Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

MC$^2$A: Enabling Algorithm-Hardware Co-Design for Efficient Markov Chain Monte Carlo Acceleration

T0 review · 3 major / 5 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 programmable MCMC accelerator using a Gumbel-max sampler claims 307.6x speedups over CPU, GPU, TPU, and prior MCMC chips by sampling from unnormalized energies without exponential or normalization operations.

desk verdict A genuinely new programmable MCMC accelerator with a clever Gumbel sampler, but the headline speedups rest on an unvalidated approximation that could bias posterior samples; still well worth a serious referee. read the letter →

arxiv 2507.12935 v1 pith:HVYDBLS6 submitted 2025-07-17 cs.LG cs.AIcs.AR

Shirui Zhao , Jun Yin , Lingyun Yao , Martin Andraud , Wannes Meert , Marian Verhelst This is my paper · ORCID
classification cs.LGcs.AIcs.AR
keywords MarkovChainMonteCarlohardwareacceleratorGumbel-maxtrickalgorithm-hardwareco-designrooflinemodelprobabilisticcomputingdiscretesamplingenergy-basedmodels
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

MC2A claims that one programmable accelerator can serve the main MCMC algorithm families — sequential Metropolis-Hastings/Gibbs, block and asynchronous Gibbs, and gradient-based samplers like PAS — across Bayes nets, MRF/Ising models, combinatorial optimization, and restricted Boltzmann machines. The paper's design method extends the processor roofline model with a third axis for sampler throughput, and uses it to pick a balanced set of hardware parameters: 64 tree-structured compute units, 64 sampler elements, and a 4.8 MB on-chip memory. The enabling trick is a Gumbel-max sampler that adds lookup-table Gumbel noise directly to unnormalized log-energies and takes the argmax, removing the exponential and normalization steps of a CDF sampler and doubling sampling throughput without area cost. As reported, the resulting 500 MHz 16 nm accelerator beats CPU, GPU, TPU, and prior MCMC accelerators by 307.6x, 1.4x, 2.0x, and 84.2x respectively on the evaluated workloads. The paper presents the Gumbel LUT accuracy only as 'good-enough' on a MaxCut problem and 100 random distributions, not as a proven bound on the stationary-distribution bias.

What carries the argument

The object that carries the argument is the reconfigurable Gumbel-based sampler unit (SU), paired with the 3D roofline model that sizes it. The Gumbel-max trick (Eq. 4) converts sampling from an unnormalized distribution into an argmax over perturbed log-energies: $g_j = -\log(-\log(u_j)) + \log p_j$, with $u_j$ uniform; real Gumbel noise makes the argmax exact. In MC2A, the noise is generated by a 16-entry, 8-bit lookup table, and the comparison is done by a chain of comparators in either temporal mode (iterating over one large distribution) or spatial mode (a 2D tree for many distributions at once). The 3D roofline extends the classic two-dimensional roofline with a third axis, sampler throughput, and expresses the achievable performance as the minimum of compute-limited, memory-limited, and sampler-limited roofs; this is what picks the reported configuration (T=64, K=3, S=64, M=6, B=320, 4.8 MB SRAM). The claim is that this combination removes the two most expensive operator classes of discrete MCMC — the exponential function and the normalization sum — and replaces a sequential CDF search with a pipelined maximum, which is why the throughput doubles with no added area.

What would settle it

Run a software emulation of MC2A's sampler with the same 16-entry, 8-bit LUT on a small Bayes net or RBM where the exact posterior is computable, and compare the empirical distribution of a long chain with the exact posterior; if the total-variation distance is significantly larger than that of a full-precision Gumbel-max sampler, the approximation biases the chain's stationary distribution.

Watch

Extended reading notes

Core claim

The central claim is that the computationally painful part of MCMC — converting energies to probabilities and then sampling — can be replaced by a much cheaper maximum-selection operation without changing the algorithm's semantics. The Gumbel-max trick states that for a categorical distribution with weights $w_j = \exp(-\beta E_j)$, the index $\arg\max_j (g_j - \beta E_j)$ with i.i.d. Gumbel noise $g_j$ is a sample from that distribution, so the sampler can work directly on energies and skip both the exponential and the normalization. MC2A implements this with a 16-entry, 8-bit lookup table that converts uniform random numbers to approximate Gumbel noise, and a comparator chain that can run temporally (one distribution over many cycles) or spatially (many distributions in parallel). Together with a tree-structured compute unit and a crossbar for irregular graph access, this makes the accelerator programmable across workloads, and the paper reports that it sustains the claimed speedups while supporting distributions as large as 1347. The paper argues that a single, flexible, co-designed architecture can popularize MCMC by making it fast enough for real-time and large-scale use.

