Pith. sign in

REVIEW 3 major objections 5 minor 36 references

EmpireDB: Data System to Accelerate Computational Sciences

T0 review · 3 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read A database system that compiles approximation tolerances into model training could accelerate computational science, especially materials discovery.

desk verdict Coherent vision paper whose single experiment is a layer-count sweep that cannot carry the weight of the abstract's claim. read the letter →

arxiv 2412.10546 v1 pith:QTVRGRCK submitted 2024-12-13 cs.DB

classification cs.DB MSC 68P15
keywords computationalsciencedatabasemanagementsystemmaterialsdiscoverygraphneuralnetworksquerylanguageapproximationtoleranceactivelearningvisionpaper
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

EmpireDB is a proposed database management system built for computational science, with the central claim that scientific domain knowledge—such as how much approximation is acceptable, what integrity constraints must hold, and how many active-learning rounds a task needs—can be expressed in a query language and then compiled into training, inference, filtering, and storage plans. The paper argues that contemporary systems fail to carry this knowledge across system layers, so approximations and domain rules are applied ad hoc. As proof of concept, it reports a small experiment on the MUTAG molecular graph dataset in which dynamically tuning the number of message-passing layers to meet an accuracy threshold (the EmpireDB-tuned mode) reaches higher training accuracy than any static-layer configuration. If the vision holds, database abstraction would bring the same rigor and optimization to scientific modeling that SQL brought to relational data.

What carries the argument

The load-bearing mechanism is the EmpireDB query language and query planner: the language extends SQL with commands for approximation tolerance and constraint-based learning, and the planner analyzes alternative execution plans and chooses the lowest-cost one. The paper argues that tolerance specifications must propagate from the query layer into the training, inference, and filter pipelines, and that active-learning stage counts should determine whether data lives in memory or on disk. The demonstrated mechanism is model-complexity tuning: rather than fixing the number of message-passing layers, the execution pipeline adjusts the layer count to meet a specified accuracy threshold, using the query-specified tolerance as the stopping criterion.

What would settle it

A concrete test would be to implement the query language and planner for one scientific task, such as DFT candidate filtering in materials discovery, and check whether a compiled plan preserves the approximation and integrity guarantees exactly as declared; a failure to express or enforce even one standard integrity constraint without manual intervention would falsify the central premise. A smaller-scale falsifier is a benchmark on a larger molecular dataset where static-layer GNN training runs to completion and matches or beats the dynamically tuned variant on held-out accuracy.

Watch

Extended reading notes

Core claim

The paper's central claim is that a database system whose query language can express approximation tolerance, integrity constraints, and active-learning requirements, and whose planner propagates these specifications through execution and storage engines, would accelerate computational science tasks such as materials discovery. The envisioned system, EmpireDB, is built around three components—query engine, execution pipelines (training, inference, filtering), and storage engines (in-memory and on-disk)—and is instantiated on the GNoME-style pipeline for discovering stable materials. The preliminary evidence is a comparison on the MUTAG dataset: a system that tunes the number of GNN layers to satisfy an accuracy threshold (91.24%, 94.84%, 94.85% for hidden dimensions 64, 256, 512) outperforms any static-layer-count baseline (91.19%, 91.71%, 93.75%) in training accuracy. The paper's conclusion is that in all tested cases, EmpireDB's dynamic tuning achieves higher accuracy.

Load-bearing premise

The system's value depends on the premise that a scientist's domain knowledge—approximation tolerance, integrity constraints, active-learning round counts—can be faithfully written in a query language and automatically compiled into training, inference, and storage plans without losing the guarantees the scientist needs.

Editorial extensions

If this is right

  • If correct, scientists would be able to declare approximation tolerance and integrity constraints in a query, and the system would automatically choose model depth, training strategy, and storage placement, eliminating manual pipeline tuning.
  • The same query-driven optimization machinery could absorb existing database techniques, such as locality-sensitive hashing for similarity search and spatial indexes for high-dimensional crystal structures, directly into scientific execution pipelines.
  • A declarative specification of DFT as a filter-pipeline component would let the system decide when running DFT on a candidate material is worth the cost, using search-space size as a signal.
  • For computational social science, privacy requirements like the U.S. Census Bureau's post-processing of noisy counts could be written as domain-level constraints that the system enforces automatically.

