Pith. sign in

REVIEW 4 major objections 5 minor 29 references

Hidden Sketch: A Space-Efficient Reversible Sketch for Tracking Frequent Items in Data Streams

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Hidden Sketch claims to record heavy-hitter keys and exact frequencies in memory close to a Bloom filter's, and reports near-perfect F1 scores at 100 KB.

desk verdict A clever hybrid sketch with strong experiments, but the central exact-frequency decoding guarantee is false as stated; the paper needs a major theoretical revision. read the letter →

arxiv 2505.12293 v1 pith:CX6T5SSG submitted 2025-05-18 cs.DB

Zicang Xu , Yuxuan Tian , Yuhan Wu , Tong Yang This is my paper · ORCID
classification cs.DB
keywords datastreamsheavyhittersreversiblesketchCount-MinBloomfilterfrequencyestimationkeyrecoveryspaceefficiency
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

Hidden Sketch is a reversible data structure that claims to record both the identity and exact frequency of frequent stream items in roughly Bloom-filter-sized memory. It stores keys implicitly in a hierarchical Reversible Bloom Filter and stores frequencies in a Count-Min sketch treated as a linear system, then decodes keys by recursive candidate-set concatenation and frequencies by bucket peeling followed by SVD or integer linear programming. The paper proves a space bound of about $1.23 n w + 1.44 n l$ bits for $n$ items, and reports F1 scores near 1 on three network-traffic datasets even at 100 KB. If correct, this removes the usual trade-off between reversibility and memory in heavy-hitter tracking.

What carries the argument

The central object is the pair consisting of the Reversible Bloom Filter and the Count-Min linear system. The RBF splits each key into $l/k$-bit segments arranged in a tree, stores leaf segments in per-block bitmaps and internal segments in ordinary Bloom Filters, so that bit positions themselves encode the key. The CM sketch is read as $\Phi \mathbf{x} = \mathbf{y}$, where $\Phi$ is the bucket-item incidence matrix; pure-bucket peeling solves the triangular part, and SVD or ILP resolves the residual, with prime-weighted increments making integer solutions sparser. This combination is what lets the structure be inverted without storing keys explicitly.

What would settle it

Take a synthetic stream with a known set of $n$ keys whose hashes make the CM matrix rank-deficient after pure-bucket peeling, so the residual system has multiple nonnegative integer solutions; run the ILP decoder with no objective and check whether it returns the true frequency vector. If it returns any other feasible vector, the claimed guaranteed reversibility fails. A simpler observable version: measure exact-decoding success on Zipf streams as $m/n$ falls below the stated constants (e.g., 1.222 for three hash functions), where the paper's own theory predicts failure with probability $O(n^{-k+2})$.

Watch

Extended reading notes

Core claim

The paper claims that key and frequency encoding can be separated: a Reversible Bloom Filter with 1-bit buckets encodes keys through bucket indices, and the CM Sketch encodes frequencies as a linear map $\Phi \mathbf{x} = \mathbf{y}$. The RBF's hierarchical decomposition of the key into segments lets the decoder recover candidate keys by taking Cartesian products of segment candidate sets and filtering through per-node Bloom Filters, avoiding full key-space traversal. Frequencies are then recovered exactly by peeling pure buckets and solving the residual system with SVD or ILP; false-positive keys are discarded because their decoded frequency is zero. The paper asserts this yields guaranteed reversibility and exact frequency recovery with high probability, and that the total memory for $n$ items is less than $1.23 n w + 1.44 n l$ bits.

Load-bearing premise

The claim that frequencies are recovered exactly rests on the unproven assertion that the integer-linear-programming step returns the true nonnegative integer frequencies even when the CM matrix is not full rank, and that the SVD step's least-norm solution can be corrected to the true counts.

Editorial extensions

If this is right

  • If the space bound holds, a stream processor can track heavy hitters with key-level fidelity using memory comparable to a Bloom filter, enabling on-chip deployment in routers and switches.
  • Exact frequency recovery for reported keys means heavy-changer detection can compare true counts between windows instead of noisy estimates.
  • The RBF's hierarchical decoding returns a candidate key set equivalent in result to full key-space traversal but with less work, so false-positive keys are cheaply filtered by their zero decoded frequency.
  • The two-stage cold filter concentrates memory on significant items: infrequent items receive estimates, while frequent items receive exact keys and counts.

