Pith. sign in

REVIEW 4 major objections 5 minor 102 references

Performance Optimization and Comparative Analysis of Generative AI Models on Advanced Accelerators

T0 review · 4 major / 5 minor · reviewed 2026-07-12 · grok-4.5

Pith's one-line read Sensitivity-guided mixed-precision quantization compresses a 1.1B LLM 2-4x with little loss, while Gaudi cards scale fine-tuning and diffusion training near-linearly across generations.

desk verdict Useful multi-generation Gaudi vs NVIDIA measurements and a concrete mixed-precision PTQ recipe, but the sensitivity-to-quantization link is unvalidated and the PTQ scope is a single small model. read the letter →

arxiv 2607.05400 v1 pith:4FQLUVY3 submitted 2026-05-05 cs.PF cs.LG

classification cs.PFcs.LG
keywords OptimizationNvidiaGPUIntelGaudiAcceleratorsBenchmarkingQuantizationLLMsmixed-precisionPTQ
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

Deploying large language and diffusion models is hard because they need too much memory, time, and specialized hardware, and those costs change when the same model moves between GPU and HPU platforms. This paper claims that a practical fix is a sensitivity-aware mixed-precision post-training quantization method: it measures how much each transformer layer or MLP submodule hurts accuracy when pruned, clusters those sensitivities, and assigns higher precision only to the fragile parts. On TinyLlama-1.1B the approach delivers roughly 2-4x compression with only small drops on language modeling, commonsense reasoning, and reading comprehension, both in simulation and in real FP16/INT8 or BF16/FP8 runs on A100 and Gaudi2. The same systems study shows that multi-card fine-tuning of 8B-70B models and diffusion-based super-resolution training scale nearly linearly on successive Gaudi generations, with clear speed-ups over earlier cards and versus NVIDIA baselines when the model fits, while DeepSpeed-style sharding is needed only when a single card cannot hold the weights. A reader who cares about running generative models on real HPC clusters gets concrete numbers and a concrete recipe for trading a little accuracy for large memory and throughput gains.

What carries the argument

Sensitivity-aware mixed-precision PTQ: magnitude-pruning sensitivity scores S_ℓ (change in perplexity or accuracy after 30% sparsity on linear layers or MLP projections) are clustered by percentile and mapped to a user bit-width list so sensitive modules stay at higher precision and robust modules drop to lower precision.

What would settle it

Re-run the identical sensitivity clustering and mixed-precision assignment on a larger or architecturally different model (or on held-out tasks) and check whether the claimed 2-4x compression still preserves accuracy within the margins reported in Table 1; a clear accuracy collapse under the same bit map would falsify the central PTQ claim.

Watch

Extended reading notes

Core claim

A two-phase sensitivity-aware mixed-precision post-training quantization framework that ranks transformer layers or MLP submodules by magnitude-pruning impact and maps the ranked clusters to higher or lower bit-widths (or to BF16 versus FP8) achieves 2-4x compression of TinyLlama-1.1B with minimal perplexity or accuracy loss on WikiText-2, HellaSwag and BoolQ on both NVIDIA A100 and Intel Gaudi2, while multi-card fine-tuning and diffusion training on Gaudi hardware scale nearly linearly and improve substantially from one accelerator generation to the next.

Load-bearing premise

The ranking of layers or submodules obtained by 30% magnitude pruning on a modest calibration set is assumed to be a faithful proxy for how those same modules will tolerate real quantization, and that ranking is assumed to transfer beyond the single 1.1B model and three tasks tested.

