Pith. sign in

REVIEW 3 major objections 5 minor 55 references

STZ: A High Quality and High Speed Streaming Lossy Compression Framework for Scientific Data

T0 review · 3 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper proposes STZ, a streaming lossy compressor that supports both progressive and random-access decompression while matching SZ3's quality and decompressing up to 6.7x faster.

desk verdict STZ is a genuinely useful streaming compressor and the first I know of to do both progressive and random-access decompression in one framework; the quality and speed claims are plausible, but the missing global error-bound accounting makes the headline speed comparison hard to audit. read the letter →

arxiv 2509.01626 v1 pith:N6QPXA5M submitted 2025-09-01 cs.DC cs.MM

Daoce Wang , Pascal Grosset , Jesus Pulido , Jiannan Tian , Tushar M. Athawale , Jinda Jia , Baixi Sun , Boyuan Zhang
show 4 more authors
Sian Jin Kai Zhao James Ahrens Fengguang Song
This is my paper · ORCID
classification cs.DCcs.MM
keywords error-boundedlossycompressionprogressivedecompressionrandom-accesshierarchicalpredictioncubicsplineinterpolationscientificdatastreaming
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 proposes STZ, a streaming lossy compression framework and claims it is the first to support both progressive decompression and random-access decompression at the same time. The central argument is that hierarchical partitioning does not have to ruin compression quality if a coarse sub-block predicts the finer levels through multi-dimensional interpolation; with that mechanism, STZ reports rate-distortion performance comparable to SZ3, a non-streaming state-of-the-art compressor, and decompression up to 6.7x faster. If the claim holds, users of terabyte- or petabyte-scale simulation data could inspect coarse previews, choose a region of interest, and reconstruct only that region without paying the quality or speed penalty that streaming features previously imposed. The evidence is a rate-distortion and timing evaluation on four large scientific datasets against SZ3, ZFP, MGARD-X, and SPERR.

What carries the argument

The central mechanism is the hierarchical partition-and-predict cascade. STZ subsamples the grid with stride 2 into eight sub-blocks; the coarsest of these (stride-4 sampling, 1/64 of the data in the three-level design) is compressed with SZ3, and every finer point is predicted from neighboring coarser points using multi-dimensional cubic spline interpolation before quantization and Huffman encoding. This makes finer levels depend only on coarser levels, which is the property that enables both random access (only needed points are predicted) and progressive reconstruction (each level adds resolution), while the interpolation recovers the spatial redundancy lost by partitioning.

What would settle it

Compress Nyx's baryon density field with STZ under a strict user-specified error bound (say 1e-4), reconstruct the full volume, and measure the pointwise maximum absolute error; if it exceeds the requested bound, the framework does not deliver its advertised global error control. The same test could vary the level-2/level-1 bound ratio around 2.5 and look for bound violations, since the paper's ratio is empirically chosen and error propagation from level 1 is acknowledged.

Watch

Extended reading notes

Core claim

The paper's core discovery is that direct hierarchical partitioning can be made compatible with high-quality prediction by reversing the direction of dependency: compress the coarsest level, then predict every finer point from a small stencil of already-reconstructed coarser neighbors. In STZ the 3D grid is partitioned into eight stride-2 sub-blocks; a three-level variant makes the coarsest level 1/64 of the data. The coarsest level is compressed with SZ3, and all finer levels are predicted from it using linear, bilinear, trilinear, or cubic spline interpolation, with only the residuals quantized and Huffman-coded. Because no point at a finer level depends on other points at that same level,

Load-bearing premise

The load-bearing premise is that the per-level error bounds (the finer level set to 2.5 times the coarser one) keep the whole reconstruction within the user's global error bound, but the paper gives no proof of that and tunes the ratio empirically.

Editorial extensions

If this is right

  • A 2D slice can be pulled from a 3D volume without reconstructing the rest; the paper measures up to 82.5% decompression-time savings for slice access on Miranda.
  • Coarse previews require decompressing only 1.6% of the data, enabling a visualization-first workflow where the user finds a region of interest and then requests full resolution for just that region.
  • The streaming features no longer force a quality trade-off: STZ stays within SZ3-level rate-distortion on the tested datasets while decompressing up to 6.7x faster.
  • The three-level scheme can be extended to four or more levels for larger grids, shrinking the coarsest footprint to 1/512 of the volume.