Reading between the lines

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

  • The same separation between a bit-level key filter and a linear frequency sketch could be ported to other linear summaries, such as Count sketch or moment sketches, to make them reversible; the paper tests only the Count-Min combination.
  • The prime-weighted increment trick suggests a general recipe: use coprime per-item weights so that null-space solutions become non-integer or sparse, but whether this preserves update speed at line rate with many items is untested.
  • The memory bound of $1.44 n l$ bits for the RBF assumes segment distributions are independent and that data skew does not blow up the Cartesian product of candidate sets; on adversarial skew the candidate sets could grow large, making the practical bound data-dependent.
  • A stress test not reported in the paper is to vary the bucket-to-key ratio $m/n$ on a Zipf stream and record exact-decoding success; the paper's Figure 6 hints the ILP step can succeed below $m=n$, but no closed-form bound for that step is proven.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes Hidden Sketch, a reversible sketch that combines a Reversible Bloom Filter (RBF) for implicit key encoding with a Count-Min (CM) sketch for frequency encoding. Keys are recovered bottom-up through hierarchical bitmaps and segment Bloom filters; frequencies are recovered by treating the CM sketch as a linear system, solved via pure-bucket peeling followed by SVD or, as an optimization, an integer linear program (ILP) with prime-weighted increments. The paper claims that this design records both keys and frequencies exactly with memory bounded by about 1.23 n w + 1.44 n l bits, and reports experiments on CAIDA, MAWI, and IMC showing F1 scores near 1 and low ARE under tight memory budgets. The theoretical appendix derives RBF memory bounds and cites prior hypergraph peeling results for the CM portion.

Significance. If the exactness and space-efficiency claims were correct, Hidden Sketch would be a meaningful contribution: it would allow exact key and frequency recovery at Bloom-filter-level memory, improving on FlowRadar's fragility and Reversible Sketch's memory overhead. The RBF design, which hierarchically encodes key segments into bitmaps and Bloom filters, is an original idea and the paper ships open-source code. However, the central advertised guarantee of exact frequency decoding is not supported and, as shown below, is false under the paper's own ILP formulation. The memory-bound proof also contains a sign error and an unproven 'best case' bound. The experimental results are promising but do not compensate for the broken theoretical core.