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 / 5 minor

Summary. The manuscript presents an empirical performance study of generative AI models (LLMs and a diffusion super-resolution model) on heterogeneous HPC accelerators: NVIDIA GPUs (V100/A100/H100 on Expanse/Perlmutter) and Intel Gaudi1/2/3 (Voyager and limited Gaudi3 access). The central technical claim is a sensitivity-aware mixed-precision post-training quantization (PTQ) procedure for TinyLlama-1.1B: magnitude-pruning of linear layers or MLP submodules at 30% sparsity yields layer/submodule sensitivity scores S_ℓ = ΔPPL/Acc, which are percentile-clustered and mapped to bit-width schedules ([16,8,4], [8,8,4], [8,4,4] simulated; BF16/FP8 or FP16/INT8 real). Table 1 reports 2–4× compression with small WikiText-2 perplexity and HellaSwag accuracy changes, and larger BoolQ drops under aggressive fake quant, on one A100 and one Gaudi2 card. Separate sections report single-epoch fine-tuning of Llama3.1-8B, Gemma2-27B, and Llama3.3-70B on Gaudi generations under DDP/DeepSpeed/FSDP (near-linear multi-card scaling; DeepSpeed preferred only when the model does not fit; flash attention 10–20% gains) and diffusion training scaling versus NVIDIA baselines (Fig. 3). A public GitHub artifact is provided.

Significance. If the PTQ procedure is shown to be genuinely sensitivity-driven rather than an arbitrary mild mixed-precision schedule, and if the multi-accelerator scaling results hold under broader models and variance reporting, the work is a useful systems contribution for the HPC/AI community: concrete cross-vendor numbers (A100 vs Gaudi2 PTQ; Gaudi1/2/3 fine-tuning; Gaudi vs V100/A100/H100 diffusion), practical guidance on when to use DDP vs ZeRO3/FSDP, and a reproducibility repository. The fine-tuning and diffusion sections already stand as solid empirical benchmarks independent of the PTQ novelty claim. Strengths to credit: hardware-specific real quantization (Intel Neural Compressor FP8 E4M3/E5M2; A100 FP16/INT8 group-wise), multi-task evaluation, and explicit multi-generation Gaudi comparison that is still relatively scarce in the literature.

major comments (4)
  1. [Section 2 (sensitivity S_ℓ, 30% sparsity, clustering)] Section 2 defines sensitivity via magnitude pruning at a fixed 30% sparsity (S_ℓ = PPL/Acc_pruned,ℓ − PPL/Acc_baseline on WikiText-2 validation 512-token chunks and 2000 HellaSwag/BoolQ training samples), then maps percentile clusters to bit-widths or BF16/FP8. Magnitude zeroing is not the same operator as group-wise (group size 128) scale/zero-point quantization or FP8 E4M3/E5M2 conversion. The manuscript never shows that the resulting ranking correlates with true quantization error, that 30% is stable, or that alternative estimators (Hessian/activation-aware, or direct fake-quant sweeps) produce the same maps. Without that link, Table 1’s 2–4× “sensitivity-aware” claim is not isolated from any mild mixed-precision schedule.
  2. [Table 1] Table 1 reports fake and real mixed-precision results but no uniform-precision or non-sensitivity (random/heuristic) mixed-precision controls at matched average bit-width/compression. Without those baselines it is impossible to attribute the favorable WikiText-2/HellaSwag numbers to the proposed clustering rather than to keeping some layers at higher precision. This is load-bearing for the abstract’s “novel mixed-precision PTQ” claim.
  3. [Section 2 / Abstract / Table 1] The entire PTQ evaluation uses only TinyLlama-1.1B. The abstract and introduction frame a general framework for LLMs and heterogeneous platforms; a single 1.1B model on three tasks does not support that scope. At minimum the paper should either restrict claims to this model or add at least one larger model (e.g., 7–8B) under the same real-quantization paths on A100 and Gaudi2.
  4. [Table 1 (BoolQ rows); abstract claim of minimal loss] Language of “minimal accuracy loss” is inconsistent with Table 1 BoolQ: FP32 ~57.9% falls to 54.4% under [16,8,4]/[8,8,4] and to ~52–53% under [8,4,4] fake quant. Real BF16/FP8 and FP16/INT8 recover closer to baseline, but the aggressive fake schedules that drive the highest compression ratios do not. Claims and abstract wording should be qualified by task and schedule, and error bars or multi-seed runs should be reported so that multi-point drops are interpretable.
minor comments (5)
  1. [Fig. 1] Fig. 1 caption and body should state explicitly how many layers/submodules fall into each precision cluster and whether the same map is used for all three tasks or re-derived per task.
  2. [Section 3] Fine-tuning hyperparameters (learning rate, batch size per card, sequence length, optimizer, DeepSpeed ZeRO stage details, FSDP wrapping policy) are only partially described; a short table would aid reproducibility beyond the GitHub link.
  3. [Fig. 3 / Section 4] Fig. 3: report absolute wall-clock or images/sec in addition to relative curves; clarify batch size and whether communication is NCCL vs HCCL so GPU–Gaudi comparisons are fair.
  4. [Throughout] Minor typos and wording: “tensore cores”, “Hellaswag” vs HellaSwag, “1 , 536” spacing, “SSCUs”, and inconsistent model name punctuation (Llama3.1:8B vs Llama3.1-8B). Normalize throughout.
  5. [Introduction / References] Related work should more clearly position the method against prior mixed-precision PTQ (including the authors’ CPTQuant arXiv:2412.03599 cited as [7]) so readers can see what is new in the sensitivity clustering and Gaudi real-quant path.

