Pith. sign in

REVIEW 4 major objections 4 minor 64 references

HybHuff: Lossless Compression for Hypergraphs via Entropy-Guided Huffman-Bitwise Coordination

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A frequency-guided split between Huffman and bitwise coding lets HybHuff compress hypergraph adjacency data up to 2.3× smaller than zip while keeping decode overhead low.

desk verdict HybHuff's engineering is plausible; its theorem isn't proven and its baselines aren't fair enough to support the 2.3x headline. read the letter →

arxiv 2506.15844 v1 pith:CNDQYCTO submitted 2025-06-18 cs.DS

Tianyu Zhao , Dongfang Zhao , Luanzheng Guo , Nathan Tallent This is my paper · ORCID
classification cs.DS
keywords hypergraphcompressionlosslessHuffmancodingbitwiseencodingentropy-guidedpartitioningadjacencyrepresentationgraphanalyticsinteger
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

This paper proposes HybHuff, a lossless compression scheme for the integer adjacency buffers that store hypergraphs, and argues that splitting the symbol stream by frequency—Huffman codes for the frequent symbols, fixed-width bitwise codes for the rest—cuts memory use sharply. The central claim is empirical: on four real-world hypergraphs, HybHuff produces compressed sizes up to 2.3× smaller than zip and up to 1.9× smaller than ZFP, with decoding overhead low enough that BFS, PageRank, and k-core run at nearly uncompressed speed. The paper also claims a theoretical guarantee: under a mild power-law assumption on symbol frequencies, an optimal split ratio always exists between the two pure encodings, so the U-shaped compression curve observed in experiments is not accidental. If correct, this gives hypergraph systems a drop-in, structure-preserving compression layer that reduces memory without changing traversal algorithms.

What carries the argument

The load-bearing object is the split ratio ρ, the fraction of the most frequent adjacency symbols assigned to Huffman coding. The coordinator builds a frequency histogram over one side of the bipartite incidence representation (whichever of vertices or hyperedges has fewer elements), ranks symbols, and evaluates candidate ρ values in O(K) using prefix sums of frequency and entropy; the encoder then writes two disjoint bitstreams—a Huffman stream and a fixed-width bitwise stream—plus per-entity metadata arrays giving each entity's degree and its number of Huffman-coded neighbors. The decoder reconstructs adjacency lists in one pass by reading that metadata and walking the two streams, with no dynamic allocation and no backtracking. The codebook is capped and the bitstream byte-aligned, which keeps decoding cost bounded by tree depth rather than by stream length.

What would settle it

Re-run the four datasets with zip at a strong compression level and with ZFP configured for lossless integer input; if the compressed sizes approach HybHuff's within a few percent, the 2.3× claim fails. Separately, computing R(ρ) for a synthetic power-law symbol stream and checking whether the minimum falls at ρ=0 or ρ=1 would test the claimed interior-optimum theorem.

Watch

Extended reading notes

Core claim

The paper's central discovery is that neither pure Huffman coding nor pure bit-packing is the right way to compress hypergraph adjacency: Huffman wastes bits and codebook space on rare symbols, while bitwise coding ignores the heavy frequency skew of common vertex or hyperedge IDs. HybHuff therefore ranks symbols by empirical frequency, sends the top ρK symbols through a Huffman code, packs the remaining (1−ρ)K with ⌈log2(M+1)⌉ bits each, and chooses ρ by an O(K) scan over prefix frequency statistics. The paper reports that across Friendster, LiveJournal, Amazon, and Slashdot, compressed size as a function of ρ is U-shaped, with the optimum typically involving only a small Huffman domain, and that at that optimum HybHuff beats zip by up to 2.3× and ZFP by up to 1.9×. It also asserts, as Theorem 1, that under a power-law tail bound R(ρ)—the per-symbol bit cost including codebook overhead—has a minimizer in the open interval (0,1), which the paper uses to justify searching for rather than assuming an endpoint.

Load-bearing premise

The empirical advantage depends on zip and ZFP being run in a fair, tuned configuration on the same lossless integer data, but the paper does not specify their compression levels, container formats, or how a floating-point-oriented compressor like ZFP was applied, so a fairer baseline could narrow the reported gap.

Editorial extensions

If this is right

  • Across the four datasets, the reported improvements translate to HybHuff output sizes around 40–55% of zip's and ZFP's at the optimal ratio, losslessly.
  • The optimal Huffman share is small—often a few percent of distinct symbols—so the codebook overhead stays negligible while still capturing most of the entropy gain.
  • Traversal workloads can run directly on the compressed representation, so compression can be an in-memory layout choice rather than an offline archival step.
  • Because the U-shape is dataset-dependent, a fixed encoder choice is suboptimal; data-driven ratio selection is required for robust performance.