major comments (4)
  1. [§3.3, Eq. (1)] The claim that Integer Linear Programming 'can yield actual frequencies even when the matrix is not full rank' is false as stated. Consider d=3 hash functions, m=5 buckets, and four candidate keys with hash sets {1,2,5}, {2,3,5}, {3,4,5}, and {4,1,5}. Let the primes be (2,3,5,7) and the true frequencies x=(100,80,60,40). The vector z=(105,-70,42,-30) lies in the null space of ΦΣ_p: at every bucket the prime-weighted contributions cancel, e.g., bucket 1 receives 2·105 + 7·(-30)=0. Therefore x'=x+z=(205,10,102,10) is another nonnegative integer solution to the same Count-Min equations. No bucket is pure (buckets 1-4 contain two keys and bucket 5 contains four), so pure-bucket peeling terminates immediately. SVD returns the least-norm real solution, which is not guaranteed to be x, and an ILP with no objective can return x' or any other feasible point. Thus the statement in §3.3 is not a guarantee, and the Abstract's 'guaranteed reversibility' is unsupported. Moreover, the assertion in §3 that false positives from the RBF can be filtered because their decoded frequencies are zero fails whenever the linear system has multiple nonnegative integer solutions.
  2. [Appendix A, Theorem A.4] The proof of Theorem A.4 contains a sign error in the intermediate memory expression. It states that an additional m = (k-l)n/ln2 bits are needed to achieve the target false positive rate. From Lemma A.1, reducing the false positive rate from 2^{-(l-k)} to O(1) requires m ≈ (l-k)n/ln2 bits, not (k-l)n/ln2. As printed, the total becomes (2k-l)n/ln2, which contradicts the theorem's nl/ln2. The final statement is correct after replacing (k-l) by (l-k), but the proof as written is invalid and must be corrected.
  3. [§3.2 and Appendix A] The claimed total memory bound of 'less than 1.23 n w + 1.44 n l bits in the best case' is not established. The only rigorous bound cited is the pure-peeling condition m > c_k n from references [23,24]. The paper states that the SVD and ILP steps lower the required number of buckets but provides no analysis, no theorem, and no bound for the resulting scheme. Consequently the 'best case' memory bound is an unsupported assertion, and the space-efficiency contribution is not proven.
  4. [§3.2, SVD step] The paper asserts that SVD provides 'mathematical guarantees for approximate or exact solutions' but does not state what guarantee applies to the decoded integer frequencies. When Φ is rank-deficient, the least-norm solution is generally a real vector, not the true integer frequency vector. The paper does not prove that the residual system after pure-bucket extraction has full column rank, nor that the least-norm real solution coincides with the true nonnegative integer solution. The empirical success rate in Figure 6, while suggestive, is not a substitute for the claimed guarantee.
minor comments (5)
  1. [Algorithm 2] Line 8 contains a typo: 'dequeque' should be 'dequeue'. Also, the algorithm's loop and variable naming (e.g., 'Bucekt' in the comment) should be cleaned up.
  2. [§3.1.1] The hash function for the bitmap block array is written as H(key)=h(key)*2^l + seg(key), which is ambiguous because seg(key) denotes a segment value, not a function selecting a bitmap. The notation should distinguish the segment index from the segment value.
  3. [Appendix A, Theorems A.2 and A.3] The proofs of Theorems A.2 and A.3 are approximate: expressions such as 'Due to the Lyapunov Central Limit Theorem' and the use of Jensen's inequality with '≈' do not constitute formal error bounds. Since Theorem A.4 depends on these approximations, the RBF memory proof should either be made rigorous or explicitly labeled as heuristic.
  4. [§5.2, Figure 6] Figure 6 reports success rates from 1000 trials but does not describe the hash functions, the prime assignment, or the frequency distribution used in the simulation. Add these experimental details and report confidence intervals or error bars.
  5. [§5.1.4] The cold filter parameters (CU sketch size and filtering threshold) are not reported in the experimental setup. Since the two-stage framework's accuracy depends on these free parameters, a sensitivity analysis or explicit parameter listing is needed.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's core derivation is self-contained and its load-bearing probability and rank bounds come from external prior work, while the main weakness is an unproven mathematical claim rather than a circular reduction.

full rationale

I walked the derivation chain for Hidden Sketch and found no step in which a prediction or first-principles result reduces to its own inputs by construction. The decoding probability bound for pure-bucket extraction is explicitly attributed to external references [23,24], not to the authors' own prior work. The memory bound for the Reversible Bloom Filter is derived from standard Bloom Filter analysis (Lemma A.1) plus the paper's own hierarchical construction; it does not fit constants to the measured F1 scores and then rename the fit as a prediction. The frequency decoding uses a linear system Phi x = y and then pure-bucket peeling, SVD, and an optional ILP step. The claim that ILP with no objective 'can yield actual frequencies even when the matrix is not full rank' is an unsupported and likely false mathematical assertion, and the sign error in Theorem A.4's memory formula is a real correctness issue, but neither is circular: the ILP claim does not define the output in terms of the input or fit a parameter to a target quantity. The self-citations in the related work (e.g., HeavyKeeper and Elastic Sketch include one of the present authors) are baseline comparisons and are not load-bearing for the Hidden Sketch derivation. Therefore, while the paper has serious correctness risks in its guaranteed-exactness claims, the circularity burden is not met, and the appropriate circularity score is 0.

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

The paper defines new data structures, the Reversible Bloom Filter and Hidden Sketch, but these are the contributions themselves and are implemented and evaluated in the paper. No unobserved external entity is postulated. The load-bearing assumptions are the hash-uniformity and full-rank/ILP decoding assumptions listed above.

