pith. machine review for the scientific record. sign in

arxiv: 2605.07999 · v1 · submitted 2026-05-08 · 💻 cs.LG · cs.AI

Recognition: 2 theorem links

· Lean Theorem

Graph-Structured Hyperdimensional Computing for Data-Efficient and Explainable Process-Structure-Property Prediction

Jingzhan Ge , Ajeeth Vellore , Ajinkya Palwe , Ahsan Khan , David Gorsich , Matthew P. Castanier , SeungYeon Kang , Farhad Imani
Authors on Pith no claims yet

Pith reviewed 2026-05-11 02:53 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords hyperdimensional computingprocess-structure-property predictiongraph-structured modelsexplainable AIadditive manufacturingsparse data3D microstructuresheet resistance
0
0 comments X

The pith

Encoding a directed PSP graph in hyperdimensional space enables accurate regime prediction and intrinsic explanations from sparse manufacturing data.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces PSP-HDC to address process-structure-property prediction when data are sparse, heterogeneous, and dominated by interactions. It encodes the directed PSP graph as a prior by mapping parameters to hypervectors with a trainable encoder and composing representations through graph-aligned binding and bundling. Prediction occurs via associative retrieval against class prototypes, which simultaneously support attribution at parameter, group, and within-group levels. On sheet-resistance regime tasks the method reaches 0.910 accuracy on random splits and 0.896 under process-fold generalization while outperforming baselines.

Core claim

PSP-HDC encodes a directed PSP graph as an internal prior in hyperdimensional space. A trainable scalar-to-hypervector encoder produces parameter-specific embeddings on a fixed basis. Sample representations are formed by graph-aligned binding and bundling along directed dependencies. Inference and explanation both rely on associative-memory retrieval against class prototypes, with memory alignment and separation tracking prototype formation.

What carries the argument

Graph-aligned binding and bundling of hypervectors derived from a directed PSP graph, followed by associative-memory retrieval against class prototypes.

If this is right

  • Prediction accuracy reaches approximately 0.91 on random splits and 0.90 under process-fold generalization for sheet-resistance regimes.
  • The same prototype memories supply explanations at parameter, group, and within-group levels without separate post-hoc tools.
  • Trainable encoders accommodate heterogeneous parameter scales and noise levels within a single fixed hyperdimensional basis.
  • Memory alignment and separation metrics provide a built-in way to monitor how class prototypes form during training.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The approach may transfer to other additive-manufacturing tasks that share similar sparsity and interaction-heavy data regimes.
  • Hypervector operations could support efficient hardware realizations for on-line monitoring in fabrication systems.
  • Combining the graph prior with mechanistic submodels might further improve performance once some calibrated physics becomes available.

Load-bearing premise

A directed PSP graph can be encoded faithfully in hyperdimensional space so that binding and bundling preserve true interactions without adding spurious correlations from sparse data.

What would settle it

A new PSP dataset whose physical dependencies contradict the assumed graph structure, yet on which PSP-HDC still outperforms baselines and produces explanations that match known mechanisms, would challenge the central claim.

Figures

Figures reproduced from arXiv: 2605.07999 by Ahsan Khan, Ajeeth Vellore, Ajinkya Palwe, David Gorsich, Farhad Imani, Jingzhan Ge, Matthew P. Castanier, SeungYeon Kang.

