation Details and Results] The experiments report only averages over five seeds; no standard deviations, confidence intervals, or significance tests are provided. In Table 1, the gap in discrete income fairness between PGFG (0.134) and DSGNN (0.141) is small, and in Table 2 FairFor achieves 0.136, close to PGFG's 0.134. Without variance information, the claim that PGFG outperforms the baselines is not statistically supported. Please report per-seed results or error bars and run paired tests (e.g., Wilcoxon signed-rank) across the five seeds.
minor comments (5)
  1. [Table 1 and Appendix] In Table 1, the entry 'SLGSGNNg' appears to be a typo for 'SLDSGNNg', and the appendix contains the misspelling 'techinical appendix'.
  2. [Figure 2(b)] Figure 2(b) is labeled 'Fair edge sampling', but the sampling procedure for continuous sensitive attributes (Eq. (10)) is explained only in prose; a small pseudo-code box would improve reproducibility and clarity.
  3. [Equation (5)] Equation (5) includes y_i,t-1 in the numerator, but the derivative in Eq. (6) is taken only with respect to upstream y_j,t; the dependence on the previous time step's own temperature is not accounted for in the influence measure, which may matter for slow-moving or large river segments.
  4. [Equation (1)] The fairness metric M_fair uses the mean absolute deviation of group RMSEs from the global RMSE; consider also reporting the standard deviation of group RMSEs, as mean absolute deviation is less commonly used in the fairness literature.
  5. [Code & Data Appendix] The paper states that data pointers will be provided after acceptance; for reproducibility, consider making the preprocessed graph, streamflow simulations, and sensitive-attribute mappings available at publication time.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the physics-based influence and edge modifier are derived from external advection theory and streamflow data, not from the fairness objective or the prediction target.

full rationale

The central fairness mechanism is self-contained in the sense required here. Influence is defined in Eq. 4 as a GNN gradient and then approximated in Eq. 7 by the physical mixing ratio q_j/(q_j+q_i), with the relation Eq. 5 attributed to external heat-transfer literature (Dugdale et al. 2017; Boyd 1996). The approximating factor is computed from observed and PRMS-SNTemp-simulated streamflow, not from the water-temperature observations being predicted, and no parameter is fitted to the fairness metric and then renamed as a prediction. The edge modifier (Eqs. 9-10) uses these precomputed influence values to reweight or resample neighbors; the fairness metrics (Eq. 1 and sliding-window worst-case RMSE) appear only in evaluation, not in the training objective of the modifier. Thus the reported fairness improvements are an empirical outcome, not forced by construction. The main theoretical weakness, that ||dŷ_j/dz_j|| is treated as constant across nodes in Eq. 7, is a questionable approximation for a nonlinear GraphSAGE stack and is a correctness risk, but it is not circularity. Self-citations (e.g., He et al. 2022, 2023a, 2024; Jia et al. 2021b) appear in related work and baseline comparisons but none is load-bearing for the derivation; the physical mixing equation is cited to external sources. Accordingly, no circular step is identified.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The derivation relies on a simplified physics model and an unverified proportionality assumption between physical sensitivity and GNN influence. No new physical entities are introduced. The only hand-chosen values are evaluation thresholds and a GraphSAGE hyperparameter.

free parameters (3)
  • Income group thresholds = 0-50k, 50-100k, >100k
    Chosen for evaluation; changes the fairness metric.
  • Education group threshold = 0.5 proportion college
    Chosen for evaluation.
  • Number of sampled neighbors per node = not reported
    GraphSAGE hyperparameter; affects the edge modifier's candidate pool and results.
assumptions (4)
  • domain assumption The advective mixing relation in Eq. 5 approximates water temperature dynamics in the river network.
    The paper states it comes from 'derivations and approximations from the heat transfer PDE' (Dugdale et al. 2017; Boyd 1996), ignoring other heat fluxes.
  • ad hoc to paper The GNN influence from neighbor j to node i can be approximated by the physical derivative dy_i/dy_j times a node-constant gradient norm.
    Eq. 7 equates ||dy_hat_i/dz_j|| with (q_j/(q_j+q_i)) * ||dy_j/dz_j|| and treats the latter as constant, with no validation.
  • ad hoc to paper Multi-hop influence is the product of per-hop mixing ratios along the stream path (Eq. 8).
    This ignores branching and attenuation beyond pairwise mixing.
  • domain assumption Sensitive attributes from census subcounty averages accurately represent each stream segment's community.
    The paper averages subcounty values over segments; this may not reflect local variations.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-Guided Fair Graph Sampling for Water Temperature Prediction in River Networks." pith.science (2026). https://pith.science/paper/M3M2CC7E