free parameters (3)
  • Segment count k = 4 for 32-bit keys
    Chosen by hand; controls the trade-off between leaf bitmap size and the number of internal Bloom filters, and appears in the candidate-set bound 2^(l-k) of Theorem A.2.
  • Hidden Sketch capacity n = 1800 items in experiments
    The paper reserves enough memory for 1800 items in its experiments; the theoretical memory bound depends on this n, and the experimental comparison is therefore tied to this specific capacity.
  • Cold filter threshold = 0.01% of total frequency, about 200 for CAIDA
    User-specified threshold controlling which items are promoted to the Hidden Sketch; it directly affects recall and precision in the two-stage framework.
assumptions (4)
  • domain assumption A hash function distributes inserted keys evenly across blocks, so n_i is approximately n/B for every block.
    Used in the proof of Theorem A.2 to replace per-block sums with the global ratio; load-bearing for the 2^(l-k) candidate-set bound.
  • standard math The CM matrix Phi with constant column sums has full rank with high probability when m > c_k n, based on references [23,24].
    Borrowed decoding guarantee for the pure-bucket extraction step; the paper does not verify that the candidate-key submatrix satisfies the conditions of the cited theorem.
  • standard math The Bloom filter hierarchy has no false negatives, so every true inserted key survives the candidate-generation process.
    Relied on by Algorithm 1; this is a standard Bloom filter property and is reasonable.
  • ad hoc to paper SVD pseudo-inverse or ILP returns the true nonnegative integer frequency vector even when the matrix is rank-deficient.
    Asserted in Section 3.3 without proof; a least-norm real solution or any feasible ILP solution is not guaranteed to equal the actual integer counts.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hidden Sketch: A Space-Efficient Reversible Sketch for Tracking Frequent Items in Data Streams." pith.science (2026). https://pith.science/paper/CX6T5SSG

@misc{pith2026250512293,
  author       = {Pith},
  title        = {Pith review of: Hidden Sketch: A Space-Efficient Reversible Sketch for Tracking Frequent Items in Data Streams},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CX6T5SSG}},
  note         = {Machine review of arXiv:2505.12293}
}
read the original abstract

Modern data stream applications demand memory-efficient solutions for accurately tracking frequent items, such as heavy hitters and heavy changers, under strict resource constraints. Traditional sketches face inherent accuracy-memory trade-offs: they either lose precision to reduce memory usage or inflate memory costs to enable high recording capacity. This paper introduces Hidden Sketch, a space-efficient reversible data structure for key and frequency encoding. Our design uniquely combines a Reversible Bloom Filter (RBF) and a Count-Min (CM) Sketch for invertible key and frequency storage, enabling precise reconstruction for both keys and their frequencies with minimal memory. Theoretical analysis establishes Hidden Sketch's space complexity and guaranteed reversibility, while extensive experiments demonstrate its substantial improvements in accuracy and space efficiency in frequent item tracking tasks. By eliminating the trade-off between reversibility and space efficiency, Hidden Sketch provides a scalable foundation for real-time stream analytics in resource-constrained environments.

Figures

Figures reproduced from arXiv: 2505.12293 by the authors.

Figure 1
Figure 1. When inserting the key 192.168.133.1, we add its internal [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 1
Figure 1. An example of Reversible Bloom Filter with 32-bit key. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Two-stage framework for tracking frequent items. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figures from the paper (4 more)
Figure 3
Figure 3. Figure 3: Accuracy comparison of our algorithm with baselines on CAIDA. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png]
Figure 4
Figure 4. Figure 4: Accuracy comparison of our algorithm with baselines on MAWI. [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Accuracy comparison of our algorithm with baselines on IMC. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Comparison of success rate. , where 𝑚′ is the remaining memory of the parent Bloom Filter. Slightly modify the expression, we have ( Î𝛼 𝑖=1 ( ( 𝑡𝑖 𝑛𝑖 − 1) × 2 − 𝑚𝑖 ln 2 𝑛 + 1) − 𝑛 Î𝛼 𝑖=1 𝑛𝑖 ) × 2 − 𝑚′ ln 2 𝑛 Î𝛼 𝑖=1 𝑡𝑖 𝑛𝑖 − 𝑛 Î𝛼 𝑖=1 𝑛𝑖 . Note that the candidate set of e…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 25 canonical work pages

  1. [1]

    A survey on big data for network traffic monitoring and analysis

    Alessandro D’Alconzo, Idilio Drago, Andrea Morichetta, Marco Mellia, and Pedro Casas. A survey on big data for network traffic monitoring and analysis. IEEE Transactions on Network and Service Management , 16(3):800–813, 2019

    work page 2019

  2. [2]