Figure 1
Figure 1. Figure 1: End-to-end PSP-HDC workflow for the multiphoton photoreduction dataset studied in this paper. Samples are fabricated under multiple process settings [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustrative PSP graph instantiated by bundling and graph-aligned [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: PSP-HDC training and inference. Each scalar [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Representative fabricated sample array produced using the 3D OHMIC [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: SEM-EDS elemental maps shown as a single montage for Sample 0 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: SEM montage for the 5% AgNO3 group under a scan-speed sweep at fixed laser power and hatch spacing (P = 50 mW, h = 5 µm). Panels (a)-(e) correspond to: (a) v = 50 µm/s, (b) v = 100 µm/s, (c) v = 200 µm/s, (d) v = 300 µm/s, and (e) v = 500 µm/s. All images are shown at the same magnification; scale bar: 100 µm. Accordingly, pore statistics are computed from SEM mi￾crographs via a standardized image-processi… view at source ↗
Figure 7
Figure 7. Figure 7: Pore segmentation and verification from an SEM micrograph of [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Each parameter is represented as a parameter-level node (encoded as a [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: reveals a pronounced speed dependence. Rs gen￾erally increases as scan speed increases, consistent with re￾duced delivered energy per unit length at higher speeds. Low￾speed settings more frequently achieve low sheet resistance, whereas high-speed settings produce larger Rs and greater dispersion, including many high-resistance realizations. The pattern is strongest for the 10% and 15% formulations. The 5%… view at source ↗
Figure 10
Figure 10. Figure 10: Sheet resistance distributions and marginal trends across key process variables in the 3D OHMIC dataset. The vertical axis is logarithmic and one [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Structure to property relations for representative composition and pore-morphology descriptors. Sheet resistance [PITH_FULL_IMAGE:figures/full_fig_p013_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Sensitivity of the adaptive scalar encoder with fixed hypervector [PITH_FULL_IMAGE:figures/full_fig_p013_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Global learning behavior and prototype evolution for PSP-HDC [PITH_FULL_IMAGE:figures/full_fig_p014_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: MAS distribution comparison for class-partitioned components before training and after convergence. The left column shows individual parameters [PITH_FULL_IMAGE:figures/full_fig_p015_14.png] view at source ↗
Figure 16
Figure 16. Figure 16: Parameter-level attribution for PSP-HDC. Bars show the class [PITH_FULL_IMAGE:figures/full_fig_p016_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Parameter-group attribution on the PSP graph. Bars show the relative [PITH_FULL_IMAGE:figures/full_fig_p016_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Within-group attribution decomposition. Each panel corresponds [PITH_FULL_IMAGE:figures/full_fig_p017_18.png] view at source ↗
read the original abstract

Multiphoton photoreduction enables high-fidelity fabrication of complex 3D microstructures, yet reliable process-structure-property (PSP) prediction remains difficult because the available data are sparse, heterogeneous, and interaction-dominated. In this regime, conventional feature-vector models are statistically underdetermined, making them prone to spurious correlations, poor regime transfer, and unstable post hoc explanations, whereas mechanistic pipelines depend on calibrated submodels that are rarely available during early process development. We present PSP-HDC, a graph-structured hyperdimensional computing framework that encodes a directed PSP graph as an internal prior for representation, inference, and explanation. A trainable scalar-to-hypervector encoder learns parameter-specific embeddings on a fixed hyperdimensional basis to accommodate heterogeneous scales and noise. Sample representations are then composed through graph-aligned binding and bundling along directed PSP dependencies, and prediction is performed by associative-memory retrieval against class prototypes. Because the same prototype memories support both decision making and attribution, PSP-HDC provides intrinsic explanations at the parameter, group, and within-group levels, while memory alignment and separation quantify prototype formation during training. On sheet-resistance regime prediction for the 3D platform, PSP-HDC achieves an accuracy of 0.910 +/- 0.077 over 1000 random splits and 0.896 under process-fold generalization, outperforming strong baselines.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper claims to introduce PSP-HDC, a graph-structured hyperdimensional computing framework for data-efficient and explainable process-structure-property (PSP) prediction in multiphoton photoreduction of 3D microstructures. It encodes a directed PSP graph as an internal prior using a trainable scalar-to-hypervector encoder, composes representations via graph-aligned binding and bundling, and performs prediction and explanation using associative memory retrieval against class prototypes. Empirical results on sheet-resistance regime prediction show accuracies of 0.910 ± 0.077 over 1000 random splits and 0.896 under process-fold generalization, outperforming strong baselines.

Significance. If the results and assumptions hold, this work has potential significance in advancing machine learning methods for scientific domains with sparse and heterogeneous data. By incorporating domain knowledge through graph priors in hyperdimensional space, it offers an alternative to data-hungry deep learning models and mechanistic simulations that require calibrated submodels. The intrinsic explainability at multiple levels and the use of memory alignment for quantifying prototype formation are notable strengths. The process-fold generalization test demonstrates robustness beyond random splits. However, the overall impact depends on confirming that the performance gains arise from the proposed mechanisms rather than unexamined factors.