@misc{pith2026241216523,
  author       = {Pith},
  title        = {Pith review of: Physics-Guided Fair Graph Sampling for Water Temperature Prediction in River Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M3M2CC7E}},
  note         = {Machine review of arXiv:2412.16523}
}
read the original abstract

This work introduces a novel graph neural networks (GNNs)-based method to predict stream water temperature and reduce model bias across locations of different income and education levels. Traditional physics-based models often have limited accuracy because they are necessarily approximations of reality. Recently, there has been an increasing interest of using GNNs in modeling complex water dynamics in stream networks. Despite their promise in improving the accuracy, GNNs can bring additional model bias through the aggregation process, where node features are updated by aggregating neighboring nodes. The bias can be especially pronounced when nodes with similar sensitive attributes are frequently connected. We introduce a new method that leverages physical knowledge to represent the node influence in GNNs, and then utilizes physics-based influence to refine the selection and weights over the neighbors. The objective is to facilitate equitable treatment over different sensitive groups in the graph aggregation, which helps reduce spatial bias over locations, especially for those in underprivileged groups. The results on the Delaware River Basin demonstrate the effectiveness of the proposed method in preserving equitable performance across locations in different sensitive groups.

Figures

Figures reproduced from arXiv: 2412.16523 by the authors.

Figure 1
Figure 1. The distribution of RMSE for a GNN model’s pre [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. (a) A diagram of the proposed model. For each node [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (a) and (b). For each sensitive group p (x-axis), we re￾port the absolute distances between RMSE achieved on each group p and the overall performance across all the groups {p|p ∈ P}. Here the low-income and low-education re￾gions have larger distances because they cover a relatively small number of locations (i.e., stream segments) compared to other groups and the models in general make larger er￾rors on these locat… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Continuous fairness comparison amongst PGFG, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The distributions of predictive root mean squared error (RMSE) by (a) GraphSAGE, (b) FairGNN, and (c) the pro [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]
Figure 6
Figure 6. Figure 6: The distributions of predictive root mean squared error (RMSE) by (a) GraphSAGE, (b) FairGNN, and (c) the pro [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 54 canonical work pages

  1. [1]

    gridMET - Climatology Lab

    2021. gridMET - Climatology Lab . http://www.climatologylab.org/gridmet.html

    work page 2021

  2. [2]

    Alasadi, J.; et al. 2019. Toward fairness in face matching algorithms. In Proceedings of the 1st International Workshop on Fairness, Accountability, and Transparency in MultiMedia, 19--25

    work page 2019

  3. [3]

    Boyd, M. S. 1996. Heat source: stream, river and open channel temperature prediction

    work page 1996

  4. [4]

    Graph Neural Networks for Improved El Ni\~no Forecasting

    Cachay, S. R.; Erickson, E.; Bucker, A. F. C.; Pokropek, E.; Potosnak, W.; Osei, S.; and L \"u tjens, B. 2020. Graph Neural Networks for Improved El Ni \ no Forecasting. arXiv preprint arXiv:2012.01598

    work page Pith review arXiv 2020

  5. [5]

    A.; D’Odorico, P.; Laio, F.; and Ridolfi, L

    Carr, J. A.; D’Odorico, P.; Laio, F.; and Ridolfi, L. 2013. Recent history and geography of virtual water trade. PloS one, 8(2): e55825

    work page 2013

  6. [6]

    Chen, S.; Appling, A.; Oliver, S.; Corson-Dosch, H.; Read, J.; Sadler, J.; Zwart, J.; and Jia, X. 2021. Heterogeneous stream-reservoir graph networks with data assimilation. In 2021 IEEE International Conference on Data Mining (ICDM), 1024--1029. IEEE

    work page 2021

  7. [7]

    Chen, S.; et al. 2022. Physics-guided graph meta learning for predicting water temperature and streamflow in stream networks. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2752--2761

    work page 2022

  8. [8]

    Cong, Z.; Shi, B.; Li, S.; Yang, J.; He, Q.; and Pei, J. 2023. FairSample: Training Fair and Accurate Graph Convolutional Neural Networks Efficiently. IEEE Transactions on Knowledge and Data Engineering

    work page 2023

Show all 62 references
  1. [9]