Load-bearing premise

The 16-entry, 8-bit lookup table produces Gumbel noise that is close enough to true i.i.d. Gumbel noise that every MCMC chain still converges to the intended target distribution.

Editorial extensions

If this is right

  • An accelerator built to these parameters sustains 307.6x, 1.4x, 2.0x, and 84.2x speedups versus CPU, GPU, TPU, and the best prior MCMC accelerator on the tested workloads.
  • The Gumbel sampler supports distributions far larger than the CDT register files of prior CDF-based samplers, since it only needs one noise value per category and a running maximum.
  • The 3D roofline gives a concrete recipe using compute intensity, memory intensity, and sampler throughput to choose CU, SU, and memory sizes for any MCMC workload.
  • The ISA plus compiler hides pipeline hazards and lets one core cover temporal, sequential sampling and spatial, parallel sampling modes, so the same hardware handles Gibbs-style and gradient-based samplers.
  • Energy efficiency improves by about 10,000x, 355x, and 197.5x over CPU, GPU, and TPU for the reported workloads.

Reading between the lines

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

  • The load-bearing assumption is that the 16-entry, 8-bit LUT approximates i.i.d. Gumbel noise closely enough to preserve the target stationary distribution; the paper validates 'good-enough accuracy' only on MaxCut and random distributions, which tests optimization quality, not posterior fidelity, so a bias on Bayes-net or RBM posteriors would make the speedups hollow for inference workloads.
  • A direct test would be to emulate the quantized Gumbel noise in software, run the same chains, and compare sample marginals against exact inference on a small Bayes net; a total-variation gap beyond the LUT's quantization error would falsify the convergence claim.
  • The 3D roofline methodology is likely portable to other probabilistic accelerators such as p-bit circuits, processing-in-memory samplers, or Ising machines, because it only depends on the compute/sample/memory abstraction, though the paper does not demonstrate that transfer.
  • Since the paper accelerates a single MCMC chain only, combining multiple MC2A cores with chain-level parallelism could narrow the remaining gap to GPU and TPU on highly parallel structured workloads; this is an easy extension the paper mentions but does not evaluate.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces MC2A, an algorithm-hardware co-design framework for accelerating Markov Chain Monte Carlo workloads. It extends the processor roofline model to three dimensions (compute intensity, memory intensity, throughput) to identify balanced hardware configurations, and proposes a parameterizable accelerator with ISA-programmable tree-structured compute units, reconfigurable Gumbel-based sampler units, a crossbar interconnect, and a customized compiler. The authors evaluate the design by RTL synthesis in 16 nm at 500 MHz and a cycle-accurate simulator on Bayes nets, MRF/Ising models, combinatorial optimization problems, and RBMs, reporting large speedups over CPU, GPU, TPU, and prior MCMC accelerators.

Significance. If the correctness and performance claims hold, MC2A would be a significant contribution: it is one of the first programmable general-purpose MCMC accelerators with a systematic design-space exploration methodology, and the idea of using a Gumbel-max sampler to avoid exponential and normalization operations is attractive and potentially impactful. The paper provides concrete engineering artifacts—SystemVerilog RTL synthesized with Cadence Genus in 16 nm, a cycle-accurate simulator, and a roofline-based parameter selection flow—which strengthen the architecture-level claims. However, the central statistical correctness of the Gumbel LUT sampler is not established to the standard required for sampling-based inference, and the co-design evaluation is partly circular with the parameter selection. These issues are load-bearing for the headline speedup claims.