    Graph stream summarization: From big bang to big crunch

    Nan Tang, Qing Chen, and Prasenjit Mitra. Graph stream summarization: From big bang to big crunch. In Proceedings of the 2016 International Conference on Management of Data, SIGMOD ’16, page 1481–1496, New York, NY, USA, 2016. Association for Computing Machinery

    work page 2016

  3. [3]

    Bryan Ball, Mark Flood, H. V. Jagadish, Joe Langsam, Louiqa Raschid, and Peratham Wiriyathammabhum. A flexible and extensible contract aggrega- tion framework (caf) for financial data stream analytics. In Proceedings of the International Workshop on Data Science for Macro-Modeling , DSMM’14, page 1–6, New York, NY, USA, 2014. Association for Computing Machinery

    work page 2014

  4. [4]

    Tamer Özsu

    Şule Gündüz and M. Tamer Özsu. A web page prediction model based on click- stream tree representation of user behavior. In Proceedings of the Ninth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining , KDD ’03, page 535–540, New York, NY, USA, 2003. Association for Computing Machin- ery

    work page 2003

  5. [5]

    An improved data stream summary: the count-min sketch and its applications

    Graham Cormode and Shan Muthukrishnan. An improved data stream summary: the count-min sketch and its applications. Journal of Algorithms , 55(1):58–75, 2005

    2005

  6. [6]

    New directions in traffic measurement and accounting

    Cristian Estan and George Varghese. New directions in traffic measurement and accounting. In Proceedings of the 2002 conference on Applications, technologies, architectures, and protocols for computer communications , pages 323–336, 2002

    work page 2002

  7. [7]

    Finding frequent items in data streams

    Moses Charikar, Kevin Chen, and Martin Farach-Colton. Finding frequent items in data streams. In Automata, Languages and Programming. Springer, 2002

    work page 2002

  8. [8]

    HeavyKeeper: An accurate algorithm for finding top-k elephant flows

    Junzhi Gong, Tong Yang, Haowei Zhang, Hao Li, Steve Uhlig, Shigang Chen, Lorna Uden, and Xiaoming Li. HeavyKeeper: An accurate algorithm for finding top-k elephant flows. In 2018 USENIX Annual Technical Conference (USENIX ATC 18), pages 909–921, Boston, MA, July 2018. USENIX Association

    work page 2018

Show all 29 references
  1. [9]

    Elastic sketch: Adaptive and fast network-wide measurements

    Tong Yang, Jie Jiang, Peng Liu, Qun Huang, Junzhi Gong, Yang Zhou, Rui Miao, Xiaoming Li, and Steve Uhlig. Elastic sketch: Adaptive and fast network-wide measurements. In Proceedings of the 2018 Conference of the ACM Special Interest Group on Data Communication , pages 561–575, 2018

    2018

  2. [10]

    Lu Tang, Qun Huang, and Patrick P. C. Lee. Mv-sketch: A fast and compact invertible sketch for heavy flow detection in network data streams. In IEEE INFOCOM 2019 - IEEE Conference on Computer Communications , pages 2026– 2034, 2019

    2019

  3. [11]

    On efficient query processing of stream counts on the cell processor

    Dina Thomas, Rajesh Bordawekar, Charu C Aggarwal, and S Yu Philip. On efficient query processing of stream counts on the cell processor. In Data En- gineering, 2009. ICDE’09. IEEE 25th International Conference on , pages 748–759. IEEE, 2009

    2009

  4. [12]

    Tight-sketch: A high-performance sketch for heavy item-oriented data stream mining with limited memory size

    Weihe Li and Paul Patras. Tight-sketch: A high-performance sketch for heavy item-oriented data stream mining with limited memory size. In Proceedings of the 32nd ACM International Conference on Information and Knowledge Management , CIKM ’23, page 1328–1337, New York, NY, USA,...