Circularity Check

1 steps flagged · score 1.0 of 10

No significant circularity: empirical hardware benchmarks and measured PTQ accuracy; only minor non-load-bearing self-citation to authors' related CPTQuant arXiv.

  1. self citation load bearing [Section 2, paragraph introducing mixed-precision PTQ]
    "Unlike full-model quantization, mixed-precision PTQs keep sensitive components at higher precision and apply lower precision to other parts of the model [7]. In this work, we introduce a novel sensitivity-aware mixed-precision post-training quantization framework."

    Citation [7] is the authors' own prior arXiv (CPTQuant). It is not load-bearing: the present paper fully defines its own magnitude-pruning sensitivity, clustering, and bit maps, then measures new results on Gaudi/A100; the citation only supplies background phrasing and does not force Table 1 or the scaling figures.

full rationale

This is a systems/performance paper whose central claims are measured wall-clock times, throughput, compression ratios, and task metrics (WikiText-2 perplexity, HellaSwag/BoolQ accuracy) on external datasets and real accelerators (A100, Gaudi1/2/3, V100/H100). The sensitivity score S_ℓ is defined from observed PPL/Acc deltas after magnitude pruning, then used as a heuristic to assign bit-widths; the subsequent Table 1 numbers are independently evaluated after real/fake quantization, not algebraically forced by the sensitivity definition. Fine-tuning (DeepSpeed/FSDP/DDP scaling) and diffusion training results are pure empirical scaling curves. The single self-citation [7] (authors' CPTQuant) merely notes the general mixed-precision idea and is not required for any uniqueness claim, uniqueness theorem, or derivation of the reported numbers. No fitted parameter is renamed a prediction, no ansatz is smuggled, and no result reduces by construction to its inputs. Score 1 only for the minor self-citation; the work is self-contained against external benchmarks.

Assumptions & free parameters 4 free parameters · 3 assumptions · 1 invented entities

The work is empirical systems benchmarking. Load-bearing choices are experimental knobs (30% sparsity, group size 128, bit-width lists, calibration sample counts, single-epoch fine-tune, Lazy-mode diffusion) rather than physical axioms or new ontological entities. No free parameters are fitted to prove a universal law; the central claims are the measured tables themselves under these stated settings.