    Dai, E.; et al. 2021. Say no to the discrimination: Learning fair graph neural networks with limited sensitive attribute information. In Proceedings of the 14th ACM International Conference on Web Search and Data Mining, 680--688

    2021

  2. [10]

    Dong, Y.; Liu, N.; Jalaian, B.; and Li, J. 2022. Edits: Modeling and mitigating data bias for graph neural networks. In Proceedings of the ACM Web Conference 2022, 1259--1269

    2022

  3. [11]

    J.; et al

    Dugdale, S. J.; et al. 2017. River temperature modelling: A review of process-based approaches and future directions. Earth-Science Reviews, 175: 97--113

    2017

  4. [12]

    Fan, J.; Bai, J.; Li, Z.; Ortiz-Bobea, A.; and Gomes, C. P. 2022. A GNN-RNN approach for harnessing geospatial and temporal information: application to crop yield prediction. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 36, 11873--11881

    2022

  5. [13]

    Ge, Y.; Zhao, X.; Yu, L.; Paul, S.; Hu, D.; Hsieh, C.-C.; and Zhang, Y. 2022. Toward Pareto efficient fairness-utility trade-off in recommendation through reinforcement learning. In Proceedings of the fifteenth ACM international conference on web search and data mining, 316--324

    2022

  6. [14]

    Hamilton, W.; et al. 2017. Inductive representation learning on large graphs. In Advances in neural information processing systems, 1024--1034

    2017

  7. [15]

    Hardt, M.; et al. 2016. Equality of opportunity in supervised learning. Advances in neural information processing systems, 29

    2016

  8. [16]

    He, E.; Xie, Y.; Jia, X.; Chen, W.; Bao, H.; Zhou, X.; Jiang, Z.; Ghosh, R.; and Ravirathinam, P. 2022. Sailing in the location-based fairness-bias sphere. In Proceedings of the 30th International Conference on Advances in Geographic Information Systems, 1--10

    2022

  9. [17]

    He, E.; Xie, Y.; Liu, L.; Chen, W.; Jin, Z.; and Jia, X. 2023 a . Physics Guided Neural Networks for Time-Aware Fairness: An Application in Crop Yield Prediction. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 37, 14223--14231

    2023

  10. [18]

    He, E.; Xie, Y.; Sun, A.; Zwart, J.; Yang, J.; Jin, Z.; Wang, Y.; Karimi, H.; and Jia, X. 2024. Fair Graph Learning Using Constraint-Aware Priority Adjustment and Graph Masking in River Networks. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, 22087--22095

    2024

  11. [19]

    He, H.; Zhang, Q.; Wang, S.; Yi, K.; Niu, Z.; and Cao, L. 2023 b . Learning informative representation for fairness-aware multivariate time-series forecasting: A group-based perspective. IEEE Transactions on Knowledge and Data Engineering

    2023

  12. [20]

    Hipsey, M.; et al. 2014. GLM-General Lake Model: Model overview and user information

    2014

  13. [21]

    Y.; et al

    Hoekstra, A. Y.; et al. 2012. The water footprint of humanity. Proceedings of the national academy of sciences, 109(9): 3232--3237

    2012

  14. [22]

    Jia, X.; Chen, S.; Zheng, C.; Xie, Y.; Jiang, Z.; and Kalanat, N. 2023. Physics-guided Graph Diffusion Network for Combining Heterogeneous Simulated Data: An Application in Predicting Stream Water Temperature. In Proceedings of the 2023 SIAM International Conference on Data Mi...

    2023

  15. [23]

    Jia, X.; Willard, J.; Karpatne, A.; Read, J.; Zwart, J.; Steinbach, M.; and Kumar, V. 2019. Physics Guided RNNs for Modeling Dynamical Systems: A Case Study in Simulating Lake Temperature Profiles. In Proceedings of SIAM International Conference on Data Mining

    2019

  16. [24]

    Jia, X.; Xie, Y.; Li, S.; Chen, S.; Zwart, J.; Sadler, J.; Appling, A.; Oliver, S.; and Read, J. 2021 a . Physics-Guided Machine Learning from Simulation Data: An Application in Modeling Lake and River Systems. In 2021 IEEE International Conference on Data Mining (ICDM), 270--...

    2021

  17. [25]

    Jia, X.; Zwart, J.; Sadler, J.; Appling, A.; Oliver, S.; Markstrom, S.; Willard, J.; Xu, S.; Steinbach, M.; Read, J.; et al. 2021 b . Physics-guided recurrent graph model for predicting flow and temperature in river networks. In Proceedings of the 2021 SIAM International Confe...