Reading between the lines

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

  • If the query-language compilation vision is realizable, it would decouple scientific modeling knowledge from system optimization, letting each field define its own domain language while reusing a common query planner—an analogy to how SQL enabled portable database applications.
  • The dynamic layer-tuning result, though based on only 188 graphs, suggests a testable extension: applying the same threshold-driven complexity control to larger molecular datasets or other GNN benchmarks could reveal whether the accuracy gain persists or saturates.
  • The active-learning feedback loop implies a concrete storage policy that the paper states only conceptually: a query-specified stage count could trigger a transition from in-memory to on-disk storage, which would be directly measurable in a prototype.
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

3 major / 5 minor

Summary. The paper presents EmpireDB, a vision for a database management system aimed at accelerating computational sciences. The proposed architecture comprises three layers---query engines, execution pipelines, and storage engines---with the goal of declaratively specifying and propagating domain knowledge (e.g., approximation tolerance, integrity constraints, active-learning counts) across the system. The paper specializes the vision to materials discovery, discussing how EmpireDB could integrate with the GNoME pipeline, and presents a small preliminary experiment on the MUTAG dataset comparing static GNN layer counts with an "EmpireDB-Tuned" variant that adjusts the number of layers to meet an accuracy threshold. The paper concludes by listing success criteria and future work.

Significance. The paper articulates a compelling research agenda: applying database system principles---query planning, cost-based optimization, storage and indexing, integrity constraints---to scientific computing pipelines, which are often built as monolithic machine-learning workflows. The connection to GNoME and the discussion of approximation tolerance and active learning are timely and highlight genuine gaps in existing systems. However, the paper is explicitly a vision paper with no implemented system, no formalization of the proposed query language or compilation steps, and no experimental validation beyond a single table of training accuracies. The experimental evidence does not support the abstract's claim that "optimized components in EmpireDB could lead to improvements in performance compared to contemporary implementations." As a vision statement, the paper has value, but the empirical claim is currently unsubstantiated and the architectural core remains at a conceptual level.

major comments (3)
  1. [Section 3, Table 1] The experiment as reported cannot support the claim that EmpireDB improves over static configurations. The "EmpireDB-Tuned" variant selects the number of GNN layers by tuning to meet an accuracy threshold, and the reported metric is the resulting training accuracy. Because the layer count is chosen using the same training-accuracy metric, it is nearly tautological that the tuned configuration matches or exceeds any static configuration. No held-out test set, test accuracy, error bars, repeated random seeds, or comparison with a contemporary system (e.g., a standard GNN library or GNoME-style pipeline) is reported. With only 188 graphs in MUTAG, the observed gaps (e.g., 94.84% vs. 91.71%) could easily be noise. The section's concluding sentence "In all cases, EmpireDB achieves higher levels of accuracy" is therefore a property of the selection procedure, not evidence for the EmpireDB architecture.
  2. [Sections 2.1.1 and 2.1.2] The central architectural premise---that approximation tolerance, integrity constraints, and active-learning stage counts can be expressed in a query language and compiled into training, inference, and filtering plans with provable guarantees---is described only at a conceptual level. No syntax or semantics for the proposed query language is given, no compilation rules from query to execution plans are specified, no cost model for the query planner is defined, and no algorithm with provable approximation guarantees is presented. This compilation step is load-bearing for the entire EmpireDB vision; without at least a formal sketch or a concrete example of how a scientific constraint would be translated into a pipeline plan, the feasibility of the architecture remains an open assumption rather than a demonstrated contribution.
  3. [Sections 2.2.1-2.2.3] The proposed optimizations for the GNoME pipeline---locality-sensitive hashing for similarity search, B-trees and quad trees for indexing, and tighter integration of DFT as a filter---are plausible research directions, but the paper provides no analysis or experimental evidence that these techniques would improve the end-to-end materials-discovery pipeline. In a vision paper, such suggestions are acceptable as a research agenda, but they should be framed explicitly as open problems rather than as evidence that "optimized components in EmpireDB could lead to improvements." As written, the paper overstates the degree of validation behind these proposals.
minor comments (5)
  1. [Abstract] The phrase "improvements in performance compared to contemporary implementations" overstates what is shown; the experiment in Section 3 does not compare against any contemporary implementation.
  2. [Section 3] The paper reports only training accuracy. For graph classification, test accuracy on held-out graphs is the standard measure of generalization, and should be reported if the goal is to demonstrate model quality.
  3. [Section 3] The "threshold of accuracy" used by EmpireDB-Tuned is not specified, nor are the range of layer counts searched, the training procedure, or the hyperparameters beyond hidden feature dimensions. These details are needed to interpret the table.
  4. [References and text] In Section 2.2, the sentence "This is what GNoME tries to accomplish [24]" cites reference [24], which is a National Academies report on reproducibility and replicability, not the GNoME paper. The citation should be corrected to the GNoME reference.
  5. [Abstract and body text] There are several typos and formatting issues, including "ofComputational Science" in the abstract and inconsistent use of italics for system names. A careful proofreading pass would improve readability.