free parameters (4)
  • magnitude-pruning sparsity target = 30%
    Fixed at 30% for sensitivity estimation; choice directly shapes layer ranking and mixed-precision maps.
  • quantization group size = 128
    Group-wise quantization uses groups of 128 consecutive weights for scales/zero-points.
  • sensitivity bit-width cluster maps = [16,8,4] / [8,8,4] / [8,4,4]; BF16 vs FP8
    User-defined lists such as [16,8,4], [8,8,4], [8,4,4] or BF16/FP8 assignment after percentile clustering.
  • calibration/evaluation sample counts and sequence lengths = 512 tokens; 2000 samples; 1 epoch
    WikiText 512-token chunks; 2000 HellaSwag/BoolQ samples for sensitivity; fixed batch sizes; single-epoch fine-tuning on ~392k psychology records.
assumptions (3)
  • ad hoc to paper Magnitude pruning sensitivity of linear/MLP weights is a valid proxy for mixed-precision quantization robustness.
    Section 2 defines S_ℓ from pruned vs baseline PPL/Acc and maps clusters to bit-widths without independent theoretical guarantee.
  • domain assumption Standard transformer LLM and diffusion training/inference semantics under PyTorch, Optimum-Habana, Intel Neural Compressor, DeepSpeed, and FSDP.
    All experiments assume these stacks correctly implement the models and parallel strategies on Gaudi and NVIDIA.
  • domain assumption Task metrics (WikiText perplexity, HellaSwag/BoolQ accuracy, wall-clock fine-tune and diffusion training time) are adequate for comparing quantization quality and accelerator performance.
    Central tables and figures rest on these standard but incomplete proxies for deployment quality.
invented entities (1)
  • sensitivity-aware mixed-precision PTQ framework (magnitude-pruning sensitivity clustering mapped to bit-widths)
    purpose: Assign higher precision to sensitive layers/submodules and lower precision to robust ones for PTQ without retraining.
    Presented as a novel evaluation/framework in Section 2; it is a procedural pipeline rather than a new physical object, with independent_evidence false beyond this paper's tables.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Performance Optimization and Comparative Analysis of Generative AI Models on Advanced Accelerators." pith.science (2026). https://pith.science/paper/4FQLUVY3

@misc{pith2026260705400,
  author       = {Pith},
  title        = {Pith review of: Performance Optimization and Comparative Analysis of Generative AI Models on Advanced Accelerators},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4FQLUVY3}},
  note         = {Machine review of arXiv:2607.05400}
}
read the original abstract

Generative AI models, such as Large Language Models (LLMs) and diffusion models, have demonstrated impressive performance across a wide range of tasks. Despite these advances, deployment remains challenging due to substantial memory requirements, extended inference latency, significant computational demands, and high hardware costs. These issues are further complicated when evaluating models across heterogeneous platforms, where differences in numerical formats, memory bandwidths, and software stacks interact with model architecture and workload characteristics in complex ways. To address these challenges, we present a systematic study focused on performance optimization and comparative analysis of several Generative AI models across diverse downstream tasks. This work introduces a novel mixed-precision post-training quantization evaluation, examines fine-tuning strategies, and assesses performance across modern high-performance computing (HPC) systems and advanced accelerators.

Figures

Figures reproduced from arXiv: 2607.05400 by the authors.

Figure 1
Figure 1. Magnitude-pruning based sensitivity clustering of TinyLlama-1.1B linear layers for simulated quantization grouped into [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Fine-tuning runtime of the Llama3.1-8B-Instruct model across three generations of Gaudi nodes; FA refers to flash attention [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Super resolution model training time for different accelerators. [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

102 extracted references · 1 canonical work pages

  1. [1]

    2021 , eprint=

    Image Super-Resolution via Iterative Refinement , author=. 2021 , eprint=

  2. [2]

    2025 , journal=

    Full-sky Models of Galactic Microwave Emission and Polarization at Subarcminute Scales for the Python Sky Model , author=. 2025 , journal=. doi:10.3847/1538-4357/adf212 , url=

  3. [3]

    2021 , publisher =

    Andrea Zonca and Ben Thorne and Nicoletta Krachmalnicoff and Julian Borrill , title =. 2021 , publisher =. doi:10.21105/joss.03783 , url =

  4. [4]

    2022 , url =

    Janspiry , title =. 2022 , url =

  5. [5]

    , title =

    Javier H Nicolau. , title =. 2025 , url =

  6. [6]

    2021 , isbn =

    Strande, Shawn and Cai, Haisong and Tatineni, Mahidhar and Pfeiffer, Wayne and Irving, Christopher and Majumdar, Amit and Mishin, Dmitry and Sinkovits, Robert and Norman, Mike and Wolter, Nicole and Cooper, Trevor and Altintas, Ilkay and Kandes, Marty and Perez, Ismael and Shantharam, Manu and Thomas, Mary and Sivagnanam, Subhashini and Hutton, Thomas , t...