Reading between the lines

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

  • A direct extension is to apply the same frequency-split idea to other sparse integer streams—ordinary graph adjacency lists, incidence matrices, or inverted indexes—since the encoder treats the stream as flat integers and never uses hypergraph-specific semantics beyond the degree metadata.
  • A stronger baseline test would compare against a graph-aware or integer-aware compressor; if HybHuff still wins there, the structure-preserving claim is more convincing, and if not, the advantage may be partly a property of the baselines chosen.
  • The paper's observed optimal ratios (roughly 1–4% of distinct symbols) suggest a cheap default policy: Huffman-code a tiny prefix of the frequency-ordered symbols and bit-pack the rest, with periodic re-estimation when stream statistics drift.
  • Checking Theorem 1 numerically on synthetic Zipfian streams would settle whether the interior minimum is a general guarantee or a property of these datasets.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes HybHuff, a lossless compression framework for integer-based hypergraph adjacency formats that splits the symbol stream between Huffman encoding and fixed-width bitwise encoding according to a ratio ρ. It claims a theoretical guarantee (Theorem 1) that an optimal interior ratio always exists, and it reports experiments on four real-world hypergraphs showing compression-rate improvements over zip and ZFP by up to 2.3×, with modest decoding overhead and negligible impact on BFS, PageRank, and k-core runtimes. The authors state that the implementation is open-source. The main evaluation evidence is a set of U-shaped curves of compressed size versus ρ, together with fitted regressions used to locate the empirical optimum. My assessment is that the empirical curves are plausible and the system design is clearly described, but the theoretical proof is not valid as written, the validation procedure is circular, and the baseline measurement protocol is under-specified in ways that bear directly on the headline claims.

Significance. If the empirical claims are confirmed under a fair and fully specified protocol, HybHuff would offer a useful structure-preserving compression layer for hypergraph analytics, and the idea of coordinating Huffman and bitwise encoding by frequency-partitioning is sensible. The paper deserves credit for releasing code, presenting concrete algorithms, and evaluating on real workloads rather than synthetic data. However, the paper's two advertised contributions—the proof that an optimal ratio always exists in (0,1) and the demonstration of consistent superiority over standard compressors—are not currently established. The U-shaped empirical curves are the strongest evidence in the paper, but they are presented as validations of a theorem whose proof is invalid, and the baselines are not specified at the level needed to support the quantitative claims. The work is potentially significant, but the manuscript needs substantial revision before the claims can be accepted.