Reading between the lines

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

  • My inference: the fixed 2.5 ratio between level-2 and level-1 error bounds is probably dataset- and field-dependent; a self-tuning version that checks the reconstructed maximum error would make the global error-bound claim testable.
  • My inference: because finer points depend only on coarser neighbors, the same cascade is likely to map well onto GPUs, extending the speed advantage beyond the OpenMP results reported here.
  • My inference: replacing Huffman coding with a random-access-friendly entropy coder would convert the near-100% prediction savings into real decoding savings for 3D region-of-interest access, since the paper shows that whole-sub-block Huffman decoding is currently the bottleneck.
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. This paper introduces STZ, a streaming lossy compression framework for scientific floating-point data that supports both resolution-progressive decompression and random-access decompression. STZ partitions the domain into a stride-sampled hierarchy; the coarsest level is compressed with SZ3, and finer levels are predicted from decompressed coarser levels using multidimensional cubic interpolation (with boundary handling), then residual-quantized and Huffman-encoded. The authors report rate-distortion results on four datasets (Nyx, WarpX, Magnetic Reconnection, Miranda) against SZ3, ZFP, MGARD-X, and SPERR, showing PSNR comparable to SZ3, and speed experiments claiming up to 6.7x faster decompression and up to 4.7x faster compression than SZ3. The paper also evaluates progressive and random-access decompression times on Miranda.

Significance. The contribution is significant if validated: it addresses a real functional gap (no existing scientific lossy compressor provides both progressive and random-access decompression) and does so with a simple partitioning plus predictive-residual design that is faster than SZ3. The manuscript contains useful algorithmic detail: precise interpolation stencils, a clear explanation of the dependency structure, and a time breakdown (Table 4) that separates decode, prediction, and reconstruction costs. The experimental breadth (four datasets, four baselines, serial and OpenMP) is reasonable. However, the error-boundedness of STZ is not established and the speed comparison omits numeric error bounds, so the headline claims are not yet fully supported.

major comments (3)
  1. [§3.1, Prediction Optimization 5; §4.3] The paper never states how a user-specified error bound eb is mapped to eb_l1 and eb_l2, and it does not prove an end-to-end error bound. Equations (6)–(8) predict level-2/3 points from decompressed level-1 data, so interpolation error from lossy level-1 reconstruction enters the final reconstruction independently of the residual quantization at levels 2/3. The text sets eb_l2 = 2.5 × eb_l1 based on 'extensive experiments' but provides no propagation analysis, no worst-case error statement, and no actual max-error measurements. Since the abstract and Section 1 place STZ in the error-bounded compressor class, this is a load-bearing gap: the comparisons to SZ3 at 'same error bound' or 'similar compression quality' cannot be interpreted without knowing STZ's effective maximum error. Please provide the mapping and a proof or conservative bound, and report actual max errors in the experimenta
  2. [§4.3, Table 3] The numeric error bounds used for each dataset and each method are not reported. The text says bounds were 'selected to produce similar compression quality' but gives no values, measured PSNR/CR, or per-row error bounds. The headline speedup claim ('up to 6.7× decompression') is therefore not independently verifiable, particularly because STZ's effective error depends on the adaptive ratio from Optimization 5. Report the exact bound configuration and quality metrics for every row of Table 3, or provide a separate fixed-error-bound speed comparison.
  3. [§4.2, Figure 11] Rate-distortion is evaluated as PSNR at fixed compression ratio. For an error-bounded compressor, the more relevant comparison is compression ratio at a fixed user-specified error bound (or at fixed measured maximum error). Figure 11 does not show which error bounds were used or whether the intended bound is respected. Please add fixed-error-bound CR curves or a table of CRs at the same nominal error bound for all methods, with actual max errors reported.
minor comments (5)
  1. [§3.1, Equations (3)–(5)] Equations (3)–(5) are typeset incorrectly (missing plus signs/operands); please correct them.
  2. [§3.1, Figure 5 and text] 'gray cycle' should be 'gray circle'; the legend labels in Figure 5 are not all defined in the text (e.g., '3-level + All').
  3. [§3.3 and §4.5] Decoding savings are stated as 57%, 58%, and 'up to 57%' in different places; please reconcile and report the exact values and measurement basis.
  4. [§4.1] Baseline compressor settings are described only as 'default settings'; please specify important parameters (e.g., ZFP accuracy vs rate mode, MGARD tolerance type, SPERR settings) to support reproducibility.
  5. [General] No code or data availability statement is included. Please clarify whether the STZ implementation and test datasets are publicly available.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the adaptive error-bound ratio is an empirically disclosed tuning parameter, the comparison to SZ3 is an external benchmark rather than a derivation from SZ3, and no prediction reduces to its fitted inputs.