major comments (2)
  1. Abstract: The headline accuracy figures (0.910 ± 0.077 on random splits and 0.896 on process-fold) are presented without dataset size, class balance, feature count, baseline architectures, or ablation results. This information is load-bearing for determining whether the reported gains are attributable to the graph-structured HDC mechanisms or to unstated implementation choices.
  2. Framework description: The central assumption that graph-aligned binding and bundling faithfully preserve PSP interactions without introducing spurious correlations (due to sparsity and heterogeneity) lacks supporting analysis such as crosstalk bounds, reconstruction-error metrics, or ablation on edge fidelity. This is load-bearing for the claim that the directed PSP graph serves as an effective internal prior.
minor comments (1)
  1. Abstract: The phrase 'strong baselines' is used without naming the models or their key hyperparameters; this should be expanded in the main text for clarity and reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the presentation of our results and strengthen the supporting analysis for the framework. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: Abstract: The headline accuracy figures (0.910 ± 0.077 on random splits and 0.896 on process-fold) are presented without dataset size, class balance, feature count, baseline architectures, or ablation results. This information is load-bearing for determining whether the reported gains are attributable to the graph-structured HDC mechanisms or to unstated implementation choices.

    Authors: We agree that the abstract would be more self-contained with these details. In the revised version we will expand the abstract to state the dataset size, class balance, feature count, the specific baseline architectures compared, and a brief reference to the ablation studies reported in the main text. This will make the performance claims easier to evaluate at a glance while respecting abstract length constraints. revision: yes

  2. Referee: Framework description: The central assumption that graph-aligned binding and bundling faithfully preserve PSP interactions without introducing spurious correlations (due to sparsity and heterogeneity) lacks supporting analysis such as crosstalk bounds, reconstruction-error metrics, or ablation on edge fidelity. This is load-bearing for the claim that the directed PSP graph serves as an effective internal prior.

    Authors: We acknowledge that additional quantitative support would strengthen the claim. The manuscript already shows empirical robustness via process-fold generalization and performance comparisons, but to directly address potential spurious correlations we will add in the revision: (i) theoretical crosstalk bounds for the graph-aligned binding operation, (ii) reconstruction-error metrics on composed hypervectors, and (iii) an explicit ablation that removes or perturbs individual directed edges and reports the resulting change in accuracy and prototype separation. These additions will provide the requested evidence that the directed PSP graph functions as a faithful internal prior. revision: yes

Circularity Check

0 steps flagged

No significant circularity; performance evaluated on held-out splits

full rationale

The reported accuracies (0.910 on 1000 random splits, 0.896 on process-fold) are measured on data partitions independent of the training set used to fit the scalar-to-hypervector encoder and form class prototypes. Graph-aligned binding/bundling and associative retrieval are standard HDC operations applied to an input graph prior; no equation reduces a held-out prediction to a training-set quantity by construction. No self-citation load-bearing steps, uniqueness theorems, or ansatz smuggling appear in the provided description. The framework is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 1 invented entities

The central claim rests on the domain assumption that a directed PSP graph faithfully captures the dominant dependencies in the fabrication process and that hyperdimensional binding/bundling operations can compose these dependencies without loss of critical information.

free parameters (1)
  • scalar-to-hypervector encoder weights
    Trainable parameters that map heterogeneous scalar inputs to a fixed hyperdimensional basis; their values are learned from data and directly affect all subsequent compositions.
axioms (2)
  • domain assumption A directed graph over process, structure, and property variables accurately encodes the causal and dependency structure of the physical system.
    Invoked when the paper states that the graph is used as an internal prior for representation, inference, and explanation.
  • domain assumption Hyperdimensional binding and bundling operations preserve the necessary interactions when applied along the directed graph edges.
    Required for the claim that sample representations composed this way support accurate associative-memory retrieval.
invented entities (1)
  • PSP-HDC framework no independent evidence
    purpose: To serve as a unified representation, inference, and explanation engine by embedding the PSP graph inside hyperdimensional memory.
    New composite method introduced in the paper; no independent evidence outside the reported experiments is supplied.

pith-pipeline@v0.9.0 · 5573 in / 1578 out tokens · 42137 ms · 2026-05-11T02:53:59.607831+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages

  1. [1]

    Process monitoring, diagnosis and control of additive manufacturing,

    Q. Fang, G. Xiong, M. C. Zhou, T. S. Tamir, C. B. Yan, H. Wu, Z. Shen, and F. Y . Wang, “Process monitoring, diagnosis and control of additive manufacturing,”IEEE Transactions on Automation Science and Engineering, vol. 21, no. 1, pp. 1041–1067, 2024. 18

    work page 2024

  2. [2]