major comments (4)
  1. [IV-C] Theorem 1 is not proven. In the proof, the authors first state that H(ρ) is strictly increasing because each added term is non-negative, then immediately say they 'flip the sign convention and treat −H(ρ) as a strictly decreasing term in R(ρ),' even though R(ρ) in Eq. (3) is defined with a positive sign for the Huffman portion. The subsequent claim that the product B(ρ) of a decreasing tail mass and an increasing bitwidth 'is unimodal and admits a minimum' is asserted without proof; a product of a decreasing function and an increasing function need not be unimodal. The limit analysis is also incomplete: the paper compares lim_{ρ→0} R(ρ) and lim_{ρ→1} R(ρ) but ignores the behavior of B(ρ) near 1 and does not show that an interior point is lower than both boundaries. Finally, the 'approximately linear' growth of M(ρ) is assumed without justification. Thus the existence of ρ* ∈ (0,1) is not established.
  2. [VI-C] The empirical validation in Section VI-C is circular. The optimal ratio is described in Section III-A as 'discovered via an empirical memory cost evaluation,' and Section IV-D uses a grid search over ρ to find the best candidate. Section VI-C then validates Theorem 1 by fitting a regression model with log and polynomial terms to the observed compressed sizes and locating the minimum of the fitted curve. Since the same observed data are used twice—first to select ρ and then to fit the curve whose minimum is reported—the agreement between the fitted optimum and the empirical optimum is not an independent test of the theory. Moreover, a model of the form y = a + bx + cx² + d log x can produce an interior minimum by construction for a wide range of data, so the reported R² values and fitted coefficients do not confirm the specific mechanism in Eq. (3).
  3. [VI] The baseline comparison is not sufficiently specified to support the abstract's claim of up to 2.3× better compression than zip and ZFP. The text only states that HybHuff is compared with 'general-purpose compressors zip and ZFP.' It does not report the zip compression level, the container or archive format, whether the compressed sizes include the Huffman tree, degree metadata, bit-count metadata, and any uncompressed side of the bipartition, or how 'Original Size' is defined for each dataset. ZFP is a lossy floating-point compressor, and Section I-A argues that such compressors are poorly suited to sparse integer domains; the paper does not say how ZFP was configured for lossless integer adjacency data. Without this information, the reported compression ratios are not reproducible and may not represent fair, tuned baselines. The authors should provide a complete measurement protocol and, ideally, include integer-aware or graph-aware compressors as additional references.
  4. [IV-B / IV-D] There is a disconnect between the theoretical cost function R(ρ) in Eq. (2) and the actual compressed size measured in Figures 2 and 3. The theory models only the expected symbol-code length plus a codebook term α′ρ, normalized by N, whereas the empirical 'Total size (KB)' includes bitstreams, serialized Huffman trees, degree arrays, bit-count metadata, and possibly an uncompressed bipartition side. The paper never derives a formula connecting Eq. (2) to the measured quantities, so the U-shape of the empirical curves does not follow from the theorem as stated. The authors should either extend the model to include all metadata or explicitly state which terms are omitted and why the comparison is still meaningful.
minor comments (4)
  1. [VI-A] There is a typo in the first sentence of Section VI-A: 'hufffman' should be 'Huffman.'
  2. [Figures 1–3] The units of the horizontal axis are inconsistent across figures. Figure 1 shows 'Ratio (%)' with values below 2.5, Figure 2 shows 'Ratio (%)' from 0 to 100, and Figure 3 zooms to 0–10. Since ρ is defined in [0,1] in the text, the figures should state clearly whether the axis is ρ, 100ρ, or the percentage of symbols assigned to Huffman coding. This ambiguity makes it difficult to reconcile the text's statement that Friendster's optimum is '10–13%' with Figure 3's fitted minimum at 1.61%.
  3. [VI-A] The text reports 'up to 2.3× smaller compressed size than zip, and up to 1.9× smaller than ZFP,' but Figure 1 only shows compression-rate percentages, and the conversion between percentages and size ratios is not stated. The authors should define the exact quantity reported and show the corresponding ratios in the figure or tables.
  4. [IV-D] The statement that 'the optimal ratio ρ* tends to lie in a narrow band, typically [0.01, 0.05]' is presented without supporting data or a citation. If this is an empirical observation from the four datasets, it should be stated as such and connected to the results in Figures 2 and 3.

Circularity Check

1 steps flagged · score 5.0 of 10

The optimality 'validation' in Section VI-C fits a curve to the measured sizes and then reads off the minimum; this is a fitted confirmation of Theorem 1, while the compression-rate measurements themselves remain independent empirical comparisons.

  1. fitted input called prediction [Section VI-C, 'Empirical Validation of Optimality via Regression']
    "To validate the theoretical insights derived in Section IV, we zoom in on the leftmost 10% of the Huffman ratio space and perform curve fitting on the observed compressed sizes as shown in Figure 3. According to Theorem 1 and our lightweight estimation method based on prefix frequency statistics, the total compressed size is expected to follow a non-linear pattern governed by the interaction between entropy reduction and prefix overhead."

    The validation fits a polynomial-log curve to the same measured compressed sizes that were already used to select each dataset's optimal Huffman ratio. The fitted coefficients are free parameters, not derived from R(ρ) in Section IV, so the minimum of the fitted curve is not an independent test of Theorem 1; it is an interpolation of the observed U-shape. The paper even reports dataset-specific fitted equations rather than a predicted curve from the theoretical model. Thus the confirmation of the theoretical optimality claim reduces by construction to a regression on the data it is supposed to validate.

full rationale

The central compression-rate comparison against zip and ZFP is an independent measurement: the paper actually encodes hypergraph datasets and reports compressed sizes, so that part is not circular. The circularity is confined to the validation of the optimal-ratio claim: Section VI-C uses a regression fitted post hoc to the observed sizes to claim empirical confirmation of Theorem 1, but the fitted curve's minimum is determined by the data, not predicted by the theory. No load-bearing self-citation appears; the paper cites standard compression and hypergraph literature, not its own prior work. The underspecified baseline configuration (zip level, ZFP mode, what counts as Original Size) is a measurement and reproducibility risk, not a circularity. Overall, the paper has one fitted-input-dressed-as-confirmation step, warranting a moderate circularity score below the range where the main empirical result itself reduces to its inputs.