major comments (3)
  1. [Section V-D, Eq. (4), Section VI-C] The Gumbel-max trick is exact only for continuous i.i.d. Gumbel noise, but the MC2A sampler uses a 16-entry, 8-bit lookup table to generate Gumbel noise. With quantized noise, the probability of selecting category k is no longer p_k in general, and the deviation can be large when several categories have similar energies. Section VI-C validates the LUT only with MaxCut solution quality and agreement on 100 random distributions sampled 10^6 times; neither test measures whether output samples match the target stationary distribution, which is the relevant correctness criterion for the Bayes net, MRF/Ising, and RBM workloads in Table I. The paper should provide a total-variation or KL bound from the LUT approximation, or at minimum empirical posterior/energy-distribution comparisons with an exact sampler on the actual benchmark workloads, before the throughput claims can be interpreted as sampling from the intended distributions.
  2. [Section VI-B, Fig. 6(d), Section VI-D] The hardware parameters T=64, K=3, B=320 are selected in Section VI-B by placing representative workloads at the apex of the 3D roofline, and the evaluation in Section VI-D uses the same workload families (Bayes net, MRF, COP, RBM) from Table I. Because the 'golden configuration' is defined as the point where CI*CU performance equals MI*BW equals SU peak performance, the later performance results partly confirm the design choice by construction. To support the claimed generality of the co-design methodology, the authors should either evaluate on held-out workloads not used for parameter selection or provide a sensitivity analysis showing that nearby configurations degrade gracefully across the workload set.
  3. [Section VI-A, Section VI-D] The baseline setup is not sufficiently specified to reproduce the headline speedups. Section VI-A lists an RTX 2080Ti GPU and Xeon CPU as baselines, while Section VI-D reports a Tesla V100 and TPU-v3 core comparison; the CPU model, GPU driver/library versions, compiler flags, energy measurement methodology (TDP versus measured power), and whether all platforms use identical data types and single-chain execution are not fully stated. The reported speedup tuple {307.6x, 1.4x, 2.0x, 84.2x} should be accompanied by reproducible measurement details, including exact software versions and power/energy measurement protocols.
minor comments (5)
  1. [Section V-D] The text states that if Xi is a random variable with a discrete distribution, then -log(-log(Xi)) follows a Gumbel distribution; this is incorrect as written and should refer to a uniform random variable, since the Gumbel-max trick requires U(0,1) inputs.
  2. [Section VI-C] The ablation study in Fig. 12 would be more informative with error bars, a statistical test, or a reported metric such as maximum selection-probability error or total variation distance over the random distributions, rather than only a qualitative 'good-enough accuracy' statement.
  3. [Section I, Section VI-D] There are several typos, including 'optimaziation' in the Introduction, 'Gumble' in Section VI-C, and 'paramters' in Section VI-D; these should be corrected.
  4. [Section V-D] The claim that the Gumbel sampler increases throughput by 2x 'without area overhead' would benefit from a direct area comparison of the Gumbel SU versus the CDF SU, including the LUT and comparator resources, since the CDF sampler also requires a register file for the CDT.
  5. [Section VI-D] In the comparison with CoopMC, the text refers to 'MC 2' where MC2A is meant; please correct this label for clarity.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the roofline-based parameter selection is a bottleneck-balancing design heuristic, and the Gumbel LUT quantization issue is an unverified correctness risk rather than a definitional reduction.

full rationale

The paper's claims do not reduce to their own inputs. The 3D roofline model (Section IV) defines global throughput as min(CI*CU performance, MI*BW, SU performance), which is a resource-bound identity. The 'golden configuration' is the point where the three roofs are equalized; choosing S=T=64, K=3, and B=320 to sit near that apex (Section VI-B) is a conventional bottleneck-balancing design choice, not a fitted prediction. The headline speedups (Section VI-D) come from RTL synthesis at 16 nm and a cycle-accurate simulator, not from reading the roofline model, so the evaluation does not confirm the model by construction. The Gumbel sampler claim (Section V-D) rests on the standard Gumbel-max trick; replacing continuous Gumbel noise with a 16-entry 8-bit LUT (Section VI-C) is an approximation whose posterior-fidelity cost is not bounded, but this is a correctness/robustness concern, not circularity: the LUT is not defined in terms of, or fitted to, the target speedups or the reported accuracies. The only self-reference (prior AIA work in Fig. 2) is illustrative and not load-bearing. No step in the paper equates a fitted parameter with a prediction or imports a contested premise via a self-citation chain.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central performance claims rest on (i) statistical validity of the LUT-approximated Gumbel sampler, (ii) fidelity of the cycle-accurate simulator, and (iii) representativeness of the benchmark workloads used both to choose hardware parameters (Section VI-B) and to report speedups (Section VI-D). No new physical entities are introduced; the free parameters are design-time choices fitted to the evaluation workloads.