Circularity Check

1 steps flagged · score 6.0 of 10

The only experimental evidence reduces to a layer-count sweep selected on training accuracy, so the reported improvements are by construction rather than independent evidence.

  1. fitted input called prediction [Section 3 'Preliminary Experimental Evidence', Table 1 and following paragraph.]
    "Table 1 shows the training accuracy when using a system with static number of layers vs. EmpireDB that is able to tune the number of layers to meet a threshold of accuracy. In all cases, EmpireDB achieves higher levels of accuracy."

    The 'EmpireDB-Tuned' variant tunes the number of GNN layers using the same training-accuracy objective that Table 1 then reports. Selecting a hyperparameter to meet an accuracy threshold and then reporting the achieved training accuracy is a fitted quantity: the selected model is guaranteed to reach the threshold, and taking the best of several layer counts will naturally dominate any fixed choice on the selection metric. The reported 'higher levels of accuracy' are therefore a property of the selection procedure, not an independent prediction by EmpireDB's design. No held-out test metric, repeated seeds, or error bars are reported; the abstract's claim of 'improvements in performance compared to contemporary implementations' rests entirely on this constructed comparison.

full rationale

The paper is primarily a vision/position statement; the architecture sections (query language, execution pipelines, storage engines) are conceptual and do not smuggle in a conclusion by definition. The load-bearing empirical step is confined to Section 3. There the comparison is circular in the specific sense: EmpireDB-Tuned is defined as the procedure that tunes the number of layers to meet an accuracy threshold, and the table reports training accuracy on the same data used for that tuning. Thus the result that the tuned variant 'achieves higher levels of accuracy' in every column is an artifact of taking a maximum or threshold-satisfying configuration over the swept layer counts, rather than evidence about the EmpireDB system's optimizations. This fits the 'fitted input called prediction' pattern. No other circular steps were found: the self-citations to prior work by the authors are used as background or as components, and no uniqueness theorem or ansatz is imported from their own prior work to force a conclusion. The score of 6 reflects that the paper's central empirical evidence reduces by construction, while the broader vision retains independent conceptual content.

Assumptions & free parameters 3 free parameters · 3 assumptions · 1 invented entities

The central claim rests on untested domain assumptions about the compilability of scientific constraints, plus experimental hyperparameters, and one proposed (unbuilt) system entity.

free parameters (3)
  • number of GNN layers = not reported
    In the EmpireDB-Tuned condition, the layer count is selected to satisfy an accuracy threshold. The reported improvement is a direct consequence of this selection.
  • accuracy threshold = not reported
    The target used to stop tuning layers is unspecified, so the optimization criterion cannot be reconstructed.
  • hidden feature dimensions = 64, 256, 512
    These are manual choices in the experimental design, not derived from any principle in the paper.
assumptions (3)
  • domain assumption Scientific data generation processes contain structure that can be exploited by domain-aware storage and index selection.
    Argued in Section 1 via the shard-key example and flat-view materialization cost, but no mechanism or evidence is given.
  • domain assumption Approximation tolerances and integrity constraints can be formally specified in a query language and propagated through training, inference, and filter pipelines with provable guarantees.
    Repeatedly asserted in Sections 2.1.1 and 2.1.2, but no language, planner, or proof is supplied.
  • domain assumption Active learning workloads fit naturally into a DBMS storage hierarchy (fresh labels in memory, older data on disk).
    Assumed in Section 2.1.3 based on the GNoME example; no experiments support this.
invented entities (1)
  • EmpireDB
    purpose: Proposed database management system integrating query, execution, and storage layers with domain-specific approximation controls.
    The paper provides an architecture diagram and prose description only; no implementation, interface, or benchmark exists.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EmpireDB: Data System to Accelerate Computational Sciences." pith.science (2026). https://pith.science/paper/QTVRGRCK