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

The only tunable parameter of the deployed encoder is rho, and it is set by brute-force searching the compression curve on each dataset, so the algorithm's performance depends on a fitted quantity. The theoretical cost model R(rho) adds three structural assumptions: a power-law tail, linear growth of the bitwise maximum value, and an unspecified codebook constant alpha; none is verified against the benchmark data. The model is used to assert, not to predict, the existence of an interior optimum, so the ledger for the central empirical claim is mostly composed of measurement choices rather than new entities. No invented entities are introduced.

free parameters (3)
  • rho (Huffman-to-bitwise split ratio) = optimal values reported as 1.25-3.75% in Figures 1-3; text in Sections VI-A and VI-B also claims 10-13%
    Selected per dataset by scanning the compression curve and keeping the best size. Section IV-A states the optimal rho is discovered via an empirical memory cost evaluation. The headline 2.3x number uses each dataset's best rho.
  • Regression coefficients in the validation fit = Friendster: y = 83980.9 + 589.4x + 22.2x^2 - 1069.6 log x, with analogous fits for LiveJournal, Amazon, and Slashdot
    Section VI-C fits these coefficients to observed compressed sizes and then takes the minimum of the fitted curve as the validated optimum. They are free parameters chosen to match the data.
  • alpha (codebook overhead constant) = unspecified, described as some small constant
    Introduced in Section IV-B-d as alpha-prime times rho in the cost model R(rho). It is never measured, and it is the only term coupling the theory to the actual implementation, which stores a serialized pre-order tree.
assumptions (4)
  • domain assumption Symbol probabilities follow a Zipfian power law p_i = 1/(i^z H_K(z)) with z > 1.
    Section IV-B-a states this as an assumed frequency model reflecting observed heavy tails, but it is never fitted or verified on the four datasets; Theorem 1's hypothesis p_i <= C i^-z inherits this assumption.
  • ad hoc to paper The maximum symbol value in the bitwise tail M(rho) grows approximately linearly with the tail index.
    Section IV-C assumes p_i <= C i^-z and that M(rho) grows approximately linearly in i for large i. No argument links frequency rank to symbol value, and the bitwise model uses M(rho) to set the bit width, so the unimodality claim for B(rho) depends on this.
  • domain assumption The side of the vertex/hyperedge bipartition with fewer elements carries the compressible redundancy.
    Algorithm 1 lines 2-8 and Section III-A choose the side with fewer elements based on an empirical observation that it tends to exhibit higher redundancy. No measurement is reported to support this.
  • standard math Huffman codeword length is approximated by the Shannon entropy term log2(1/p_i).
    Section IV-B-b and Theorem 1 use this standard within-one-bit approximation as an equality in R(rho), and ignore the ceil() discretization when combining with the bitwise term.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HybHuff: Lossless Compression for Hypergraphs via Entropy-Guided Huffman-Bitwise Coordination." pith.science (2026). https://pith.science/paper/CNDQYCTO

@misc{pith2026250615844,
  author       = {Pith},
  title        = {Pith review of: HybHuff: Lossless Compression for Hypergraphs via Entropy-Guided Huffman-Bitwise Coordination},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CNDQYCTO}},
  note         = {Machine review of arXiv:2506.15844}
}
read the original abstract

Hypergraphs provide a natural representation for many-to-many relationships in data-intensive applications, yet their scalability is often hindered by high memory consumption. While prior work has improved computational efficiency, reducing the space overhead of hypergraph representations remains a major challenge. This paper presents a hybrid compression framework for integer-based hypergraph adjacency formats, which adaptively combines Huffman encoding and bitwise encoding to exploit structural redundancy. We provide a theoretical analysis showing that an optimal encoding ratio exists between the two schemes, and introduce an empirical strategy to approximate this ratio for practical use. Experiments on real-world hypergraphs demonstrate that our method consistently outperforms standard compressors such as Zip and ZFP in compression rate by up to 2.3x with comparable decoding overhead. To assess practical utility, we integrate our framework with three common hypergraph workloads: breadth-first search, PageRank, and k-core label propagation, and show that compression incurs negligible performance loss. Extensive evaluations across four benchmark datasets confirm the efficiency and applicability of our approach.

Figures

Figures reproduced from arXiv: 2506.15844 by the authors.