    Intelligent additive manufacturing architecture for enhancing uniformity of surface roughness and mechanical properties of laser powder bed fusion components,

    H. C. Tran, Y . L. Lo, H. C. Yang, H. C. Hsiao, F. T. Cheng, and T. H. Kuo, “Intelligent additive manufacturing architecture for enhancing uniformity of surface roughness and mechanical properties of laser powder bed fusion components,”IEEE Transactions on Automation Science and Engineering, vol. 20, no. 4, pp. 2527–2538, 2023

    work page 2023

  3. [3]

    Online distortion simulation using generative machine learning models: A step toward digital twin of metallic additive manufacturing,

    H. Mu, F. He, L. Yuan, H. Hatamian, P. Commins, and Z. Pan, “Online distortion simulation using generative machine learning models: A step toward digital twin of metallic additive manufacturing,”Journal of Industrial Information Integration, vol. 38, p. 100563, 2024

    work page 2024

  4. [4]

    In- terpretable machine learning approach for exploring process-structure- property relationships in metal additive manufacturing,

    Q. Liu, W. Chen, V . Yakubov, J. J. Kruzic, C. H. Wang, and X. Li, “In- terpretable machine learning approach for exploring process-structure- property relationships in metal additive manufacturing,”Additive Man- ufacturing, vol. 85, p. 104187, 2024

    work page 2024

  5. [5]

    An integrated process–structure–property modeling framework for additive manufacturing,

    W. Yan, Y . Lian, C. Yu, O. L. Kafka, Z. Liu, W. K. Liu, and G. J. Wagner, “An integrated process–structure–property modeling framework for additive manufacturing,”Computer Methods in Applied Mechanics and Engineering, vol. 339, pp. 184–204, 2018

    work page 2018

  6. [6]

    A deep-learning-based surro- gate model for thermal signature prediction in laser metal deposition,

    S. Guo, W. Guo, L. Bian, and Y . B. Guo, “A deep-learning-based surro- gate model for thermal signature prediction in laser metal deposition,” IEEE Transactions on Automation Science and Engineering, vol. 20, no. 1, pp. 482–494, 2023

    work page 2023

  7. [7]

    Dataset of process-structure-property feature relationships for alsi10mg material fabricated using laser powder bed fusion additive manufacturing,

    Q. Luo, N. Huang, T. Fu, J. Wang, D. L. Bartles, T. W. Simpson, and A. M. Beese, “Dataset of process-structure-property feature relationships for alsi10mg material fabricated using laser powder bed fusion additive manufacturing,”Data in Brief, vol. 53, p. 110130, 2024

    work page 2024

  8. [8]

    Metal am process-structure-property relational linkages using gaussian process surrogates,

    R. N. Saunders, K. Teferra, A. Elwany, J. G. Michopoulos, and D. Lagoudas, “Metal am process-structure-property relational linkages using gaussian process surrogates,”Additive Manufacturing, vol. 62, p. 103398, 2023

    work page 2023

  9. [9]

    Learning and predicting shape deviations of smooth and non-smooth 3d geometries through mathe- matical decomposition of additive manufacturing,

    Y . Wang, C. Ruiz, and Q. Huang, “Learning and predicting shape deviations of smooth and non-smooth 3d geometries through mathe- matical decomposition of additive manufacturing,”IEEE Transactions on Automation Science and Engineering, vol. 20, no. 3, pp. 1527–1538, 2022

    work page 2022

  10. [10]

    Surrogate-based model chains for establishing process-structure-property linkages with quantified uncer- tainties in metal additive manufacturing,

    J. Ye, R. N. Saunders, and A. Elwany, “Surrogate-based model chains for establishing process-structure-property linkages with quantified uncer- tainties in metal additive manufacturing,”Manufacturing Letters, vol. 35, pp. 750–759, 2023

    work page 2023

  11. [11]

    Semi-supervised classification with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” inInternational Conference on Learning Representations (ICLR), 2017. [Online]. Available: https: //openreview.net/forum?id=SJU4ayYgl

    work page 2017

  12. [12]

    Physics-informed weakly- supervised learning for quality prediction of manufacturing processes,