full rationale

The paper's central claims are (1) first simultaneous progressive and random-access decompression, (2) compression quality comparable to SZ3, and (3) up to 6.7x speedups. None of these reduces to its own inputs by construction. The most suspicious point is Prediction Optimization 5, where eb_l2 = 2.5 * eb_l1 is set 'based on extensive experiments.' This is an empirically tuned hyperparameter, disclosed as such, not a derived prediction. The paper does not claim to prove this ratio from first principles, so this is tuning, not circularity. SZ3 appears both as a component (used to compress the coarsest hierarchical level) and as a baseline. This is not circular: STZ uses SZ3 on only 1.6%/12.5% of the data, whereas the baseline SZ3 compresses the full dataset. The quality comparison is an external benchmark, not a self-referential derivation. The only in-text self-citation is the cubic spline formula credited to [47], a prior paper by co-author Kai Zhao. The formula is standard numerical interpolation and is not load-bearing as a 'first-principles result' or as a uniqueness claim. It does not smuggle in an unverified ansatz that the central claim depends on. The paper does have a verification gap: Section 4.3 says error bounds were 'selected to produce similar compression quality' but Table 3 omits the numeric error bounds, and the paper never states how a user-specified global error bound maps to (eb_l1, eb_l2) or proves the end-to-end reconstruction error is bounded. This is a reproducibility/correctness concern, not a circularity: the speed and quality numbers could be affected by error-bound choices, but the paper's claims are not equivalent to its inputs by construction. Overall, no circular step is exhibited, so the circularity score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central method rests on one tuned scalar (the 2.5x error-bound ratio), unreported per-dataset error-bound choices, and domain assumptions about spatial coherence and SZ3's error control. No new physical entities are introduced.

free parameters (2)
  • Adaptive error-bound ratio eb_l2/eb_l1 = 2.5
    Section 3.1, Prediction Optimization 5: the ratio 2.5 was chosen after 'extensive experiments on different datasets' to maximize overall compression performance. It is a hand-tuned global hyperparameter, not derived from an error-bound requirement.
  • Per-dataset error bounds used in speed comparison = not reported
    Section 4.3 states 'we selected error bounds that produce similar compression quality for SZ3 and our solution', but the values are not listed, so the speed/quality comparison cannot be independently reproduced.
assumptions (4)
  • domain assumption Regular-grid spatial coherence: neighboring samples in scientific simulation data are correlated enough for multidimensional interpolation to reduce prediction residuals.
    Required throughout Section 3.1; the whole prediction scheme assumes spatial redundancy between sub-blocks.
  • domain assumption SZ3 is a valid error-bounded compressor when applied to the coarsest level sub-blocks.
    Section 3.1 uses SZ3 to compress level 1 and relies on its error control and speed without re-verifying its behavior on subsampled data.
  • domain assumption Prediction residuals after multidimensional interpolation are sufficiently decorrelated that a second prediction step is unnecessary.
    Prediction Optimization 3, Section 3.1, asserts the differences are relatively random and that a second SZ3 prediction would waste time and degrade quality.
  • ad hoc to paper Error propagation from level 1 into levels 2 and 3 remains within a controllable bound under the 2.5x error-bound ratio.
    Implicitly assumed in Section 3.1, Prediction Optimization 5; the paper states level-1 error propagates but never derives a worst-case global error bound.

how reviews work

0 comments
Cite this review

Pith. "Pith review of STZ: A High Quality and High Speed Streaming Lossy Compression Framework for Scientific Data." pith.science (2026). https://pith.science/paper/N6QPXA5M