Figure 1
Figure 1. Huffman encoding ratio vs. Compression rate; (Huffman encoding ratio, Compression rate); The red dots on the left indicate the initial state; the right ones are the optimal. layout aligns well with the structural sparsity of real-world hypergraphs, particularly in low-entropy regimes where offset￾based adjacency exhibits significant symbol reuse. As the Huffman ratio increases, we observe a non￾monotonic trend: comp… view at source ↗
Figure 2
Figure 2. Huffman encoding ratio vs. size of compressed hypergraph; (Encoding ratio, Size of compressed hypergraph). 0 2 4 6 8 10 Ratio (%) 84000 85000 86000 87000 88000 89000 90000 91000 Total size (KB) (1.61%, 84478 KB) Friendster HybHuff Fitted Curve 0 2 4 6 8 10 Ratio (%) 19000 19500 20000 20500 21000 21500 22000 Total size (KB) (1.27%, 18763 KB) LiveJournal HybHuff Fitted Curve 0 2 4 6 8 10 Ratio (%) 4400 4600 4800 5000 … view at source ↗
Figure 3
Figure 3. Huffman encoding ratio vs. size of compressed hypergraph (Ratio ≤ 10%). The red dots here represent the optimal on the regression curves. (Huffman Encoding Ratio, Size of compressed hypergraph). polynomial terms capture the rising overhead as Huffman coverage expands. Empirically, we observe that this regression model provides an excellent fit to the actual compression behavior across all datasets. For example, the … view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Huffman encoding ratio vs. execution time of encoding, decoding, and both. BFS K-core LP PageRank 0 20 40 60 80 100 120 Total Time (s) Friendster HybHuff Zip Zfp BFS K-core LP PageRank 0 50 100 150 200 Total Time (s) LiveJournal HybHuff Zip Zfp BFS K-core LP PageRank 0…
Figure 5
Figure 5. Figure 5: Comparison of time with compression methods across hypergraph applications, including both the decompression time and the application execution time. all tasks. In Amazon, all methods perform similarly due to the dataset’s small size. The only exception occurs in the B…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 63 canonical work pages

  1. [1]

    Abeywickrama, V

    T. Abeywickrama, V . Liang, and K.-L. Tan. Optimizing bipartite matching in real-world applications by incremental cost computation. Proc. VLDB Endow., 14(7):1150–1158, Mar. 2021

    work page 2021

  2. [2]

    Abeywickrama, V

    T. Abeywickrama, V . Liang, and K.-L. Tan. Bipartite matching: What to do in the real world when computing assignment costs dominates finding the optimal assignment. SIGMOD Rec., 51(1):51–58, June 2022

    work page 2022

  3. [3]

    Amburg, N

    I. Amburg, N. Veldt, and A. Benson. Clustering in graphs and hypergraphs with categorical edge labels. In Proceedings of The Web Conference 2020 , WWW ’20, page 706–717, New York, NY , USA,

    work page 2020

  4. [4]

    Azad and A

    A. Azad and A. Buluc ¸. Distributed-memory algorithms for maximum cardinality matching in bipartite graphs. In 2016 IEEE International Parallel and Distributed Processing Symposium (IPDPS) , pages 32–42, 2016

    work page 2016

  5. [5]

    Azami, A

    N. Azami, A. Fallin, and M. Burtscher. Efficient lossless compression of scientific floating-point data on cpus and gpus. ASPLOS ’25, page 395–409, New York, NY , USA, 2025. Association for Computing Machinery

    work page 2025

  6. [6]

    Balakrishnan and R

    R. Balakrishnan and R. K. Sahoo. Lossless compression for large scale cluster logs. IPDPS’06, page 362, USA, 2006. IEEE Computer Society

    work page 2006

  7. [7]

    Boldi and S

    P. Boldi and S. Vigna. The webgraph framework i: Compression techniques. Proceedings of the 13th International World Wide Web Conference (WWW), pages 595–602, 2004

    work page 2004

  8. [8]

    Burtscher and P

    M. Burtscher and P. Ratanaworabhan. Fpc: A high-speed compressor for double-precision floating-point data. IEEE Trans. Comput., 58(1):18–31, Jan. 2009

    work page 2009

Show all 64 references
  1. [9]

    C ¸ ataly¨urek, K

    U. C ¸ ataly¨urek, K. Devine, M. Faraj, L. Gottesb ¨uren, T. Heuer, H. Mey- erhenke, P. Sanders, S. Schlag, C. Schulz, D. Seemaier, and D. Wagner. More recent advances in (hyper)graph partitioning. ACM Comput. Surv., 55(12), Mar. 2023

    2023

  2. [10]

    Chakrabarti, Y

    D. Chakrabarti, Y . Zhan, and C. Faloutsos. R-mat: A recursive model for graph mining. In Proceedings of the 4th SIAM International Conference on Data Mining (SDM) , pages 442–446, 2004

    2004

  3. [11]

    X. Chen, J. Tian, I. Beaver, C. Freeman, Y . Yan, J. Wang, and D. Tao. Fcbench: Cross-domain benchmarking of lossless compression for floating-point data. Proc. VLDB Endow. , 17(6):1418–1431, Feb. 2024

    2024

  4. [12]

    Y . Chen, X. Wang, and C. Chen. Hyperedge importance estimation via identity-aware hypergraph attention network. In Proceedings of the 33rd ACM International Conference on Information and Knowledge Management, CIKM ’24, page 334–343, New York, NY , USA, 2024. Association for C...