    D. F. Alenezi, M. Biehler, J. Shi, and J. Li, “Physics-informed weakly- supervised learning for quality prediction of manufacturing processes,” IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 2006–2019, 2025

    work page 2006

  13. [13]

    Graph attention networks,

    P. Veli ˇckovi´c, G. Cucurull, A. Casanova, A. Romero, P. Li `o, and Y . Bengio, “Graph attention networks,” inInternational Conference on Learning Representations (ICLR), 2018. [Online]. Available: https://openreview.net/forum?id=rJXMpikCZ

    work page 2018

  14. [14]

    Revolutionizing 3d electronics: Single-step femtosecond laser fabrication of conductive embedded structures and circuitry,

    S. Awasthi and S. Kang, “Revolutionizing 3d electronics: Single-step femtosecond laser fabrication of conductive embedded structures and circuitry,”Materials Today Advances, vol. 25, p. 100544, Mar. 2025. [Online]. Available: https://doi.org/10.1016/j.mtadv.2024.100544

    work page doi:10.1016/j.mtadv.2024.100544 2025

  15. [15]

    Dataset of process- structure-property feature relationship for laser powder bed fusion addi- tive manufactured ti-6al-4v material

    Q. Luo, L. Yin, T. W. Simpson, and A. M. Beese, “Dataset of process- structure-property feature relationship for laser powder bed fusion addi- tive manufactured ti-6al-4v material.”Data in Brief, vol. 46, p. 108911, 2023

    work page 2023

  16. [16]

    Laser powder bed fusion dataset for relative density prediction of commercial metallic alloys,

    G. O. Barrionuevo, I. La F ´e-Perdomo, and J. A. Ramos-Grez, “Laser powder bed fusion dataset for relative density prediction of commercial metallic alloys,”Scientific Data, vol. 12, no. 1, p. 375, 2025

    work page 2025

  17. [17]

    Aachen-heerlen annotated steel microstruc- ture dataset,

    D. Iren, M. Ackermann, J. Gorfer, G. Pujar, S. Wesselmecking, U. Krupp, and S. Bromuri, “Aachen-heerlen annotated steel microstruc- ture dataset,”Scientific data, vol. 8, no. 1, p. 140, 2021

    work page 2021

  18. [18]

    Thermography of the metal bridge structures fabricated for the 2018 additive manufacturing benchmark test series (am-bench 2018),

    J. C. Heigel, B. M. Lane, L. Levine, T. Q. Phan, and J. G. Whiting, “Thermography of the metal bridge structures fabricated for the 2018 additive manufacturing benchmark test series (am-bench 2018),”Journal of Research of the National Institute of Standards and Technology, vol. 125, p. 005, Jan. 2020. [Online]. Available: https://doi.org/10.6028/jres.125.005

    work page doi:10.6028/jres.125.005 2018

  19. [19]

    A compilation of experimental data on the mechanical properties and microstructural features of ti-alloys,

    C. A. F. Salvador, E. L. Maia, F. H. Costa, J. D. Escobar, and J. P. Oliveira, “A compilation of experimental data on the mechanical properties and microstructural features of ti-alloys,” Scientific Data, vol. 9, p. 188, Apr. 2022. [Online]. Available: https://doi.org/10.1038/s41597-022-01283-9

    work page doi:10.1038/s41597-022-01283-9 2022

  20. [20]

    Process monitoring, diagnosis and control of additive manufacturing,

    Q. Fang, G. Xiong, M. C. Zhou, T. S. Tamir, C. B. Yan, H. Wu, Z. Shen, and F. Y . Wang, “Process monitoring, diagnosis and control of additive manufacturing,”IEEE Transactions on Automation Science and Engineering, vol. 21, no. 1, pp. 1041–1067, 2024. [Online]. Available: https://doi.org/10.1109/TASE.2022.3215258

    work page doi:10.1109/tase.2022.3215258 2024

  21. [21]

    A deep-learning- based surrogate model for thermal signature prediction in laser metal deposition,

    S. Guo, W. Guo, L. Bian, and Y . Guo, “A deep-learning- based surrogate model for thermal signature prediction in laser metal deposition,”IEEE Transactions on Automation Science and Engineering, vol. 20, no. 1, pp. 482–494, 2023. [Online]. Available: https://doi.org/10.1109/TASE.2022.3158204

    work page doi:10.1109/tase.2022.3158204 2023

  22. [22]