    2023

  5. [13]

    Cormode and S

    G. Cormode and S. Muthukrishnan. What’s new: finding significant differences in network data streams. IEEE/ACM Transactions on Networking, 13(6):1219–1232, 2005

    2005

  6. [14]

    Reversible sketches: enabling monitoring and analysis over high-speed data streams

    Robert Schweller, Zhichun Li, Yan Chen, et al. Reversible sketches: enabling monitoring and analysis over high-speed data streams. IEEE/ACM Transactions on Networking (ToN), 15(5):1059–1072, 2007

    2007

  7. [15]

    Tian Bu, Jin Cao, Aiyou Chen, and Patrick P.C. Lee. Sequential hashing: A flexible approach for unveiling significant patterns in high speed networks. Computer Networks, 54(18):3309–3326, 2010

    2010

  8. [16]

    Simple and efficient identification of heavy hitters based on bitcount

    Feng Wang and Lixin Gao. Simple and efficient identification of heavy hitters based on bitcount. In 2019 IEEE 20th International Conference on High Performance Switching and Routing (HPSR) , pages 1–6, 2019

    2019

  9. [17]

    Flowradar: a better netflow for data centers

    Yuliang Li, Rui Miao, Changhoon Kim, and etal. Flowradar: a better netflow for data centers. In NSDI, 2016

    2016

  10. [18]

    Qun Huang, Patrick P. C. Lee, and Yungang Bao. Sketchlearn: Relieving user burdens in approximate measurement with automated statistical inference. In Proceedings of the 2018 Conference of the ACM Special Interest Group on Data Com- munication, SIGCOMM ’18, page 576–590, New Y...

    2018

  11. [19]

    https://github.com/HiddenSketch/ HiddenSketchCode/, 2023

    Source code related to hidden sketch. https://github.com/HiddenSketch/ HiddenSketchCode/, 2023

    2023

  12. [20]

    Huang and P

    Q. Huang and P. P. C. Lee. Ld-sketch: A distributed sketching design for accurate and scalable anomaly detection in network data streams. In INFOCOM, 2014

    2014

  13. [21]

    Haas, and Chris Jermaine

    Graham Cormode, Minos Garofalakis, Peter J. Haas, and Chris Jermaine. 2011

    2011

  14. [22]

    Toward Nearly-Zero-Error sketching via compressive sensing

    Qun Huang, Siyuan Sheng, Xiang Chen, Yungang Bao, Rui Zhang, Yanwei Xu, and Gong Zhang. Toward Nearly-Zero-Error sketching via compressive sensing. In 18th USENIX Symposium on Networked Systems Design and Implementation (NSDI 21), pages 1027–1044. USENIX Association, April 2021

    2021

  15. [23]

    A. M. Odlyzko. On the ranks of some (0, 1)-matrices with constant row sums. Journal of the Australian Mathematical Society. Series A. Pure Mathematics and Statistics, 31(2):193–201, 1981

    1981

  16. [24]

    Invertible bloom lookup tables

    Michael T Goodrich and Michael Mitzenmacher. Invertible bloom lookup tables. In Proceedings of the 49th Annual Allerton Conference on Communication, Control, and Computing, pages 792–799. IEEE, 2011

    2011

  17. [25]

    https://catalog.caida.org/details/dataset/ passive_2018_pcap

    Anonymized internet traces 2018. https://catalog.caida.org/details/dataset/ passive_2018_pcap. Accessed: 2022-6-29

    2018

  18. [26]

    http://mawi.wide.ad.jp/mawi/

    MAWI Working Group Traffic Archive. http://mawi.wide.ad.jp/mawi/

  19. [27]

    Network traffic charac- teristics of data centers in the wild

    Theophilus Benson, Aditya Akella, and David A Maltz. Network traffic charac- teristics of data centers in the wild. In𝑃𝑟𝑜𝑐. IMC, pages 267–280, 2010

    2010

  20. [28]

    One sketch to rule them all: Rethinking network flow monitoring with univmon

    Zaoxing Liu, Antonis Manousis, Gregory Vorsanger, Vyas Sekar, and Vladimir Braverman. One sketch to rule them all: Rethinking network flow monitoring with univmon. In Proceedings of the 2016 ACM SIGCOMM Conference , pages 101–114, 2016

    2016

  21. [29]

    Dillinger and Panagiotis Manolios

    Peter C. Dillinger and Panagiotis Manolios. Bloom filters in probabilistic ver- ification. In Alan J. Hu and Andrew K. Martin, editors, Formal Methods in Computer-Aided Design, pages 367–381, Berlin, Heidelberg, 2004. Springer Berlin Heidelberg. A Mathematical Analysis We prov...

    2004

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.