    2024

  5. [13]

    Z. Chen, J. Gehrke, and F. Korn. Query optimization in compressed database systems. SIGMOD Rec., 30(2):271–282, May 2001

    2001

  6. [14]

    Z. Chen, F. Zhang, J. Guan, J. Zhai, X. Shen, H. Zhang, W. Shu, and X. Du. Compressgraph: Efficient parallel graph analytics with rule-based compression. Proc. ACM Manag. Data , 1(1), May 2023

    2023

  7. [15]

    Chiosa, F

    M. Chiosa, F. Maschi, I. M ¨uller, G. Alonso, and N. May. Hardware acceleration of compression and encryption in sap hana. Proc. VLDB Endow., 15(12):3277–3291, Aug. 2022

    2022

  8. [16]

    Y . Choi, K. Park, and H. Kim. Bice: Exploring compact search space by using bipartite matching and cell-wide verification. Proc. VLDB Endow., 16(9):2186–2198, May 2023

    2023

  9. [17]

    G. Cormode. Compact summaries over large datasets. In Proceedings of the 34th ACM SIGMOD-SIGACT-SIGAI Symposium on Principles of Database Systems , PODS ’15, page 157–158, New York, NY , USA,

  10. [18]

    Engelson, D

    V . Engelson, D. Fritzson, and P. Fritzson. Lossless compression of high-volume numerical data from simulations. In Proceedings of the Conference on Data Compression , DCC ’00, page 574, USA, 2000. IEEE Computer Society

    2000

  11. [19]

    M. H. Faykus, J. Calhoun, and M. Smith. Lossy and lossless compres- sion for biofilm optical coherence tomography (oct). In Proceedings of the SC ’23 Workshops of the International Conference on High Performance Computing, Network, Storage, and Analysis , SC-W ’23, page 281–28...

    2023

  12. [20]

    Q. Gong, J. Chen, B. Whitney, X. Liang, V . Reshniak, T. Banerjee, J. Lee, A. Rangarajan, L. Wan, N. Vidal, Q. Liu, A. Gainaru, N. Pod- horszki, R. Archibald, S. Ranka, and S. Klasky. Mgard: A multigrid framework for high-performance, error-controlled data compression and refa...

    2023

  13. [21]

    Guzun and G

    G. Guzun and G. Canahuate. Hybrid query optimization for hard-to- compress bit-vectors. The VLDB Journal , 25(3):339–354, June 2016

    2016

  14. [22]

    W.-S. Han, S. Lee, K. Park, J.-H. Lee, M.-S. Kim, J. Kim, and H. Yu. Turbograph: A fast parallel graph engine handling billion-scale graphs in a single pc. Proceedings of the 19th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD) , pages 77–85, 2013

    2013

  15. [23]

    Ilkhechi, A

    A. Ilkhechi, A. Crotty, A. Galakatos, Y . Mao, G. Fan, X. Shi, and U. Cetintemel. Deepsqueeze: Deep semantic compression for tabular data. In Proceedings of the 2020 ACM SIGMOD International Con- ference on Management of Data , SIGMOD ’20, page 1733–1746, New York, NY , USA, 2...

    2020

  16. [24]

    Islam, N

    M. Islam, N. Nurain, M. Kaykobad, S. Chellappan, and A. B. M. A. A. Islam. Helios: huffman coding based lightweight encryption scheme for data transmission. In Proceedings of the 16th EAI International Conference on Mobile and Ubiquitous Systems: Computing, Networking and Serv...

    2020

  17. [25]

    S. K. Jensen, T. B. Pedersen, and C. Thomsen. Modelardb: modular model-based time series management with spark and cassandra. Proc. VLDB Endow., 11(11):1688–1701, July 2018

    2018

  18. [26]

    Jiang, C

    H. Jiang, C. Liu, Q. Jin, J. Paparrizos, and A. J. Elmore. Pids: attribute decomposition for improved compression and query performance in columnar storage. Proc. VLDB Endow., 13(6):925–938, Feb. 2020

    2020

  19. [27]

    A. Kipf, D. Chromejko, A. Hall, P. Boncz, and D. G. Andersen. Cuckoo index: a lightweight secondary index structure. Proc. VLDB Endow. , 13(13):3559–3572, Sept. 2020

    2020

  20. [28]

    Knorr, P

    F. Knorr, P. Thoman, and T. Fahringer. ndzip-gpu: efficient lossless compression of scientific floating-point data on gpus. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis , SC ’21, New York, NY , USA, 2021. Assoc...