    2021

  18. [26]

    Jia, X.; Zwart, J.; Sadler, J.; Appling, A.; Oliver, S.; Markstrom, S.; Willard, J.; Xu, S.; Steinbach, M.; Read, J.; et al. 2021 c . Physics-guided recurrent graph model for predicting flow and temperature in river networks. In Proceedings of the 2021 SIAM International Confe...

    2021

  19. [27]

    S.; and Gebru, T

    Jo, E. S.; and Gebru, T. 2020. Lessons from archives: Strategies for collecting sociocultural data in machine learning. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, 306--316

    2020

  20. [28]

    Kamishima, T.; et al. 2011. Fairness-aware learning through regularization approach. In 2011 IEEE 11th International Conference on Data Mining Workshops, 643--650. IEEE

    2011

  21. [29]

    Laclau, C.; Redko, I.; Choudhary, M.; and Largeron, C. 2021. All of the fairness for edge prediction with optimal transport. In International Conference on Artificial Intelligence and Statistics, 1774--1782. PMLR

    2021

  22. [30]

    Lahoti, P.; Beutel, A.; Chen, J.; Lee, K.; Prost, F.; Thain, N.; Wang, X.; and Chi, E. 2020. Fairness without demographics through adversarially reweighted learning. Advances in neural information processing systems, 33: 728--740

    2020

  23. [31]

    Lahoti, P.; et al. 2019 a . ifair: Learning individually fair data representations for algorithmic decision making. In 2019 ieee 35th international conference on data engineering (icde), 1334--1345. IEEE

    2019

  24. [32]

    Lahoti, P.; et al. 2019 b . Operationalizing individual fairness with pairwise fair representations. arXiv preprint arXiv:1907.01439

    2019 arXiv

  25. [33]

    Li, P.; Wang, Y.; Zhao, H.; Hong, P.; and Liu, H. 2021. On dyadic fairness: Exploring and mitigating bias in graph connections. In International Conference on Learning Representations

    2021

  26. [34]

    Liu, L.; Zhou, W.; Guan, K.; Peng, B.; Xu, S.; Tang, J.; Zhu, Q.; Till, J.; Jia, X.; Jiang, C.; et al. 2024. Knowledge-guided machine learning can improve carbon cycle quantification in agroecosystems. Nature communications, 15(1): 357

    2024

  27. [35]

    Manson, S. M. 2022. IPUMS national historical geographic information system: version 17.0

    2022

  28. [36]

    Markstrom, S. L. 2012. P2S--coupled Simulation with the Precipitation-Runoff Modeling System (PRMS) and the Stream Temperature Network (SNTemp) Models. US Department of the Interior, US Geological Survey

    2012

  29. [37]

    Mehrabi, N.; Morstatter, F.; Saxena, N.; Lerman, K.; and Galstyan, A. 2021. A survey on bias and fairness in machine learning. ACM Computing Surveys (CSUR), 54(6): 1--35

    2021

  30. [38]

    Moshe, Z.; Metzger, A.; Elidan, G.; Kratzert, F.; Nevo, S.; and El-Yaniv, R. 2020. Hydronets: Leveraging river structure for hydrologic modeling. arXiv preprint arXiv:2007.00595

    2020 arXiv

  31. [39]

    Muralidhar, N.; Bu, J.; Cao, Z.; He, L.; Ramakrishnan, N.; Tafti, D.; and Karpatne, A. 2020. Phynet: Physics guided neural networks for particle drag force prediction in assembly. In Proceedings of the 2020 SIAM International Conference on Data Mining, 559--567. SIAM

    2020

  32. [40]

    K.; et al

    Oliver, S. K.; et al. 2021. Predicting water temperature in the Delaware River Basin. U.S. Geological Survey Data Release

    2021

  33. [41]

    Raissi, M.; et al. 2017. Physics informed deep learning (part i): Data-driven solutions of nonlinear partial differential equations. arXiv preprint arXiv:1711.10561

    2017 arXiv

  34. [42]

    K.; et al

    Read, E. K.; et al. 2017. Water quality data for national-scale aquatic research: The Water Quality Portal. Water Resources Research

    2017

  35. [43]

    S.; et al

    Regan, R. S.; et al. 2018. Description of the national hydrologic model for use with the precipitation-runoff modeling system (PRMS). Technical report, US Geological Survey

    2018

  36. [44]