  7. [7]

    and Chen, Jiunn-Yeu and Duarte, Javier M

    Amaro, Rommie E. and Chen, Jiunn-Yeu and Duarte, Javier M. and Hutton, Thomas E. and Irving, Christopher and Kandes, Martin C. and Majumdar, Amit and Mishin, Dmitry Y. and Nguyen, Mai H. and Rodriguez, Paul and Silva, Fernando and Sinkovits, Robert S. and Strande, Shawn M. and Tatineni, Mahidhar and Tran, Leon Si and Wolter, Nicole , title =. 2023 , isbn ...

  8. [8]

    2021 , howpublished =

    Perlmutter Supercomputer , author =. 2021 , howpublished =

Show all 102 references
  1. [9]

    Proceedings of the AAAI Symposium Series , volume=

    Building communication efficient asynchronous peer-to-peer federated llms with blockchain , author=. Proceedings of the AAAI Symposium Series , volume=

  2. [10]

    arXiv preprint arXiv:2412.03599 , year=

    CPTQuant-A Novel Mixed Precision Post-Training Quantization Techniques for Large Language Models , author=. arXiv preprint arXiv:2412.03599 , year=

  3. [11]

    arXiv preprint arXiv:2402.06196 , year=

    Large language models: A survey , author=. arXiv preprint arXiv:2402.06196 , year=

  4. [12]

    IEEE transactions on knowledge and data engineering , volume=

    A survey on generative diffusion models , author=. IEEE transactions on knowledge and data engineering , volume=. 2024 , publisher=

  5. [13]

    Abril and Robert Plant

    Patricia S. Abril and Robert Plant. The patent holder's dilemma: Buy, sell, or troll?. Communications of the ACM. doi:10.1145/1188913.1188915

  6. [14]

    Deciding equivalances among conjunctive aggregate queries

    Sarah Cohen and Werner Nutt and Yehoshua Sagic. Deciding equivalances among conjunctive aggregate queries. doi:10.1145/1219092.1219093

  7. [15]

    Special issue: Digital Libraries. 1996

  8. [16]

    Understanding Policy-Based Networking

    David Kosiur. Understanding Policy-Based Networking

  9. [19]

    doi:10.1007/3-540-09237-4

    The title of book two. doi:10.1007/3-540-09237-4

  10. [20]

    Asad Z. Spector. Achieving application requirements. Distributed Systems. doi:10.1145/90417.90738

  11. [21]

    Douglass and David Harel and Mark B

    Bruce P. Douglass and David Harel and Mark B. Trakhtenbrot. Statecarts in use: structured analysis and object-orientation. Lectures on Embedded Systems. doi:10.1007/3-540-65193-4_29

  12. [22]

    Donald E. Knuth. The Art of Computer Programming, Vol. 1: Fundamental Algorithms (3rd. ed.)

  13. [23]

    Donald E. Knuth. The Art of Computer Programming

  14. [24]

    Structured Variational Inference Procedures and their Realizations (as incol)

    Dan Geiger and Christopher Meek. Structured Variational Inference Procedures and their Realizations (as incol). Proceedings of Tenth International Workshop on Artificial Intelligence and Statistics, The Barbados

  15. [25]

    Stan W. Smith. An experiment in bibliographic mark-up: Parsing metadata for XML export. Proceedings of the 3rd. annual workshop on Librarians and Computers

  16. [26]

    Catch me, if you can: Evading network signatures with web-based polymorphic worms

    Matthew Van Gundy and Davide Balzarotti and Giovanni Vigna. Catch me, if you can: Evading network signatures with web-based polymorphic worms. Proceedings of the first USENIX workshop on Offensive Technologies

  17. [27]

    Predicate Path expressions

    Sten Andler. Predicate Path expressions. Proceedings of the 6th. ACM SIGACT-SIGPLAN symposium on Principles of Programming Languages. doi:10.1145/567752.567774

  18. [28]