@misc{pith2026250901626,
  author       = {Pith},
  title        = {Pith review of: STZ: A High Quality and High Speed Streaming Lossy Compression Framework for Scientific Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N6QPXA5M}},
  note         = {Machine review of arXiv:2509.01626}
}
abstract

Error-bounded lossy compression is one of the most efficient solutions to reduce the volume of scientific data. For lossy compression, progressive decompression and random-access decompression are critical features that enable on-demand data access and flexible analysis workflows. However, these features can severely degrade compression quality and speed. To address these limitations, we propose a novel streaming compression framework that supports both progressive decompression and random-access decompression while maintaining high compression quality and speed. Our contributions are three-fold: (1) we design the first compression framework that simultaneously enables both progressive decompression and random-access decompression; (2) we introduce a hierarchical partitioning strategy to enable both streaming features, along with a hierarchical prediction mechanism that mitigates the impact of partitioning and achieves high compression quality -- even comparable to state-of-the-art (SOTA) non-streaming compressor SZ3; and (3) our framework delivers high compression and decompression speed, up to 6.7$\times$ faster than SZ3.

Figures

Figures reproduced from arXiv: 2509.01626 by the authors.

Figure 1
Figure 1. Iso-surface of original (top, 256×256×2048, 1𝐺𝐵) and down￾sampled (bottom, 128 × 128 × 1024, 128𝑀𝐵) WarpX dataset. Visual differences between them are virtually imperceptible (SSIM=0.96), highlighting the need for progressive decompression. Although lossy compression can significantly reduce the storage and transmission overhead, traditional compression workflows require the entire compressed data to be fully de￾com… view at source ↗
Figure 2
Figure 2. Overview of our proposed STZ 3 Design Methodology In this section, we introduce our proposed streaming lossy compression framework, STZ, illustrated in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Visual comparison (one 2× zoom in 2D slice) of original data and decompressed data produced by partition, SZ3, and our STZ on Nyx’s “baryon density” field. The “Rainbow Blended White” colormap in ParaView is used. The CR of partition, SZ3, and ours are 204, 205, and 206 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: 2D example of 2-level partition. 0 50 100 150 200 250 300 350 110 120 130 CR PSNR Partition Direct pred Multi-dim Interp Multi-dim + Qt Cubic-Multi + Qt SZ3 Cubic-Multi-Qt + Adp 3-level + All [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Rate-distortion comparison of direct partition, our optimiza [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: 2D example of 2-level direct prediction (left) and multi [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: Visualization of different interpolation cases in 3D data. [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 9
Figure 9. Figure 9: 2D example of three-level partition (right) and hierarchical interpolation prediction (left & center), linear interpolation was used for [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Vis of the ROI subset (right, 0.69% of the dataset) extracted [PITH_FULL_IMAGE:figures/full_fig_p008_10.png]
Figure 11
Figure 11. Figure 11: Rate-distortion of our solution and baselines on different datasets (top-right is better). [PITH_FULL_IMAGE:figures/full_fig_p009_11.png]
Figure 12
Figure 12. Figure 12: Visual comparison (iso-surface and zoom in 2D slice) of the original data and decompressed data produced by baseline compressors [PITH_FULL_IMAGE:figures/full_fig_p010_12.png]
Figure 13
Figure 13. Figure 13: Volume rendering of the original data of the Miranda dataset and progressively decompressed data produced by our STZ at different [PITH_FULL_IMAGE:figures/full_fig_p012_13.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 48 canonical work pages

  1. [1]

    Mark Ainsworth, Ozan Tugluk, Ben Whitney, and Scott Klasky

    work page

  2. [2]

    Allison H Baker, Dorit M Hammerling, and Terece L Turton

    work page

  3. [3]

    Rafael Ballester-Ripoll, Peter Lindstrom, and Renato Pajarola

    work page

  4. [4]

    Harsh Bhatia, Duong Hoang, Nate Morrical, Valerio Pascucci, Peer-Timo Bremer, and Peter Lindstrom. 2022. AMM: Adaptive multilinear meshes. IEEE Transactions on Visualization and Computer Graphics28, 6 (2022), 2350–2363

    work page 2022

  5. [5]