free parameters (5)
  • CU size T = 64 = 64
    Chosen in Section VI-B 'from our budget' and to fit the benchmark workloads; directly sets the compute roof in the roofline model.
  • CU tree depth K = 3 = 3
    Set in Section VI-B to push the spatial-mode roof apex toward the PAS workloads; affects the maximum PE input size 2^K + 1.
  • SU size S = 64, sampler depth M = 6 = S = 64, M = 6
    Chosen in Section VI-B from the resource budget and workload analysis; sets the SU peak throughput roof.
  • Memory bandwidth B = 320 banks = 320
    Set in Section VI-B after 'the memory bottleneck is cleared'; drives the memory-bound corner of the roofline and the 4.8 MB SRAM size.
  • Gumbel LUT size and precision = 16 entries, 8 bits
    Selected by ablation in Section VI-C on MaxCut and 100 random distributions; the resulting sampling bias is not characterized outside these cases.
assumptions (4)
  • domain assumption The Gumbel-max trick with quantized, truncated noise (16-entry, 8-bit LUT) yields samples statistically close enough to the target categorical distribution for MCMC correctness
    Section V-D replaces the i.i.d. standard Gumbel noise required by Eq. (4) with LUT-quantized noise; Section VI-C validates only MaxCut accuracy and 100 random draws, so distributional fidelity on other workloads is assumed.
  • domain assumption The 3D roofline abstraction (computation intensity, memory intensity, throughput) captures the dominant system bottlenecks
    Section IV states that internal CU-SU communication constraints 'can be addressed during the tightly-coupled hardware design', but the crossbar interconnect (size B:T*2^K) is precisely where irregular-access contention appears, so the model ignores a potential bottleneck.
  • domain assumption The cycle-accurate simulator faithfully reproduces the synthesized RTL and the RTL matches the architectural model
    Section VI-A reports RTL synthesis with Cadence Genus at 16 nm and 500 MHz, but all performance figures come from the simulator; no silicon or post-layout comparison is provided, so simulator fidelity is assumed.
  • domain assumption Log-domain fixed-point storage of CPTs and weights preserves MCMC accuracy at the chosen bit widths
    Section VI-A stores CPTs in logarithmic form in 32-bit banks; Section VI-D states int32 data is used for fairness with PGMA; no precision analysis is given, so adequacy of the representation is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MC$^2$A: Enabling Algorithm-Hardware Co-Design for Efficient Markov Chain Monte Carlo Acceleration." pith.science (2026). https://pith.science/paper/HVYDBLS6

@misc{pith2026250712935,
  author       = {Pith},
  title        = {Pith review of: MC$^2$A: Enabling Algorithm-Hardware Co-Design for Efficient Markov Chain Monte Carlo Acceleration},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HVYDBLS6}},
  note         = {Machine review of arXiv:2507.12935}
}
abstract

An increasing number of applications are exploiting sampling-based algorithms for planning, optimization, and inference. The Markov Chain Monte Carlo (MCMC) algorithms form the computational backbone of this emerging branch of machine learning. Unfortunately, the high computational cost limits their feasibility for large-scale problems and real-world applications, and the existing MCMC acceleration solutions are either limited in hardware flexibility or fail to maintain efficiency at the system level across a variety of end-to-end applications. This paper introduces \textbf{MC$^2$A}, an algorithm-hardware co-design framework, enabling efficient and flexible optimization for MCMC acceleration. Firstly, \textbf{MC$^2$A} analyzes the MCMC workload diversity through an extension of the processor performance roofline model with a 3rd dimension to derive the optimal balance between the compute, sampling and memory parameters. Secondly, \textbf{MC$^2$A} proposes a parametrized hardware accelerator architecture with flexible and efficient support of MCMC kernels with a pipeline of ISA-programmable tree-structured processing units, reconfigurable samplers and a crossbar interconnect to support irregular access. Thirdly, the core of \textbf{MC$^2$A} is powered by a novel Gumbel sampler that eliminates exponential and normalization operations. In the end-to-end case study, \textbf{MC$^2$A} achieves an overall {$307.6\times$, $1.4\times$, $2.0\times$, $84.2\times$} speedup compared to the CPU, GPU, TPU and state-of-the-art MCMC accelerator. Evaluated on various representative MCMC workloads, this work demonstrates and exploits the feasibility of general hardware acceleration to popularize MCMC-based solutions in diverse application domains.

Figures

Figures reproduced from arXiv: 2507.12935 by the authors.