    LOGICS of Programs: AXIOMATICS and DESCRIPTIVE POWER

    David Harel. LOGICS of Programs: AXIOMATICS and DESCRIPTIVE POWER

  19. [29]

    Anisi , title =

    David A. Anisi , title =

  20. [30]

    Clarkson

    Kenneth L. Clarkson. Algorithms for Closest-Point Problems (Computational Geometry)

  21. [31]

    Introduction to Bayesian Statistics

    Harry Thornburg. Introduction to Bayesian Statistics. 2001

  22. [32]

    CLIFFORD: a Maple 11 Package for Clifford Algebra Computations, version 11

    Rafal Ablamowicz and Bertfried Fauser. CLIFFORD: a Maple 11 Package for Clifford Algebra Computations, version 11. 2007

  23. [33]

    Stats and Analysis

    Poker-Edge.Com. Stats and Analysis. 2006

  24. [34]

    A more perfect union

    Barack Obama. A more perfect union

  25. [35]

    The fountain of youth

    Joseph Scientist. The fountain of youth

  26. [36]

    Solder man

    Dave Novak. Solder man. ACM SIGGRAPH 2003 Video Review on Animation theater Program: Part I - Vol. 145 (July 27--27, 2003). doi:10.945/woot07-S422

  27. [37]

    Interview with Bill Kinder: January 13, 2005

    Newton Lee. Interview with Bill Kinder: January 13, 2005. Comput. Entertain. doi:10.1145/1057270.1057278

  28. [38]

    The Enabling of Digital Libraries

    Bernard Rous. The Enabling of Digital Libraries. Digital Libraries

  29. [40]

    (new) Finding minimum congestion spanning trees , journal =

    Werneck, Renato and Setubal, Jo\. (new) Finding minimum congestion spanning trees , journal =. doi:10.1145/351827.384253 , acmid = 384253, publisher =

  30. [42]

    and Mei, Alessandro , title =

    Conti, Mauro and Di Pietro, Roberto and Mancini, Luigi V. and Mei, Alessandro , title =. Inf. Fusion , volume =. 2009 , issn =. doi:10.1016/j.inffus.2009.01.002 , acmid =

  31. [43]

    and Hutchful, David K

    Li, Cheng-Lun and Buyuktur, Ayse G. and Hutchful, David K. and Sant, Natasha B. and Nainwal, Satyendra K. , title =. CHI '08 extended abstracts on Human factors in computing systems , year =. doi:10.1145/1358628.1358946 , acmid =

  32. [44]

    , title =

    Hollis, Billy S. , title =. 1999 , isbn =

  33. [45]

    Goossens, Michel and Rahtz, S. P. and Moore, Ross and Sutor, Robert S. , title =. 1999 , isbn =

  34. [46]

    and Rosenberg, Arnold L

    Buss, Jonathan F. and Rosenberg, Arnold L. and Knott, Judson D. , title =. 1987 , source =

  35. [47]

    CHI '08: CHI '08 extended abstracts on Human factors in computing systems , year =

    , note =. CHI '08: CHI '08 extended abstracts on Human factors in computing systems , year =

  36. [48]

    Algorithms for Closest-Point Problems (Computational Geometry) , year =

    Clarkson, Kenneth Lee , advisor =. Algorithms for Closest-Point Problems (Computational Geometry) , year =

  37. [49]

    SIGCOMM Comput. Commun. Rev. , year =

  38. [50]

    2004 , isbn =

    IEEE TCSC Executive Committee , booktitle =. 2004 , isbn =. doi:http://dx.doi.org/10.1109/ICWS.2004.64 , acmid =

  39. [51]

    Distributed systems (2nd Ed.) , year =

  40. [52]

    , title =

    Petrie, Charles J. , title =. 1986 , source =

  41. [53]

    Donald E. Knuth. Seminumerical Algorithms. 1981

  42. [54]

    E-commerce and cultural values , year =

    Kong, Wei-Chang , Title =. E-commerce and cultural values , year =

  43. [55]

    E-commerce and cultural values , year =

    Kong, Wei-Chang , type =. E-commerce and cultural values , year =

  44. [56]