    Franck Cappello, Sheng Di, Sihuan Li, Xin Liang, Ali Murat Gok, Dingwen Tao, Chun Hong Yoon, Xin-Chuan Wu, Yuri Alexeev, and Frederic T Chong. 2019. Use cases of lossy compression for floating-point data in scientific data sets.The International Journal of High Performance Computing Applications(2019)

    work page 2019

  6. [6]

    Xinyu Chen, Jiannan Tian, Ian Beaver, Cynthia Freeman, Yan Yan, Jianguo Wang, and Dingwen Tao. 2024. FCBench: Cross- Domain Benchmarking of Lossless Compression for Floating-Point Data. Proceedings of the VLDB Endowment17, 6 (2024), 1418– 1431

    work page 2024

  7. [7]

    Cook, William Cabot, and Paul L

    Andrew W. Cook, William Cabot, and Paul L. Miller. 2004. The Mixing Transition in Rayleigh-Taylor Instability. Journal of Fluid Mechanics511 (2004), 333–362. https://doi.org/10.1017/ S0022112004009681

    work page 2004

  8. [8]

    cuZFP. 2023. https://github.com/LLNL/zfp/tree/develop/src/ cuda_zfp. Online

    work page 2023

Show all 55 references
  1. [9]

    Marc Davis, George Efstathiou, Carlos S Frenk, and Simon DM White. 1985. The evolution of large-scale structure in a universe dominated by cold dark matter.The Astrophysical Journal292 (1985), 371–394

    1985

  2. [10]

    1996.GZIP file format specification version 4.3

    Peter Deutsch. 1996.GZIP file format specification version 4.3. Technical Report

    1996

  3. [11]

    Sheng Di and Franck Cappello. 2016. Fast error-bounded lossy HPC data compression with SZ. In2016 IEEE International Parallel and Distributed Processing Symposium. IEEE, 730–739

    2016

  4. [12]

    Sheng Di, Jinyang Liu, Kai Zhao, Xin Liang, Robert Under- wood, Zhaorui Zhang, Milan Shah, Yafan Huang, Jiajun Huang, Xiaodong Yu, et al. 2024. A Survey on Error-Bounded Lossy Com- pression for Scientific Datasets.arXiv preprint arXiv:2404.02840 (2024)

    2024 arXiv

  5. [13]

    Bo Fang, Daoce Wang, Sian Jin, Quincey Koziol, Zhao Zhang, Qiang Guan, Surendra Byna, Sriram Krishnamoorthy, and Ding- wen Tao. 2021. Characterizing Impacts of Storage Faults on HPC Applications: A Methodology and Insights. 409–420. https: //doi.org/10.1109/Cluster48925.2021.00048

    2021

  6. [14]

    Fedeli, A

    L. Fedeli, A. Huebl, F. Boillod-Cerneux, T. Clark, K. Gott, C. Hillairet, S. Jaure, A. Leblanc, R. Lehe, A. Myers, C. Piechurski, M. Sato, N. Zaim, W. Zhang, J. Vay, and H. Vincenti. 2022. Pushing the Frontier in the Design of Laser-Based Electron Ac- celerators with Groundbre...

    2022 arXiv

  7. [15]

    Qian Gong, Jieyang Chen, Ben Whitney, Xin Liang, Viktor Reshniak, Tania Banerjee, Jaemoon Lee, Anand Rangarajan, Lipeng Wan, Nicolas Vidal, Qing Liu, Ana Gainaru, Norbert Podhorszki, Richard Archibald, Sanjay Ranka, and Scott Klasky

  8. [16]

    Fan Guo, Hui Li, William Daughton, and Yi-Hsin Liu. 2014. Formation of Hard Power Laws in the Energetic Particle Spectra Resulting from Relativistic Magnetic Reconnection.Phys. Rev. Lett. 113 (oct 2014), 155005. Issue 15. https://doi.org/10.1103/ PhysRevLett.113.155005

    2014

  9. [17]

    Yafan Huang, Sheng Di, Guanpeng Li, and Franck Cappello. 2024. cuSZp2: A GPU Lossy Compressor with Extreme Throughput and Optimized Compression Ratio. InSC24: International Con- ference for High Performance Computing, Networking, Storage and Analysis. IEEE, 1–18

    2024

  10. [18]

    Axel Huebl, René Widera, Felix Schmitt, Alexander Matthes, Norbert Podhorszki, Jong Youl Choi, Scott Klasky, and Michael Bussmann. 2017. On the Scalability of Data Reduction Techniques in Current and Upcoming HPC Systems from an Application Perspective. InHigh Performance Comp...