    Schwartz, R.; Vassilev, A.; Greene, K.; Perine, L.; Burt, A.; Hall, P.; et al. 2022. Towards a standard for identifying and managing bias in artificial intelligence. NIST special publication, 1270(10.6028)

    2022

  37. [45]

    Serna, I.; Morales, A.; Fierrez, J.; Cebrian, M.; Obradovich, N.; and Rahwan, I. 2020. Sensitiveloss: Improving accuracy and fairness of face representations with discrimination-aware deep learning. arXiv preprint arXiv:2004.11246

    2020 arXiv

  38. [46]

    Steed, R.; and Caliskan, A. 2021. Image representations learned with unsupervised pre-training contain human-like biases. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 701--713

    2021

  39. [47]

    Y.; Jiang, P.; Mudunuru, M

    Sun, A. Y.; Jiang, P.; Mudunuru, M. K.; and Chen, X. 2021. Explore spatio-temporal learning of large sample hydrology using graph neural networks. Water Resources Research, 57(12): e2021WR030394

    2021

  40. [48]

    Sweeney, C.; et al. 2020. Reducing sentiment polarity for demographic attributes in word embeddings using adversarial learning. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, 359--368

    2020

  41. [49]

    Tang, X.; Yao, H.; Sun, Y.; Wang, Y.; Tang, J.; Aggarwal, C.; Mitra, P.; and Wang, S. 2020. Investigating and mitigating degree-related biases in graph convoltuional networks. In Proceedings of the 29th ACM International Conference on Information & Knowledge Management, 1435--1444

    2020

  42. [50]

    N.; Barclay, J.; Diaz, J.; Sun, A

    Topp, S. N.; Barclay, J.; Diaz, J.; Sun, A. Y.; Jia, X.; Lu, D.; Sadler, J. M.; and Appling, A. P. 2023. Stream temperature prediction in a shifting environment: Explaining the influence of deep learning architecture. Water Resources Research, 59(4): e2022WR033880

    2023

  43. [51]

    USGS. 2016. US Geological Survey. National water information system data available on the world wide web (USGS water data for the nation). doi:10.5066/F7P55KJN

    2016 doi

  44. [52]

    Wang, X.; Ma, Y.; Wang, Y.; Jin, W.; Wang, X.; Tang, J.; Jia, C.; and Yu, J. 2020. Traffic flow prediction via spatial temporal graph neural network. In Proceedings of the web conference 2020, 1082--1092

    2020

  45. [53]

    Willard, J.; Jia, X.; Xu, S.; Steinbach, M.; and Kumar, V. 2022. Integrating scientific knowledge with machine learning for engineering and environmental systems. ACM Computing Surveys, 55(4): 1--37

    2022

  46. [54]

    N.; et al

    Williamson, T. N.; et al. 2015. Summary of hydrologic modeling for the Delaware River Basin using the Water Availability Tool for Environmental Resources (WATER). Technical report, U.S. Geological Survey Scientific Investigations Report

    2015

  47. [55]

    Xie, Y.; He, E.; Jia, X.; Chen, W.; Skakun, S.; Bao, H.; Jiang, Z.; Ghosh, R.; and Ravirathinam, P. 2022. Fairness by “Where”: A Statistically-Robust and Model-Agnostic Bi-Level Learning Framework. In Proceedings of the AAAI Conference on Artificial Intelligence

    2022

  48. [56]

    Yan, A.; et al. 2019. Fairst: Equitable spatial and temporal demand prediction for new mobility systems. In Proceedings of the 27th ACM SIGSPATIAL International Conference on Advances in Geographic Information Systems, 552--555

    2019

  49. [57]

    Yang, K.; Qinami, K.; Fei-Fei, L.; Deng, J.; and Russakovsky, O. 2020. Towards fairer datasets: Filtering and balancing the distribution of the people subtree in the imagenet hierarchy. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, 547--558

    2020

  50. [58]

    B.; Valera, I.; Gomez Rodriguez, M.; and Gummadi, K

    Zafar, M. B.; Valera, I.; Gomez Rodriguez, M.; and Gummadi, K. P. 2017. Fairness beyond disparate treatment & disparate impact: Learning classification without disparate mistreatment. In Proceedings of the 26th international conference on world wide web, 1171--1180

    2017

  51. [59]

    Zhang, C.; et al. 2014. Fairness in multi-agent sequential decision-making. Advances in Neural Information Processing Systems, 27

    2014

  52. [60]

    Zhang, H.; et al. 2021. Towards Fair Deep Anomaly Detection. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency, 138--148

    2021

  53. [61]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  54. [62]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.