Figure 1
Figure 1. Markov Chain Monte Carlo (MCMC) methods are [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Comparison of MCMC accelerator structures: (a) The [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The overview of the MCMC processing flow, in view of (a) the key energy functions in different MCMC applications [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Illustration of the Random Variable (RV) level paral [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: The demonstration of MCMC hardware challenges [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: MC2A 3D Roofline Model: (a) The 3D rectangular frustum that models the hybrid MCMC CU-SU hardware system. The peak hardware performances and their mutual data dependencies form three different performance boundary roofs and bottleneck corners. The purple star highlight…
Figure 7
Figure 7. Figure 7: The overview of MC2A top-level architecture: (a) The 4-stage pipelined and parameterizable architecture for tightly￾coupled computing-sampling MCMC acceleration. (b) The pipelined mapping of control signals to each hardware modules. (c) The definition of the MC2A Instr…
Figure 8
Figure 8. Figure 8: The structure of reconfigurable core elements: (a) [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: The design comparison of CDF and Gumbel sampler [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: The schedule of flexible processing flows for different MCMC applications: Block Gibbs sampling for (a) Bayes Net [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 12
Figure 12. Figure 12: The accuracy performance of different Gumbel LUT [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]
Figure 15
Figure 15. Figure 15: Energy Efficiency Improvement on structured graph. [PITH_FULL_IMAGE:figures/full_fig_p012_15.png]
Figure 14
Figure 14. Figure 14: Latency comparison. MC2A works under temporal mode for Bayes Net and MRF. For the key step in PAS sampling algorithm where the SU is under spatial mode. Compared to other computing platforms, MC2A is shown higher throughput and flexibility for various workloads. PGMA[…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. MCHA: A Memory-Centric Hierarchical Architecture for Parallel-Sequential Computing

    cs.AR 2026-08 conditional novelty 6.0 of 10

    MCHA combines a distributed memory fabric with data-driven triggers to execute parallel-sequential workloads, claiming 153x to 2456x MARL speedups over an A100 GPU.

Reference graph

Works this paper leans on

52 extracted references · 42 canonical work pages · cited by 1 Pith paper

  1. [1]

    LIGO Algorithm Library - LALSuite,

    LIGO Scientific Collaboration, Virgo Collaboration, and KAGRA Col- laboration, “LIGO Algorithm Library - LALSuite,” Free software (GPL), 2018

    work page 2018

  2. [2]

    Near-optimal mimo detection using gradient-based mcmc in discrete spaces,

    X. Zhou, L. Liang, J. Zhang, C.-K. Wen, and S. Jin, “Near-optimal mimo detection using gradient-based mcmc in discrete spaces,” IEEE Transactions on Signal Processing , 2024

    work page 2024

  3. [3]

    Revisiting sampling for combinatorial optimization,

    H. Sun, K. Goshvadi, A. Nova, D. Schuurmans, and H. Dai, “Revisiting sampling for combinatorial optimization,” in Proceedings of the 40th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research, A. Krause, E. Brunskill, K. Cho, B. Engelhardt, S. Sabato, and J. Scarlett, Eds., vol. 202. PMLR, 23–29 Jul 2023, pp. 32 859–...

    work page 2023

  4. [4]

    Gurobi Optimizer Reference Manual,

    Gurobi Optimization, LLC, “Gurobi Optimizer Reference Manual,”

  5. [5]

    Dimes: A differentiable meta solver for combinatorial optimization problems,

    R. Qiu, Z. Sun, and Y . Yang, “Dimes: A differentiable meta solver for combinatorial optimization problems,” Advances in Neural Information Processing Systems, vol. 35, pp. 25 531–25 546, 2022

    work page 2022

  6. [6]

    Combinatorial optimization with graph convolutional networks and guided tree search,

    Z. Li, Q. Chen, and V . Koltun, “Combinatorial optimization with graph convolutional networks and guided tree search,” Advances in neural information processing systems , vol. 31, 2018

    work page 2018

  7. [7]

    Difusco: Graph-based diffusion solvers for combinatorial optimization,

    Z. Sun and Y . Yang, “Difusco: Graph-based diffusion solvers for combinatorial optimization,” Advances in neural information processing systems, vol. 36, pp. 3706–3731, 2023

    work page 2023

  8. [8]

    Guest Editors Introduction to the top 10 algorithms ,

    J. Dongarra and F. Sullivan, “ Guest Editors Introduction to the top 10 algorithms ,” Computing in Science & Engineering , vol. 2, no. 01, pp. 22–23, Jan. 2000. [Online]. Available: https: //doi.ieeecomputersociety.org/10.1109/MCISE.2000.814652

    work page arXiv

Show all 52 references
  1. [9]

    G. E. Box and G. C. Tiao, Bayesian inference in statistical analysis . John Wiley & Sons, 2011

    2011

  2. [10]

    Probabilistic machine learning and artificial intelli- gence,

    Z. Ghahramani, “Probabilistic machine learning and artificial intelli- gence,” Nature, vol. 521, no. 7553, pp. 452–459, 2015