@misc{pith2026241210546,
  author       = {Pith},
  title        = {Pith review of: EmpireDB: Data System to Accelerate Computational Sciences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QTVRGRCK}},
  note         = {Machine review of arXiv:2412.10546}
}
read the original abstract

The emerging discipline of Computational Science is concerned with using computers to simulate or solve scientific problems. These problems span the natural, political, and social sciences. The discipline has exploded over the past decade due to the emergence of larger amounts of observational data and large-scale simulations that were previously unavailable or unfeasible. However, there are still significant challenges with managing the large amounts of data and simulations. The database management systems community has always been at the forefront of the development of the theory and practice of techniques for formalizing and actualizing systems that access or query large datasets. In this paper, we present EmpireDB, a vision for a data management system to accelerate computational sciences. In addition, we identify challenges and opportunities for the database community to further the fledgling field of computational sciences. Finally, we present preliminary evidence showing that the optimized components in EmpireDB could lead to improvements in performance compared to contemporary implementations.

Figures

Figures reproduced from arXiv: 2412.10546 by the authors.

Figure 1
Figure 1. shows the architecture of the EmpireDB system. Scientific computing tasks often rely on approximation methods to define and evaluate hypotheses. We envision a system where the toler￾ance of the approximation can be specified via a query language (the EmpireDB query language) which would be communicated to other parts of the system, especially during the training and infer￾ence stages. The training and inference pipe… view at source ↗
Figure 2
Figure 2. Illustration of the systems architecture of GNoME-based materials discovery [23]. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 24 canonical work pages

  1. [1]

    Degree Distribution Identifiability of Stochastic Kronecker Graphs

    Daniel Alabi and Dimitris Kalimeris. Degree distribution identifiability of sto- chastic kronecker graphs, 2023. arXiv:2310.00171

  2. [2]

    Pfunk-h: approximate query processing using perceptual models

    Daniel Alabi and Eugene Wu. Pfunk-h: approximate query processing using perceptual models. In Proceedings of the Workshop on Human-In-the-Loop Data Analytics, HILDA@SIGMOD 2016, San Francisco, CA, USA, June 26 - July 01, 2016 , page 10. ACM, 2016

  3. [3]

    Axel D. Becke. Density-functional thermochemistry. III. The role of exact exchange. The Journal of Chemical Physics , 98(7):5648–5652, 04

  4. [4]

    Paul G. Brown. Overview of scidb: large scale array storage, processing and analysis. In Proceedings of the 2010 ACM SIGMOD , SIGMOD ’10, page 963–968, New York, NY, USA, 2010. Association for Computing Machinery.doi:10.1145/ 1807167.1807271

  5. [5]

    Kolda, Stefan M

    Aydin Buluc, Tamara G. Kolda, Stefan M. Wild, Mihai Anitescu, Anthony De- gennaro, John D. Jakeman, Chandrika Kamath, Ramakrishnan Kannan, Miles E. Lopes, Per-Gunnar Martinsson, Kary Myers, Jelani Nelson, Juan Restrepo, C. Se- shadri, Draguna Vrabie, Brendt Wohlberg, Stephen J. Wright, Chao Yang, and Peter Zwart. Randomized algorithms for scientific compu...

  6. [6]

    Mirrokni

    Mayur Datar, Nicole Immorlica, Piotr Indyk, and Vahab S. Mirrokni. Locality- sensitive hashing scheme based on p-stable distributions. In Proceedings of the Twentieth Annual Symposium on Computational Geometry , SCG ’04, page 253–262, New York, NY, USA, 2004. Association for Computing Machinery. doi:10.1145/997817.997857

  7. [7]

    Mauvedb: supporting model-based user views in database systems

    Amol Deshpande and Samuel Madden. Mauvedb: supporting model-based user views in database systems. In Proceedings of the 2006 ACM SIGMOD, SIGMOD ’06, page 73–84, New York, NY, USA, 2006. Association for Computing Machinery

  8. [8]

    Altman, Roger Wattenhofer, and Kubilay Atasu

    Béni Egressy, Luc von Niederhäusern, Jovan Blanusa, Erik R. Altman, Roger Wattenhofer, and Kubilay Atasu. Provably powerful graph neural networks for directed multigraphs. In AAAI, pages 11838–11846. AAAI Press, 2024

Show all 36 references
  1. [9]