    2017

  11. [19]

    Jinda Jia, Cong Xie, Hanlin Lu, Daoce Wang, Hao Feng, Cheng- ming Zhang, Baixi Sun, Haibin Lin, Zhi Zhang, Xin Liu, and Dingwen Tao. 2024. SDP4Bit: Toward 4-bit Communication Quantization in Sharded Data Parallelism for LLM Training. In Advances in Neural Information Processin...

    2024

  12. [20]

    Wenqi Jia, Zhewen Hu, Youyuan Liu, Boyuan Zhang, Jinzhen Wang, Jinyang Liu, Wei Niu, Stavros Kalafatis, Junzhou Huang, Sian Jin, et al. 2024. NeurLZ: An Online Neural Learning-Based Method to Enhance Scientific Lossy Compression.arXiv preprint arXiv:2409.05785 (2024)

    2024 arXiv

  13. [21]

    Sian Jin, Sheng Di, Frédéric Vivien, Daoce Wang, Yves Robert, Dingwen Tao, and Franck Cappello. 2024. Concealing Compression-accelerated I/O for HPC Applications through In Situ Task Scheduling. InEuroSys 2024

    2024

  14. [22]

    Shaomeng Li, Peter Lindstrom, and John Clyne. 2023. Lossy Sci- entific Data Compression With SPERR. In2023 IEEE Interna- tional Parallel and Distributed Processing Symposium (IPDPS). 1007–1017. https://doi.org/10.1109/IPDPS54959.2023.00104

    2023

  15. [23]

    Xin Liang, Sheng Di, Dingwen Tao, Sihuan Li, Shaomeng Li, Hanqi Guo, Zizhong Chen, and Franck Cappello. 2018. Error- controlled lossy compression optimized for high compression ratios of scientific datasets. In2018 IEEE International Conference on Big Data. IEEE, 438–447

    2018

  16. [24]

    Xin Liang, Qian Gong, Jieyang Chen, Ben Whitney, Lipeng Wan, Qing Liu, David Pugmire, Rick Archibald, Norbert Podhorszki, and Scott Klasky. 2021. Error-controlled, progressive, and adapt- able retrieval of scientific data with multilevel decomposition. In Proceedings of the In...

    2021

  17. [25]

    Peter Lindstrom. 2014. Fixed-rate compressed floating-point arrays. IEEE Transactions on Visualization and Computer Graphics 20, 12 (2014), 2674–2683

    2014

  18. [26]

    Jinyang Liu, Jiannan Tian, Shixun Wu, Sheng Di, Boyuan Zhang, Robert Underwood, Yafan Huang, Jiajun Huang, Kai Zhao, Guan- peng Li, et al. 2024. CUSZ-i: High-Ratio Scientific Lossy Com- pression on GPUs with Optimized Multi-Level Interpolation. In SC24: International Conferenc...

    2024

  19. [27]

    Tao Lu, Qing Liu, Xubin He, Huizhang Luo, Eric Suchyta, Jong Choi, Norbert Podhorszki, Scott Klasky, Mathew Wolf, Tong Liu, et al. 2018. Understanding and modeling lossy compression schemes on HPC scientific data. In2018 IEEE International Parallel and Distributed Processing S...