    2015

  3. [11]

    An introduction to mcmc for machine learning,

    C. Andrieu, N. De Freitas, A. Doucet, and M. I. Jordan, “An introduction to mcmc for machine learning,” Machine learning , vol. 50, pp. 5–43, 2003

    2003

  4. [12]

    Koller and N

    D. Koller and N. Friedman, Probabilistic graphical models: principles and techniques. MIT press, 2009

    2009

  5. [13]

    J. S. Liu and J. S. Liu, Monte Carlo strategies in scientific computing . Springer, 2001, vol. 10

    2001

  6. [14]

    DISCS: A Benchmark for Discrete Sampling,

    Katayoon Goshvadi, Haoran Sun, Xingchao Liu, Azade Nova, Ruqi Zhang, Will Grathwohl, Dale Schuurmans, and Hanjun Dai, “DISCS: A Benchmark for Discrete Sampling,” Neural Information Processing Systems, 2023

    2023

  7. [15]

    A tutorial on energy-based learning,

    Y . LeCun, S. Chopra, R. Hadsell, M. Ranzato, F. Huanget al., “A tutorial on energy-based learning,” Predicting structured data , vol. 1, no. 0, 2006

    2006

  8. [16]

    Fast, accurate training and sampling of restricted boltzmann machines,

    N. B ´ereux, A. Decelle, C. Furtlehner, L. Rosset, and B. Seoane, “Fast, accurate training and sampling of restricted boltzmann machines,” arXiv preprint arXiv:2405.15376, 2024

    2024

  9. [17]

    Parameter estimation with gravitational waves,

    N. Christensen and R. Meyer, “Parameter estimation with gravitational waves,” Reviews of modern physics , vol. 94, no. 2, p. 025001, 2022

    2022

  10. [18]

    A novel monte-carlo-sampling-based receiver for large-scale uplink multiuser mimo systems,

    T. Datta, N. A. Kumar, A. Chockalingam, and B. S. Rajan, “A novel monte-carlo-sampling-based receiver for large-scale uplink multiuser mimo systems,” IEEE Transactions on Vehicular Technology , vol. 62, no. 7, pp. 3019–3038, 2013

    2013

  11. [19]

    Large-scale mimo detection using mcmc approach with blockwise sampling,

    L. Bai, T. Li, J. Liu, Q. Yu, and J. Choi, “Large-scale mimo detection using mcmc approach with blockwise sampling,” IEEE Transactions on Communications, vol. 64, no. 9, pp. 3697–3707, 2016

    2016

  12. [20]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” Communications of the ACM, vol. 60, no. 6, pp. 84–90, 2017

    2017

  13. [21]

    Natural language processing,

    K. Chowdhary and K. Chowdhary, “Natural language processing,” Fundamentals of artificial intelligence , pp. 603–649, 2020

    2020

  14. [22]

    Monte carlo sampling methods using markov chains and their applications,

    W. K. Hastings, “Monte carlo sampling methods using markov chains and their applications,” 1970

    1970

  15. [23]

    Illustration of bayesian inference in normal data models using gibbs sampling,

    A. E. Gelfand, S. E. Hills, A. Racine-Poon, and A. F. Smith, “Illustration of bayesian inference in normal data models using gibbs sampling,” Journal of the American Statistical Association , vol. 85, no. 412, pp. 972–985, 1990

    1990

  16. [24]

    Blocking gibbs sampling in very large probabilistic expert systems,

    C. S. Jensen, U. Kjærulff, and A. Kong, “Blocking gibbs sampling in very large probabilistic expert systems,” International Journal of Human-Computer Studies, vol. 42, no. 6, pp. 647–666, 1995

    1995

  17. [25]

    Asynchronous gibbs sampling,

    A. Terenin, D. Simpson, and D. Draper, “Asynchronous gibbs sampling,” in International Conference on Artificial Intelligence and Statistics . PMLR, 2020, pp. 144–154

    2020

  18. [26]

    Path auxiliary proposal for mcmc in discrete space,

    H. Sun, H. Dai, W. Xia, and A. Ramamurthy, “Path auxiliary proposal for mcmc in discrete space,” in International Conference on Learning Representations, 2021

    2021

  19. [27]

    A langevin-like sampler for discrete dis- tributions,

    R. Zhang, X. Liu, and Q. Liu, “A langevin-like sampler for discrete dis- tributions,” in International Conference on Machine Learning . PMLR, 2022, pp. 26 375–26 396

    2022

  20. [28]

    A 3mm 2 programmable bayesian inference accelerator for unsupervised machine perception using parallel gibbs sampling in 16nm,

    G. G. Ko, Y . Chai, M. Donato, P. N. Whatmough, T. Tambe, R. A. Rutenbar, D. Brooks, and G.-Y . Wei, “A 3mm 2 programmable bayesian inference accelerator for unsupervised machine perception using parallel gibbs sampling in 16nm,” in 2020 IEEE Symposium on VLSI Circuits . IEEE,...