    A query language perspective on graph learning

    Floris Geerts. A query language perspective on graph learning. In Proceedings of the 42nd ACM SIGMOD-SIGACT-SIGAI Symposium on Principles of Database Systems, PODS 2023, Seattle, W A, USA, June 18-23, 2023 , pages 373–379. ACM, 2023

  2. [10]

    S.P. Ghosh. Data Base Organization for Data Management . Computer Science and Applied Mathematics. A Series of Monographs and Textobooks. Academic Press, 1977. URL: https://books.google.com/books?id=h-omAAAAMAAJ

  3. [11]

    Hohenberg and W

    P. Hohenberg and W. Kohn. Inhomogeneous electron gas. Phys. Rev., 136:B864– B871, Nov 1964. URL: https://link.aps.org/doi/10.1103/PhysRev.136.B864, doi: 10.1103/PhysRev.136.B864

  4. [12]

    Saibot: A differentially private data search platform

    Zezhou Huang, Jiaxiang Liu, Daniel Gbenga Alabi, Raul Castro Fernandez, and Eugene Wu. Saibot: A differentially private data search platform. Proceedings of the VLDB Endowment, 16(11):3057–3070, 2023

  5. [13]

    Anubhav Jain, Shyue Ping Ong, Geoffroy Hautier, Wei Chen, William Davidson Richards, Stephen Dacek, Shreyas Cholia, Dan Gunter, David Skinner, Gerbrand Ceder, and Kristin A. Persson. Commentary: The Materials Project: A mate- rials genome approach to accelerating materials inn...

  6. [14]

    Noscope: optimizing neural network queries over video at scale

    Daniel Kang, John Emmons, Firas Abuzaid, Peter Bailis, and Matei Zaharia. Noscope: optimizing neural network queries over video at scale. Proc. VLDB Endow., 10(11):1586–1597, aug 2017. doi:10.14778/3137628.3137664

  7. [15]

    Kenny, Cory McCartan, Shiro Kuriwaki, Tyler Simko, and Kosuke Imai

    Christopher T. Kenny, Cory McCartan, Shiro Kuriwaki, Tyler Simko, and Kosuke Imai. Evaluating bias and noise induced by the u.s. census bureau’s privacy protec- tion methods. Science Advances, 10(18):eadl2524, 2024. URL: https://www.science. org/doi/abs/10.1126/sciadv.adl2524,...

  8. [16]

    Kersten, Stratos Idreos, Stefan Manegold, and Erietta Liarou

    Martin L. Kersten, Stratos Idreos, Stefan Manegold, and Erietta Liarou. The researcher’s guide to the data deluge: Querying a scientific database in just a few seconds. Proc. VLDB Endow., 4(12):1474–1477, 2011. URL: http://www.vldb. org/pvldb/vol4/p1474-kersten.pdf

  9. [17]

    Fischer, David N

    Benedikt Knüsel, Marius Zumwald, Christophe Baumberger, Gertrude Hirsch Hadorn, Erich M. Fischer, David N. Bresch, and Reto Knutti. Applying big data beyond small problems in climate research. Nature Climate Change, 9:196–202,

  10. [18]

    Kohn and L

    W. Kohn and L. J. Sham. Self-consistent equations including exchange and correlation effects. Phys. Rev., 140:A1133–A1138, Nov 1965. URL: https://link. aps.org/doi/10.1103/PhysRev.140.A1133, doi:10.1103/PhysRev.140.A1133

  11. [19]

    David M. J. Lazer, Alex Pentland, Duncan J. Watts, Sinan Aral, Susan Athey, Noshir Contractor, Deen Freelon, Sandra Gonzalez-Bailon, Gary King, Helen Margetts, Alondra Nelson, Matthew J. Salganik, Markus Strohmaier, Alessan- dro Vespignani, and Claudia Wagner. Computational so...

  12. [20]

    Mai, Pengyu Wang, Azza Abouzied, Matteo Brucato, Peter J

    Anh L. Mai, Pengyu Wang, Azza Abouzied, Matteo Brucato, Peter J. Haas, and Alexandra Meliou. Scaling package queries to a billion tuples via hierarchical partitioning and customized optimization. Proc. VLDB Endow., 17(5):1146–1158, 2024

  13. [21]

    Leo: An autonomic query optimizer for db2

    Volker Markl, Guy M Lohman, and Vijayshankar Raman. Leo: An autonomic query optimizer for db2. IBM Systems Journal, 42(1):98–106, 2003

  14. [22]

    Franklin, Reza Zadeh, Matei Zaharia, and Ameet Talwalkar

    Xiangrui Meng, Joseph Bradley, Burak Yavuz, Evan Sparks, Shivaram Venkatara- man, Davies Liu, Jeremy Freeman, DB Tsai, Manish Amde, Sean Owen, Doris Xin, Reynold Xin, Michael J. Franklin, Reza Zadeh, Matei Zaharia, and Ameet Talwalkar. Mllib: machine learning in apache spark. ...