    2018

  20. [28]

    Huizhang Luo, Dan Huang, Qing Liu, Zhenbo Qiao, Hong Jiang, Jing Bi, Haitao Yuan, Mengchu Zhou, Jinzhen Wang, and Zhenlu Qin. 2019. Identifying Latent Reduced Models to Precondition Lossy Compression. In2019 IEEE International Parallel and Distributed Processing Symposium. IEEE

    2019

  21. [29]

    Victor AP Magri and Peter Lindstrom. 2023. A general framework for progressive data compression and retrieval.IEEE Transac- tions on Visualization and Computer Graphics30, 1 (2023), Wang et al. 1358–1368

    2023

  22. [30]

    NYX simulation. 2019. https://amrex-astro.github.io/Nyx/. On- line

    2019

  23. [31]

    Oak Ridge Leadership Computing Facility. 2023. WarpX, granted early access to the exascale supercomputer Fron- tier, receives the high-performance computing world’s high- est honor. https://www.olcf.ornl.gov/2022/11/17/plasma- simulation-code-wins-2022-acm-gordon-bell-prize/ Online

    2023

  24. [32]

    Gregory Pauloski, Jiannan Tian, Jinda Jia, Daoce Wang, Boyuan Zhang, Mingkai Zheng, Sheng Di, Sian Jin, Zhao Zhang, Xiaodong Yu, Kamil A

    Baixi Sun, Weijin Liu, J. Gregory Pauloski, Jiannan Tian, Jinda Jia, Daoce Wang, Boyuan Zhang, Mingkai Zheng, Sheng Di, Sian Jin, Zhao Zhang, Xiaodong Yu, Kamil A. Iskra, Pete Beckman, Guangming Tan, and Dingwen Tao. 2025. COMPSO: Optimizing GradientCompressionforDistributedTr...

    2025

  25. [33]

    Dingwen Tao, Sheng Di, Zizhong Chen, and Franck Cappello. 2017. Significantly improving lossy compression for scientific data sets based on multidimensional prediction and error-controlled quan- tization. In2017 IEEE International Parallel and Distributed Processing Symposium....

    2017

  26. [34]

    Jiannan Tian, Sheng Di, Xiaodong Yu, Cody Rivera, Kai Zhao, Sian Jin, Yunhe Feng, Xin Liang, Dingwen Tao, and Franck Cappello. 2021. Optimizing error-bounded lossy compression for scientific data on GPUs. In2021 IEEE International Conference on Cluster Computing (CLUSTER). IEE...

    2021

  27. [35]

    Jiannan Tian, Sheng Di, Kai Zhao, Cody Rivera, Megan Hickman Fulp, Robert Underwood, Sian Jin, Xin Liang, Jon Calhoun, Dingwen Tao, and Franck Cappello. 2020. cuSZ: An Efficient GPU-Based Error-Bounded Lossy Compression Framework for Scientific Data. (2020), 3–15

    2020

  28. [36]

    Alex Ulmer, Marco Angelini, Jean-Daniel Fekete, Jörn Kohlham- mer, and Thorsten May. 2024. A Survey on Progressive Visu- alization. IEEE Transactions on Visualization and Computer Graphics 30, 9 (2024), 6447–6467. https://doi.org/10.1109/ TVCG.2023.3346641

    2024

  29. [37]

    Daoce Wang. 2025. Designing Efficient Data Reduction Ap- proaches for Multi-Resolution Simulations on HPC Systems. Ph.D. Dissertation. Indiana University

    2025

  30. [38]

    Athawale, Jiannan Tian, Kai Zhao, Zarija Lukić, Axel Huebl, Zhe Wang, James Ahrens, and Dingwen Tao

    Daoce Wang, Pascal Grosset, Jesus Pulido, Tushar M. Athawale, Jiannan Tian, Kai Zhao, Zarija Lukić, Axel Huebl, Zhe Wang, James Ahrens, and Dingwen Tao. 2024. A High-Quality Work- flow for Multi-Resolution Scientific Data Reduction and Visu- alization. InSC24: International Co...

    2024 arXiv

  31. [39]

    Daoce Wang, Jesus Pulido, Pascal Grosset, Sian Jin, Jiannan Tian, James Ahrens, and Dingwen Tao. 2022. TAC: Optimizing Error-Bounded Lossy Compression for Three-Dimensional Adap- tive Mesh Refinement Simulations. InProceedings of the 31st International Symposium on High-Perfor...