    2021

  21. [29]

    Kuschewski, D

    M. Kuschewski, D. Sauerwein, A. Alhomssi, and V . Leis. Btrblocks: Efficient columnar compression for data lakes. Proc. ACM Manag. Data, 1(2), June 2023

    2023

  22. [30]

    S. Lal, J. Lucas, and B. Juurlink. E ˆ2mc: Entropy encoding based memory compression for gpus. In 2017 IEEE International Parallel and Distributed Processing Symposium (IPDPS) , pages 1119–1128, 2017

    2017

  23. [31]

    S. Lal, M. Renz, J. Hartmer, and B. Juurlink. Memory access granularity aware lossless compression for gpus. In 2022 IEEE International Parallel and Distributed Processing Symposium (IPDPS) , pages 1074– 1084, 2022

    2022

  24. [32]

    G. Lee, J. Ko, and K. Shin. Hypergraph motifs: concepts, algorithms, and discoveries. Proc. VLDB Endow., 13(12):2256–2269, July 2020

    2020

  25. [33]

    Leskovec, K

    J. Leskovec, K. J. Lang, A. Dasgupta, and M. W. Mahoney. Community structure in large networks: Natural cluster sizes and the absence of large well-defined clusters. Internet Mathematics, 6(1):29–123, 2009

    2009

  26. [34]

    R. Li, Z. Li, Y . Wu, C. Chen, and Y . Zheng. Elf: Erasing-based lossless floating-point compression. Proc. VLDB Endow., 16(7):1763–1776, Mar. 2023

    2023

  27. [35]

    Y . Li, A. Kashyap, W. Chen, Y . Guo, and X. Lu. Accelerating lossy and lossless compression on emerging bluefield dpu architectures. In 2024 IEEE International Parallel and Distributed Processing Symposium (IPDPS), pages 373–385, 2024

    2024

  28. [36]

    Liakos, K

    P. Liakos, K. Papakonstantinopoulou, and Y . Kotidis. Chimp: efficient lossless floating point compression for time series databases. Proc. VLDB Endow., 15(11):3058–3070, July 2022

    2022

  29. [37]

    C. Lin, B. Mandel, Y . Papakonstantinou, and M. Springer. Fast in-memory sql analytics on typed graphs. Proc. VLDB Endow. , 10(3):265–276, Nov. 2016

    2016

  30. [38]

    Lindstrom

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

    2014

  31. [39]

    C. Liu, H. Jiang, J. Paparrizos, and A. J. Elmore. Decomposed bounded floats for fast compression and queries. Proc. VLDB Endow. , 14(11):2586–2598, July 2021

    2021

  32. [40]

    C. Liu, A. Pavlenko, M. Interlandi, and B. Haynes. A deep dive into common open formats for analytical dbmss. Proc. VLDB Endow., 16(11):3044–3056, July 2023

    2023

  33. [41]

    Maleki, U

    S. Maleki, U. Agarwal, M. Burtscher, and K. Pingali. Bipart: a parallel and deterministic hypergraph partitioner. PPoPP ’21, page 161–174, New York, NY , USA, 2021. Association for Computing Machinery

    2021

  34. [42]

    Milidiu, E

    R. Milidiu, E. Laber, and A. Pessoa. A work-efficient parallel algo- rithm for constructing huffman codes. In Proceedings DCC’99 Data Compression Conference (Cat. No. PR00096) , pages 277–286, 1999

    1999

  35. [43]

    A. Ozsoy. Culzss-bit: a bit-vector algorithm for lossless data compres- sion on gpgpus. In Proceedings of the 2014 International Workshop on Data Intensive Scalable Computing Systems , DISCS ’14, page 57–64. IEEE Press, 2014

    2014

  36. [44]

    J. Park, R. Azizi, G. F. Oliveira, M. Sadrosadati, R. Nadig, D. Novo, J. G ´omez-Luna, M. Kim, and O. Mutlu. Flash-cosmos: In-flash bulk bitwise operations using inherent computation capability of nand flash memory. In Proceedings of the 55th Annual IEEE/ACM International Symp...

    2023

  37. [45]

    R. A. Patel, Y . Zhang, J. Mak, A. Davidson, and J. D. Owens. Parallel lossless data compression on the gpu. In 2012 Innovative Parallel Computing (InPar), pages 1–9, 2012

    2012

  38. [46]

    Y . Qiao, Y . Gao, and H. Zhang. Blitzcrank: Fast semantic compression for in-memory online transaction processing. Proc. VLDB Endow. , 17(10):2528–2540, June 2024

    2024

  39. [47]

    M. Shah, X. Yu, S. Di, M. Becchi, and F. Cappello. Lightweight huffman coding for efficient gpu compression. In Proceedings of the 37th ACM International Conference on Supercomputing , ICS ’23, page 99–110, New York, NY , USA, 2023. Association for Computing Machinery