    2020

  21. [29]

    Coopmc: Algorithm-architecture co-optimization for markov chain monte carlo accelerators,

    Y . Chai, G. G. Ko, L. Bailey, D. Brooks, G.-Y . Wei et al. , “Coopmc: Algorithm-architecture co-optimization for markov chain monte carlo accelerators,” in 2022 IEEE International Symposium on High-Performance Computer Architecture (HPCA) . IEEE, 2022, pp. 38–52. 14

    2022

  22. [30]

    Proca: Programmable probabilistic processing unit archi- tecture with accept/reject prediction & multicore pipelining for causal inference,

    Y . Fu et al., “Proca: Programmable probabilistic processing unit archi- tecture with accept/reject prediction & multicore pipelining for causal inference,” in 2025 IEEE International Symposium on High-Performance Computer Architecture (HPCA). IEEE, 2025

    2025

  23. [31]

    Statistical robustness of markov chain monte carlo accelerators,

    X. Zhang, R. Bashizade, Y . Wang, S. Mukherjee, and A. R. Lebeck, “Statistical robustness of markov chain monte carlo accelerators,” in Proceedings of the 26th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, ser. ASPLOS ’2...

    2021

  24. [32]

    Massively parallel probabilistic computing with sparse ising machines,

    N. A. Aadit, A. Grimaldi, M. Carpentieri, L. Theogarajan, J. M. Martinis, G. Finocchio, and K. Y . Camsari, “Massively parallel probabilistic computing with sparse ising machines,” Nature Electronics, vol. 5, no. 7, pp. 460–468, 2022

    2022

  25. [33]

    Towards 3d cmos+x ising machines: Addressing the connectivity prob- lem with back-end-of-line fefets,

    T. H. Pantha, A. Khanna, H. Ye, S. Niazi, B. Chakraborty, E. Weinstock, N. Babu, S. Mukhopadhyay, S. Datta, K. Camsari, and S. Dutta, “Towards 3d cmos+x ising machines: Addressing the connectivity prob- lem with back-end-of-line fefets,” in 2024 IEEE International Electron Dev...

    2024

  26. [34]

    Ising machines as hard- ware solvers of combinatorial optimization problems,

    N. Mohseni, P. L. McMahon, and T. Byrnes, “Ising machines as hard- ware solvers of combinatorial optimization problems,” Nature Reviews Physics, vol. 4, no. 6, pp. 363–379, 2022

    2022

  27. [35]

    A full-stack view of probabilistic computing with p-bits: Devices, architectures, and algorithms,

    S. Chowdhury, A. Grimaldi, N. A. Aadit, S. Niazi, M. Mohseni, S. Kanai, H. Ohno, S. Fukami, L. Theogarajan, G. Finocchio et al. , “A full-stack view of probabilistic computing with p-bits: Devices, architectures, and algorithms,” IEEE Journal on Exploratory Solid-State Computa...

    2023

  28. [36]

    Acmc 2: Accelerating markov chain monte carlo algorithms for probabilistic models,

    S. S. Banerjee, Z. T. Kalbarczyk, and R. K. Iyer, “Acmc 2: Accelerating markov chain monte carlo algorithms for probabilistic models,” in Proceedings of the Twenty-Fourth International Conference on Architectural Support for Programming Languages and Operating Systems, ser. AS...