    2022

  32. [40]

    Daoce Wang, Jesus Pulido, Pascal Grosset, Sian Jin, Jiannan Tian, Kai Zhao, James Ahrens, and Dingwen Tao. 2024. TAC+: Optimizing Error-Bounded Lossy Compression for 3D AMR Simu- lations. IEEE Transactions on Parallel and Distributed Systems 35, 3 (2024), 421–438

    2024

  33. [41]

    Daoce Wang, Jesus Pulido, Pascal Grosset, Jiannan Tian, James Ahrens, and Dingwen Tao. 2023. Analyzing impact of data re- duction techniques on visualization for amr applications using amrex framework. InProceedings of the SC’23 Workshops of The International Conference on Hig...

    2023

  34. [42]

    Daoce Wang, Jesus Pulido, Pascal Grosset, Jiannan Tian, Sian Jin, Houjun Tang, Jean Sexton, Sheng Di, Kai Zhao, Bo Fang, et al

  35. [43]

    Zhou Wang, Alan C Bovik, Hamid R Sheikh, and Eero P Simon- celli. 2004. Image quality assessment: from error visibility to structural similarity.IEEE transactions on image processing13, 4 (2004), 600–612

    2004

  36. [44]

    Xuan Wu, Qian Gong, Jieyang Chen, Qing Liu, Norbert Pod- horszki, Xin Liang, and Scott Klasky. 2024. Error-controlled Progressive Retrieval of Scientific Data under Derivable Quan- tities of Interest. InSC24: International Conference for High Performance Computing, Networking,...

    2024

  37. [45]

    Zhuoxun Yang, Sheng Di, Longtao Zhang, Ruoyu Li, Ximiao Li, Jiajun Huang, Jinyang Liu, Franck Cappello, and Kai Zhao. 2025. IPComp: Interpolation Based Progressive Lossy Compression for Scientific Applications. arXiv:2502.04093 [cs.DC] https: //arxiv.org/abs/2502.04093

    2025 arXiv

  38. [46]

    Emanuel Zgraggen, Alex Galakatos, Andrew Crotty, Jean-Daniel Fekete, and Tim Kraska. 2017. How Progressive Visualiza- tions Affect Exploratory Analysis.IEEE Transactions on Vi- sualization and Computer Graphics23, 8 (2017), 1977–1987. https://doi.org/10.1109/TVCG.2016.2607714

    2017

  39. [47]

    Kai Zhao, Sheng Di, Maxim Dmitriev, Thierry-Laurent D Tonellot, Zizhong Chen, and Franck Cappello. 2021. Optimizing error- bounded lossy compression for scientific data by dynamic spline interpolation. In2021 IEEE 37th International Conference on Data Engineering (ICDE). IEEE,...

    2021

  40. [48]

    InPro- ceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis

    AMRIC: A Novel In Situ Lossy Compression Framework for Efficient I/O in Adaptive Mesh Refinement Applications. InPro- ceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis. 1–15

  41. [49]

    Zstandard. 2020. http://facebook.github.io/zstd/

    2020

  42. [54]

    Qing Zheng, Brian Atkinson, Daoce Wang, Jason Lee, John Patch- ett, Dominic Manno, and Gary Grider. 2024. Accelerating Viz Pipelines Using Near-Data Computing: An Early Experience. In SC24-W: Workshops of the International Conference for High Performance Computing, Networking,...

    2024

  43. [2018]

    Multilevel techniques for compression and reduction of scientific data—the univariate case.Computing and Visualization in Science19, 5–6 (2018), 65–76

    2018

  44. [2019]

    In Computer Graphics Forum, Vol

    Evaluating image quality measures to assess the impact of lossy data compression applied to climate simulation data. In Computer Graphics Forum, Vol. 38. Wiley Online Library, 517–528

  45. [2020]

    https://doi.org/10.1109/ TVCG.2019.2904063

    TTHRESH: Tensor Compression for Multidimensional Visual Data.IEEE Transactions on Visualization and Computer Graphics 26, 9 (2020), 2891–2903. https://doi.org/10.1109/ TVCG.2019.2904063

    2020

  46. [2023]

    MGARD: A multigrid framework for high-performance, error-controlled data compression and refactoring.SoftwareX 24 (2023), 101590

    2023

  47. [8759]

    https://proceedings.neurips.cc/paper_files/paper/2024/ file/10826a1a80f816ea98d559d7c7a97973-Paper-Conference.pdf

    2024

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.