    2023

  40. [48]

    J. Shun. Practical parallel hypergraph algorithms. In Proceedings of the 25th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, PPoPP ’20, page 232–249, New York, NY , USA, 2020. Association for Computing Machinery

    2020

  41. [49]

    J. Shun, L. Dhulipala, and G. E. Blelloch. Smaller and faster: Parallel processing of compressed graphs with ligra+. In 2015 Data Compression Conference, pages 403–412, 2015

    2015

  42. [50]

    J. Tian, S. Di, K. Zhao, C. Rivera, M. H. Fulp, R. Underwood, S. Jin, X. Liang, J. Calhoun, D. Tao, and F. Cappello. cusz: An efficient gpu-based error-bounded lossy compression framework for scientific data. In Proceedings of the ACM International Conference on Parallel Archi...

    2020

  43. [51]

    J. Tian, C. Rivera, S. Di, J. Chen, X. Liang, D. Tao, and F. Cappello. Revisiting huffman coding: Toward extreme performance on modern gpu architectures. In 2021 IEEE International Parallel and Distributed Processing Symposium (IPDPS) , pages 881–891, 2021

    2021

  44. [52]

    Wang and M

    J. Wang and M. Athanassoulis. Cubit: Concurrent updatable bitmap indexing. Proc. VLDB Endow., 18(2):399–412, Oct. 2024

    2024

  45. [53]

    Z. Wang, Q. Wang, P. Wang, T. Palpanas, and W. Wang. Dumpy: A compact and adaptive index for large data series collections. Proc. ACM Manag. Data, 1(1), May 2023

    2023

  46. [54]

    Weißenberger and B

    A. Weißenberger and B. Schmidt. Massively parallel huffman decoding on gpus. In Proceedings of the 47th International Conference on Parallel Processing, ICPP ’18, New York, NY , USA, 2018. Association for Computing Machinery

    2018

  47. [55]

    Wu, H.-W

    C.-H. Wu, H.-W. Zhang, C.-W. Liu, T.-C. Yu, and C.-Y . Yang. A dynamic huffman coding method for reliable tlc nand flash memory. ACM Trans. Des. Autom. Electron. Syst. , 26(5), June 2021

    2021

  48. [56]

    Q. Xu, J. Yang, F. Zhang, J. Pan, K. Chen, Y . Shen, A. C. Zhou, and X. Du. Tribase: A vector data query engine for reliable and lossless pruning compression using triangle inequalities. Proc. ACM Manag. Data, 3(1), Feb. 2025

    2025

  49. [57]

    Yang and J

    J. Yang and J. Leskovec. Defining and evaluating network communities based on ground-truth. In Proceedings of the 2012 IEEE 12th Inter- national Conference on Data Mining (ICDM) , pages 745–754. IEEE, 2012

    2012

  50. [58]

    H. Yin, K. Wang, W. Zhang, Y . Zhang, R. Wu, and X. Lin. Efficient computation of hyper-triangles on hypergraphs. Proc. VLDB Endow. , 18(3):729–742, Nov. 2024

    2024

  51. [59]

    N. Yin, F. Feng, Z. Luo, X. Zhang, W. Wang, X. Luo, C. Chen, and X.-S. Hua. Dynamic hypergraph convolutional network. In 2022 IEEE 38th International Conference on Data Engineering (ICDE) , pages 1621– 1634, 2022

    2022

  52. [60]

    X. Yu, Y . Peng, F. Li, S. Wang, X. Shen, H. Mai, and Y . Xie. Two-level data compression using machine learning in time series database. In 2020 IEEE 36th International Conference on Data Engineering (ICDE) , pages 1333–1344, 2020

    2020

  53. [61]

    Zhang, J

    B. Zhang, J. Tian, S. Di, X. Yu, M. Swany, D. Tao, and F. Cappello. Gpulz: Optimizing lzss lossless compression for multi-byte data on modern gpus. In Proceedings of the 37th ACM International Conference on Supercomputing, ICS ’23, page 348–359, New York, NY , USA, 2023. Assoc...

    2023

  54. [62]

    Y . Zhao, X. Luo, W. Ju, C. Chen, X.-S. Hua, and M. Zhang. Dynamic hy- pergraph structure learning for traffic flow forecasting. In International Conference on Data Engineering (ICDE) 2023 , pages 2303–2316, 04 2023

    2023

  55. [63]

    L. Zhou, K. S. Candan, and J. Zou. Deepmapping: Learned data mapping for lossless compression and efficient lookup. In 2024 IEEE 40th International Conference on Data Engineering (ICDE) , pages 1– 14, 2024

    2024

  56. [2015]

    Association for Computing Machinery

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.