    2019

  29. [37]

    12.2 p-circuits: Neither digital nor analog,

    M.-C. Li, A. Ghosh, R. Jaiswal, L. A. Ghantasala, B. Behin-Aein, S. Sen, and S. Datta, “12.2 p-circuits: Neither digital nor analog,” in 2025 IEEE International Solid-State Circuits Conference (ISSCC) , vol. 68. IEEE, 2025, pp. 1–3

    2025

  30. [38]

    P. J. Van Laarhoven, E. H. Aarts, P. J. van Laarhoven, and E. H. Aarts, Simulated annealing. Springer, 1987

    1987

  31. [39]

    PASS: An Asynchronous Probabilistic Processor for Next Generation Intelligence,

    S. Patel, P. Canoza, A. Datar, S. Lu, C. Garg, and S. Salahuddin, “PASS: An Asynchronous Probabilistic Processor for Next Generation Intelligence,” Sep. 2024

    2024

  32. [40]

    Discrete langevin samplers via wasserstein gradient flow,

    H. Sun, H. Dai, B. Dai, H. Zhou, and D. Schuurmans, “Discrete langevin samplers via wasserstein gradient flow,” in International Conference on Artificial Intelligence and Statistics . PMLR, 2023, pp. 6290–6313

    2023

  33. [41]

    No free lunch theorems for optimization,

    D. H. Wolpert and W. G. Macready, “No free lunch theorems for optimization,” IEEE transactions on evolutionary computation , vol. 1, no. 1, pp. 67–82, 1997

    1997

  34. [42]

    CausaLearn: Automated Framework for Scalable Streaming-based Causal Bayesian Learning using FPGAs,

    Bita Darvish Rouhani, B. D. Rouhani, Mohammad Ghasemzadeh, M. Ghasemzadeh, Farinaz Koushanfar, and F. Koushanfar, “CausaLearn: Automated Framework for Scalable Streaming-based Causal Bayesian Learning using FPGAs,” pp. 1–10, Feb. 2018

    2018

  35. [43]

    PMBA: A Parallel MCMC Bayesian Computing Accelerator,

    Yufei Ni, Y . Ni, Yansha Deng, Y . Deng, Songlin Li, and S. Li, “PMBA: A Parallel MCMC Bayesian Computing Accelerator,” IEEE Access, vol. 9, pp. 65 536–65 546, Apr. 2021

    2021

  36. [44]

    Running markov chain monte carlo on modern hardware and software,

    P. Sountsov, C. Carroll, and M. D. Hoffman, “Running markov chain monte carlo on modern hardware and software,” arXiv preprint arXiv:2411.04260, 2024

    2024 arXiv

  37. [45]

    Simple, distributed, and accelerated probabilistic program- ming,

    D. Tran, M. W. Hoffman, D. Moore, C. Suter, S. Vasudevan, and A. Radul, “Simple, distributed, and accelerated probabilistic program- ming,” Advances in Neural Information Processing Systems , vol. 31, 2018

    2018

  38. [46]

    Pyro: Deep universal probabilistic programming,

    E. Bingham, J. P. Chen, M. Jankowiak, F. Obermeyer, N. Pradhan, T. Karaletsos, R. Singh, P. Szerlip, P. Horsfall, and N. D. Goodman, “Pyro: Deep universal probabilistic programming,” Journal of machine learning research, vol. 20, no. 28, pp. 1–6, 2019

    2019

  39. [47]

    AIA: A 16nm Multicore SoC for Approximate Inference Acceleration Exploiting Non-normalized Knuth-Yao Sampling and Inter-Core Register Sharing,

    Shirui Zhao, Nimish Shah, Wannes Meert, and Marian Verhelst, “AIA: A 16nm Multicore SoC for Approximate Inference Acceleration Exploiting Non-normalized Knuth-Yao Sampling and Inter-Core Register Sharing,” 2024 IEEE European Solid-State Electronics Research Conference (ES- SERC), 2024

    2024

  40. [48]

    High performance monte carlo simulation of ising model on tpu clusters,

    K. Yang, Y .-F. Chen, G. Roumpos, C. Colby, and J. Anderson, “High performance monte carlo simulation of ising model on tpu clusters,” in Proceedings of the international conference for high performance computing, networking, storage and analysis , 2019, pp. 1–15

    2019

  41. [49]

    Bayes net repository,

    M. Scutari, “Bayes net repository,” 2025. [Online]. Available: https://www.bnlearn.com/bnrepository

    2025

  42. [50]

    agrum/pyagrum: a tool- box to build models and algorithms for probabilistic graphical models in python,

    G. Ducamp, C. Gonzales, and P.-H. Wuillemin, “agrum/pyagrum: a tool- box to build models and algorithms for probabilistic graphical models in python,” in International Conference on Probabilistic Graphical Models. PMLR, 2020

    2020

  43. [51]

    Bayeslib,

    “Bayeslib,” 2025. [Online]. Available: https://github.com/mspronesti/ baylib

    2025

  44. [2024]

    Available: https://www.gurobi.com

    [Online]. Available: https://www.gurobi.com

Pith tools

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