    Physics-informed weakly-supervised learning for quality prediction of manufacturing processes,

    D. F. Alenezi, M. Biehler, J. Shi, and J. Li, “Physics-informed weakly-supervised learning for quality prediction of manufacturing processes,”IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 2006–2019, 2025. [Online]. Available: https://doi.org/10. 1109/TASE.2024.3374098

    work page arXiv 2006

  23. [23]

    Architecture-driven physics-informed deep learning for temperature prediction in laser powder bed fusion additive manufacturing with limited data,

    S. Ghungrad, M. Faegh, B. Gould, S. J. Wolff, and A. Haghighi, “Architecture-driven physics-informed deep learning for temperature prediction in laser powder bed fusion additive manufacturing with limited data,”Journal of Manufacturing Science and Engineering, vol. 145, no. 8, p. 081007, 2023

    work page 2023

  24. [24]

    Ten- sile strength prediction in directed energy deposition through physics- informed machine learning and shapley additive explanations,

    C. Cooper, J. Zhang, J. Huang, J. Bennett, J. Cao, and R. X. Gao, “Ten- sile strength prediction in directed energy deposition through physics- informed machine learning and shapley additive explanations,”Journal of Materials Processing Technology, vol. 315, p. 117908, 2023

    work page 2023

  25. [25]

    A physics-informed convolu- tional neural network with custom loss functions for porosity prediction in laser metal deposition,

    E. McGowan, V . Gawade, and W. Guo, “A physics-informed convolu- tional neural network with custom loss functions for porosity prediction in laser metal deposition,”Sensors, vol. 22, no. 2, p. 494, 2022

    work page 2022

  26. [26]

    A physics-informed machine learning method for predicting grain structure characteristics in directed energy deposition,

    D. Kats, Z. Wang, Z. Gan, W. K. Liu, G. J. Wagner, and Y . Lian, “A physics-informed machine learning method for predicting grain structure characteristics in directed energy deposition,”Computational Materials Science, vol. 202, p. 110958, 2022

    work page 2022

  27. [27]

    New insight into the multivariate relationships among process, structure, and properties in laser powder bed fusion alsi10mg,

    Q. Luo, N. Huang, T. Fu, J. Wang, D. L. Bartles, T. W. Simpson, and A. M. Beese, “New insight into the multivariate relationships among process, structure, and properties in laser powder bed fusion alsi10mg,” Additive Manufacturing, vol. 77, p. 103804, 2023

    work page 2023

  28. [28]

    Data-driven analysis of process, structure, and properties of additively manufactured inconel 718 thin walls,

    L. Fang, L. Cheng, J. A. Glerum, J. Bennett, J. Cao, and G. J. Wagner, “Data-driven analysis of process, structure, and properties of additively manufactured inconel 718 thin walls,”npj Computational Materials, vol. 8, no. 1, p. 126, 2022

    work page 2022

  29. [29]

    Machine learning of microstructure–property relationships in materials leveraging microstructure representation from foundational vision transformers

    S. E. Whitman and M. I. Latypov, “Machine learning of microstructure–property relationships in materials leveraging microstructure representation from foundational vision transformers,” Acta Materialia, vol. 296, p. 121217, Sep. 2025. [Online]. Available: https://doi.org/10.1016/j.actamat.2025.121217

    work page doi:10.1016/j.actamat.2025.121217 2025

  30. [30]

    Graph-based variation propagation network for modeling and prediction of hybrid multi-stage manufacturing systems,

    D. F. Alenezi, J. Shi, and J. Li, “Graph-based variation propagation network for modeling and prediction of hybrid multi-stage manufacturing systems,”IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 14 317–14 331, 2025. [Online]. Available: https://doi.org/10.1109/TASE.2025.3560174

    work page doi:10.1109/tase.2025.3560174 2025

  31. [31]

    Inductive representation learning on large graphs,

    W. L. Hamilton, R. Ying, and J. Leskovec, “Inductive representation learning on large graphs,” inAdvances in Neural Information Processing Systems (NeurIPS), vol. 30, 2017. [Online]. Available: https://proceedings.neurips.cc/paper files/paper/ 2017/hash/5dd9db5e033da9c6fb5ba83c7a7ebea9-Abstract.html

    work page 2017

  32. [32]

    Hyperdimensional computing: An introduction to comput- ing in distributed representation with high-dimensional random vectors,