    Chapter 9 , booktitle =

    Kong, Wei-Chang , editor =. Chapter 9 , booktitle =

  45. [57]

    E-commerce and cultural values , editor =

    Kong, Wei-Chang , title =. E-commerce and cultural values , editor =. 2003 , isbn =

  46. [58]

    E-commerce and cultural values - (InBook-num-in-chap) , chapter =

    Kong, Wei-Chang , editor =. E-commerce and cultural values - (InBook-num-in-chap) , chapter =. 2004 , address =

  47. [59]

    E-commerce and cultural values (Inbook-text-in-chap) , chapter =

    Kong, Wei-Chang , editor =. E-commerce and cultural values (Inbook-text-in-chap) , chapter =. 2005 , address =

  48. [60]

    E-commerce and cultural values (Inbook-num chap) , chapter =

    Kong, Wei-Chang , editor =. E-commerce and cultural values (Inbook-num chap) , chapter =. 2006 , address =

  49. [61]

    Microelectron

    Mehdi Saeedi and Morteza Saheb Zamani and Mehdi Sedighi , title =. Microelectron. J. , volume =. 2010 , pages =

  50. [62]

    Mehdi Saeedi and Morteza Saheb Zamani and Mehdi Sedighi and Zahra Sasanian , title =. J. Emerg. Technol. Comput. Syst. , volume =

  51. [63]

    Kirschmer, Markus and Voight, John , title =. SIAM J. Comput. , issue_date =. 2010 , issn =. doi:https://doi.org/10.1137/080734467 , acmid =

  52. [64]

    Hoare, C. A. R. , title =. Structured programming (incoll) , editor =. 1972 , isbn =

  53. [65]

    History of programming languages I (incoll) , editor =

    Lee, Jan , title =. History of programming languages I (incoll) , editor =. 1981 , isbn =. doi:http://doi.acm.org/10.1145/800025.1198348 , acmid =

  54. [66]

    , title =

    Dijkstra, E. , title =. Classics in software engineering (incoll) , year =

  55. [67]

    , title =

    Wenzel, Elizabeth M. , title =. Multimedia interface design (incoll) , year =. doi:10.1145/146022.146089 , acmid =

  56. [68]

    , title =

    Mumford, E. , title =. Critical issues in information systems research (incoll) , year =

  57. [69]

    and Golden, Donald G

    McCracken, Daniel D. and Golden, Donald G. , title =. 1990 , isbn =

  58. [70]

    The analysis of linear partial differential operators

    H. The analysis of linear partial differential operators. 1985 , PAGES =

  59. [71]

    IEEE", address =

    A. Adya and P. Bahl and J. Padhye and A.Wolman and L. Zhou , title =. Proceedings of the IEEE 1st International Conference on Broadnets Networks (BroadNets'04) , publisher = "IEEE", address = "Los Alamitos, CA", year =

  60. [72]

    I. F. Akyildiz and W. Su and Y. Sankarasubramaniam and E. Cayirci , title =. Comm. ACM , volume = 38, number = "4", year =

  61. [73]

    I. F. Akyildiz and T. Melodia and K. R. Chowdhury , title =. Computer Netw. , volume = 51, number = "4", year =

  62. [74]

    ACM", address =

    P. Bahl and R. Chancre and J. Dungeon , title =. Proceeding of the 10th International Conference on Mobile Computing and Networking (MobiCom'04) , publisher = "ACM", address = "New York, NY", year =

  63. [75]

    8 (Special Issue on Sensor Networks)

    D. Culler and D. Estrin and M. Srivastava , title =. IEEE Comput. , volume = 37, number = "8 (Special Issue on Sensor Networks)", publisher = "IEEE", address = "Los Alamitos, CA", year =

  64. [76]

    Natarajan and M

    A. Natarajan and M. Motani and B. de Silva and K. Yap and K. C. Chua , title =. Network Architectures , editor =. 960935712

  65. [77]

    Tzamaloukas and J

    A. Tzamaloukas and J. J. Garcia-Luna-Aceves , title =

  66. [78]