  15. [23]

    Scaling deep learning for materials discovery

    Amil Merchant, Simon Batzner, Samuel Schoenholz, Muratahan Aykol, Gowoon Cheon, and Ekin Cubuk. Scaling deep learning for materials discovery. Nature, 624:1–6, 11 2023. doi:10.1038/s41586-023-06735-9

  16. [24]

    Reproducibility and Replicability in Science

    National Academies of Sciences & Engineering & and Medicine. Reproducibility and Replicability in Science . The National Academies Press, Washington, DC,

  17. [25]

    Governor hochul unveils fifth proposal of 2024 state of the state: Empire ai consortium to make new york the national leader in ai research and innovation

    New York Governor Press Office. Governor hochul unveils fifth proposal of 2024 state of the state: Empire ai consortium to make new york the national leader in ai research and innovation. https://www.governor.ny.gov/news/governor- hochul-unveils-fifth-proposal-2024-state-state...

  18. [26]

    Parr and Y

    R.G. Parr and Y. Weitao. Density-Functional Theory of Atoms and Molecules . International Series of Monographs on Chemistry. Oxford University Press, 1994. URL: https://books.google.com/books?id=mGOpScSIwU4C

  19. [27]

    URL: https://nap.nationalacademies.org/catalog/25303/reproducibility- and-replicability-in-science, doi:10.17226/25303

  20. [28]

    everyone wants to do the model work, not the data work

    Nithya Sambasivan, Shivani Kapania, Hannah Highfill, Diana Akrong, Praveen Paritosh, and Lora M Aroyo. “everyone wants to do the model work, not the data work”: Data cascades in high-stakes ai. In Proceedings of the 2021 CHI Conference , CHI ’21, 2021

  21. [29]

    Efficient model-relational data manage- ment: Challenges and opportunities

    Viktor Sanca and Anastasia Ailamaki. Efficient model-relational data manage- ment: Challenges and opportunities. IEEE Transactions on Knowledge and Data Engineering, pages 1–12, 2024. doi:10.1109/TKDE.2024.3384276

  22. [30]

    Database Management Systems

    Raghu Ramakrishnan and Johannes Gehrke. Database Management Systems . McGraw-Hill, Inc., USA, 3 edition, 2002

  23. [31]

    Spglib: a software library for crystal symmetry search, 2018

    Atsushi Togo and Isao Tanaka. Spglib: a software library for crystal symmetry search, 2018. arXiv:1808.01590

  24. [32]

    Wenisch, R.E

    T.F. Wenisch, R.E. Wunderlich, M. Ferdman, A. Ailamaki, B. Falsafi, and J.C. Hoe. Simflex: Statistical sampling of computer system simulation. IEEE Micro, 26(4):18–31, 2006

  25. [33]

    An autonomous laboratory for the accelerated synthesis of novel materials

    Nathan Szymanski, Bernardus Rendy, Yuxing Fei, Rishi Kumar, Tanjin He, David Milsted, Matthew McDermott, Max Gallant, Ekin Cubuk, Amil Merchant, Hae- gyeom Kim, Anubhav Jain, Chris Bartel, Kristin Persson, Yan Zeng, and Gerbrand Ceder. An autonomous laboratory for the accelera...

  26. [36]

    Sciql: bridging the gap between science and relational dbms

    Ying Zhang, Martin Kersten, Milena Ivanova, and Niels Nes. Sciql: bridging the gap between science and relational dbms. In Proceedings of the 15th symposium on international database engineering & applications , pages 124–133, 2011

  27. [1993]

    arXiv:https://pubs.aip.org/aip/jcp/article-pdf/98/7/5648/ 19277469/5648\_1\_online.pdf, doi:10.1063/1.464913

  28. [2019]

    URL: https://api.semanticscholar.org/CorpusID:91553899

Pith tools

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