    P. Kanerva, “Hyperdimensional computing: An introduction to comput- ing in distributed representation with high-dimensional random vectors,” Cognitive computation, vol. 1, no. 2, pp. 139–159, 2009

    work page 2009

  33. [33]

    Efficient biosignal processing using hyperdimensional computing: Network templates for combined learning and classification of exg signals,

    A. Rahimi, P. Kanerva, L. Benini, and J. M. Rabaey, “Efficient biosignal processing using hyperdimensional computing: Network templates for combined learning and classification of exg signals,”Proceedings of the IEEE, vol. 107, no. 1, pp. 123–143, 2018

    work page 2018

  34. [34]

    Neural computation for robust and holo- graphic face detection,

    M. Imani, A. Zakeri, H. Chen, T. Kim, P. Poduval, H. Lee, Y . Kim, E. Sadredini, and F. Imani, “Neural computation for robust and holo- graphic face detection,” inProceedings of the 59th ACM/IEEE Design Automation Conference, 2022, pp. 31–36

    work page 2022

  35. [35]

    Brain-inspired computing for in-process melt pool characterization in additive manufacturing,

    R. Chen, M. Sodhi, M. Imani, M. Khanzadeh, A. Yadollahi, and F. Imani, “Brain-inspired computing for in-process melt pool characterization in additive manufacturing,”CIRP Journal of Manufacturing Science and Technology, vol. 41, pp. 380–390, 2023

    work page 2023

  36. [36]

    Hierarchical representation and interpretable learning for accelerated quality monitoring in machin- ing process,

    D. Hoang, H. Errahmouni, H. Chen, S. Rachuri, N. Mannan, R. ElKhar- boutly, M. Imani, R. Chen, and F. Imani, “Hierarchical representation and interpretable learning for accelerated quality monitoring in machin- ing process,”CIRP Journal of Manufacturing Science and Technology, vol. 50, pp. 198–212, 2024

    work page 2024

  37. [37]

    Graphd: Graph-based hyperdimensional mem- orization for brain-like cognitive learning,

    P. Poduval, H. Alimohamadi, A. Zakeri, F. Imani, M. H. Najafi, T. Gi- vargis, and M. Imani, “Graphd: Graph-based hyperdimensional mem- orization for brain-like cognitive learning,”Frontiers in Neuroscience, vol. 16, p. 757125, 2022. 19

    work page 2022

  38. [38]

    Composition- based multi-relational graph convolutional networks,

    S. Vashishth, S. Sanyal, V . Nitin, and P. Talukdar, “Composition- based multi-relational graph convolutional networks,”arXiv preprint arXiv:1911.03082, 2019

    work page arXiv 1911

  39. [39]

    A theoretical perspective on hyperdimensional computing,

    A. Thomas, S. Dasgupta, and T. Rosing, “A theoretical perspective on hyperdimensional computing,”Journal of Artificial Intelligence Re- search, vol. 72, pp. 215–249, 2021

    work page 2021

  40. [40]

    Adaptive document image binarization,

    J. Sauvola and M. Pietik ¨ainen, “Adaptive document image binarization,” Pattern recognition, vol. 33, no. 2, pp. 225–236, 2000

    work page 2000

  41. [41]

    Porespy: A python toolkit for quantitative analysis of porous media images,

    J. T. Gostick, Z. A. Khan, T. G. Tranter, M. D. Kok, M. Agnaou, M. Sadeghi, and R. Jervis, “Porespy: A python toolkit for quantitative analysis of porous media images,”Journal of Open Source Software, vol. 4, no. 37, p. 1296, 2019

    work page 2019

  42. [42]

    Measurement of sheet resistivities with the four-point probe,

    F. M. Smits, “Measurement of sheet resistivities with the four-point probe,”Bell System Technical Journal, vol. 37, no. 3, pp. 711–718, 1958

    work page 1958

  43. [43]

    Correction factor tables for four-point probe resistivity measurements on thin, circular semiconductor samples,

    L. J. Swartzendruber, “Correction factor tables for four-point probe resistivity measurements on thin, circular semiconductor samples,” National Bureau of Standards, Technical Note 199, 1964, issued April 15, 1964. [Online]. Available: https://nvlpubs.nist.gov/nistpubs/Legacy/ TN/nbstechnicalnote199.pdf

    work page 1964