    Zhou and J

    G. Zhou and J. Lu and C.-Y. Wan and M. D. Yarvis and J. A. Stankovic , title =

  67. [79]

    Mapping Powerlists onto Hypercubes

    Jacob Kornerup. Mapping Powerlists onto Hypercubes. 1994

  68. [80]

    Automatic Parallelization for Distributed-Memory Multiprocessing Systems

    Michael Gerndt. Automatic Parallelization for Distributed-Memory Multiprocessing Systems

  69. [81]

    J. E. Archer, Jr. and R. Conway and F. B. Schneider. User recovery and reversal in interactive systems. ACM Trans. Program. Lang. Syst

  70. [82]

    D. D. Dunlop and V. R. Basili. Generalizing specifications for uniformly implemented loops. ACM Trans. Program. Lang. Syst

  71. [83]

    Heering and P

    J. Heering and P. Klint. Towards monolingual programming environments. ACM Trans. Program. Lang. Syst

  72. [84]

    Donald E. Knuth. The book

  73. [85]

    Korach and D

    E. Korach and D. Rotem and N. Santoro. Distributed algorithms for finding centers and medians in networks. ACM Trans. Program. Lang. Syst

  74. [86]

    : A Document Preparation System

    Leslie Lamport. : A Document Preparation System

  75. [87]

    F. Nielson. Program transformations in a denotational setting. ACM Trans. Program. Lang. Syst

  76. [88]

    Brian K. Reid. A high-level approach to computer document formatting. Proceedings of the 7th Annual Symposium on Principles of Programming Languages

  77. [89]

    and Abdelzaher, Tarek F

    Zhou, Gang and Wu, Yafeng and Yan, Ting and He, Tian and Huang, Chengdu and Stankovic, John A. and Abdelzaher, Tarek F. , title =. ACM Trans. Embed. Comput. Syst. , issue_date =. doi:10.1145/1721695.1721705 , acmid = 1721705, publisher =

  78. [90]

    Institutional members of the Users Group

  79. [91]

    Boris Veytsman , title =

  80. [92]

    Robin Schneider , title =

  81. [93]

    and Peterson, Larry L

    Bowman, Mic and Debray, Saumya K. and Peterson, Larry L. , title =. ACM Trans. Program. Lang. Syst. , volume =. 1993 , doi =

  82. [94]

    TUGboat , volume =

    Braams, Johannes , title =. TUGboat , volume =

  83. [95]

    Post Congress Tristesse

    Malcolm Clark. Post Congress Tristesse. TeX90 Conference Proceedings

  84. [96]

    ACM Trans

    Herlihy, Maurice , title =. ACM Trans. Program. Lang. Syst. , volume =. 1993 , doi =

  85. [97]

    Salas and Einar Hille

    S.L. Salas and Einar Hille. Calculus: One and Several Variable. 1978

  86. [98]

    Publication quality tables in

    Simon Fear , month =. Publication quality tables in

  87. [99]

    Using the amsthm Package , organization =

  88. [100]

    2019 , url =

    R: A Language and Environment for Statistical Computing , author =. 2019 , url =

  89. [101]

    Sam Anzaroot and Andrew McCallum , title =

  90. [102]

    Brad and Haunschild, Robin , title =

    Bornmann, Lutz and Wray, K. Brad and Haunschild, Robin , title =

  91. [103]

    2014 , archivePrefix =

    Sam Anzaroot and Alexandre Passos and David Belanger and Andrew McCallum , title =. 2014 , archivePrefix =. 1403.1349 , doi =

  92. [104]

    Proceedings of the 20th International Colloquium on Automata, Languages and Programming , series =

    Maintaining Discrete Probability Distributions Optimally , author =. Proceedings of the 20th International Colloquium on Automata, Languages and Programming , series =

  93. [105]

    Reiser , year = 2014, title =

    Brian J. Reiser , year = 2014, title =

  94. [106]

    Baggett and M

    R. Baggett and M. Simecek and C. Chambellan and K. Tsui and M. Fraune , year = 2025, title =

Pith tools

Reviewed July 12, 2026 · model on record in the stance chip above.