pith. sign in

arxiv: 2606.27361 · v1 · pith:PDXJCDKEnew · submitted 2026-06-25 · 💻 cs.LG · cs.AI

Autoregressive Boltzmann Generators

Danyal Rehman , Charlie B. Tan , Yoshua Bengio , Avishek Joey Bose , Alexander Tong This is my paper

Pith reviewed 2026-06-26 04:48 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords autoregressive modelsBoltzmann generatorsmolecular samplingnormalizing flowsimportance samplingpeptide systemsgenerative modelsthermodynamic equilibrium
0
0 comments X

The pith

Autoregressive models generate Boltzmann-distributed molecular samples without the constraints of normalizing flows.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces Autoregressive Boltzmann Generators (ArBG) to sample molecular systems at thermodynamic equilibrium by replacing flow-based generators with autoregressive architectures. This change removes invertibility requirements and allows use of scalable sequential models, yielding better accuracy on peptide benchmarks. A 132-million-parameter model named Robin trained in this framework cuts zero-shot energy error by more than 60 percent on 8-residue systems relative to prior methods. The approach keeps exact likelihoods so that importance sampling still recovers the target distribution. Results are strongest on larger systems such as the 10-residue Chignolin.

Core claim

ArBG replaces normalizing-flow Boltzmann generators with an autoregressive framework that produces sequential proposals whose exact likelihoods, after importance-sampling correction, recover the target Boltzmann distribution; the resulting models outperform flow-based baselines on all tested peptide systems and enable a 132-million-parameter transferable model that reduces E-W₂ error by over 60 percent on 8-residue cases.

What carries the argument

Autoregressive proposal model that supplies exact likelihoods for importance-sampling correction in place of invertible flow transformations.

If this is right

  • ArBG removes the topological and computational limits that restrict flow-based Boltzmann generators.
  • Sequential architectures effective for language modeling become directly usable for molecular sampling.
  • Inference-time interventions on individual residues become feasible because generation is sequential.
  • Performance gains are largest on bigger peptide systems where flow expressivity previously limited accuracy.
  • A single 132-million-parameter model can be trained to transfer across multiple residue lengths.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The framework may extend to other chain-like molecular systems such as proteins or RNA where sequential generation aligns with physical structure.
  • Combining ArBG proposals with modern large-scale training regimes could further lower the cost of equilibrium sampling in drug-discovery pipelines.
  • Because likelihoods remain exact, hybrid schemes that mix autoregressive steps with occasional flow or MCMC corrections could be explored without losing theoretical guarantees.

Load-bearing premise

An autoregressive model can be trained so that its proposals, once reweighted by importance sampling, recover the exact target Boltzmann distribution.

What would settle it

Training an autoregressive proposal model on a known small peptide and then measuring whether the importance-weighted energy histogram matches the reference Boltzmann distribution to within sampling noise.

Figures

Figures reproduced from arXiv: 2606.27361 by Alexander Tong, Avishek Joey Bose, Charlie B. Tan, Danyal Rehman, Yoshua Bengio.

Figure 1
Figure 1. Figure 1: (Left) The training loss curves of models of varying scale (ranging from 8M to 126M parameters) as a function of the total number of floating point operations (FLOPs) on the decapeptide: Chignolin. (Right) Global and local performance metrics as a function of the number of energy evaluations demonstrating inference-time scaling between our transferable model ROBIN, the previous state-of-the-art BG Prose, a… view at source ↗
Figure 2
Figure 2. Figure 2: Energy histogram of the proposal and re-sampled distribution compared to ground truth MD data for all single peptide systems. ¡¼ ¡¼ 2 0 ¼ 2 ¼ ¡¼ ¡¼ 2 0 ¼ 2 ¼ Ã Ground Truth TYR2 ¡¼ ¡¼ 2 0 ¼ 2 ¼ ¡¼ ¡¼ 2 0 ¼ 2 ¼ ASP3 ¡¼ ¡¼ 2 0 ¼ 2 ¼ ¡¼ ¡¼ 2 0 ¼ 2 ¼ PRO4 ¡¼ ¡¼ 2 0 ¼ 2 ¼ ¡¼ ¡¼ 2 0 ¼ 2 ¼ GLU5 ¡¼ ¡¼ 2 0 ¼ 2 ¼ ¡¼ ¡¼ 2 0 ¼ 2 ¼ THR6 ¡¼ ¡¼ 2 0 ¼ 2 ¼ ¡¼ ¡¼ 2 0 ¼ 2 ¼ GLY7 ¡¼ ¡¼ 2 0 ¼ 2 ¼ ¡¼ ¡¼ 2 0 ¼ 2 ¼ THR8 ¡¼ ¡¼ 2 0… view at source ↗
Figure 3
Figure 3. Figure 3: Ramachandran plots for Chignolin (top row: ground truth test set; bottom row: ARBG’s predictions) [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: TICA plots comparing the true MD distribution against predictions from ROBIN for the octapeptide: CGSWHKQR. Increasing resolution [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Left: Performance with increasing bin count for AL3. Right: Optimal temperature on AL3 across E-W2 and T-W2. Sampling Temperature. We perform inference-time ablations over the transformer’s sampling temperature to identify conditions that yield optimal generative perfor￾mance. As shown in Figure A.5, the energy distribution varies systematically with temperature: at low temperatures, probability mass conce… view at source ↗
read the original abstract

Efficient sampling of molecular systems at thermodynamic equilibrium is a hallmark challenge in statistical physics. This challenge has driven the development of Boltzmann Generators (BGs), which allow rapid generation of uncorrelated equilibrium samples by combining a generative model with exact likelihoods and an importance sampling correction. However, modern BGs predominantly rely on normalizing flows (NFs), which either suffer from limited expressivity due to strict invertibility constraints (discrete time) or computationally expensive likelihoods (continuous time). In this paper, we propose Autoregressive Boltzmann Generators (ArBG) -- a novel autoregressive modelling framework -- that overcomes these limitations by departing from the flow-based BG paradigm. ArBG circumvents the topological constraints of flows and enables sequential inference-time interventions, while offering enhanced scalability by leveraging architectures effective in Large Language Models. We empirically demonstrate that ArBG leads to significant improvements over flow-based models across all benchmarks, but particularly in larger peptide systems such as the 10-residue Chignolin. Furthermore, we introduce Robin, a 132 million parameter transferable model trained with the ArBG framework which improves over the previous state-of-the-art, reducing the zero-shot energy error, E-W$_2$, on 8-residue systems by over 60$\%$. The code can be found at the following link: https://github.com/danyalrehman/autobg.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper proposes Autoregressive Boltzmann Generators (ArBG), a framework replacing normalizing flows in Boltzmann Generators with autoregressive models for sampling equilibrium configurations of molecular systems. It claims that ArBG overcomes invertibility and computational limits of flows, enables sequential interventions, and yields significant empirical improvements over flow-based models on all benchmarks, especially larger peptides such as 10-residue Chignolin. It further introduces Robin, a 132-million-parameter transferable model trained under the ArBG framework that reduces zero-shot energy error E-W₂ by over 60% on 8-residue systems relative to prior state-of-the-art.

Significance. If the central claims hold, ArBG would represent a meaningful advance by demonstrating that autoregressive factorizations can supply tractable exact densities suitable for importance-sampling correction to the Boltzmann distribution, while scaling via LLM-style architectures to larger systems where flows are limited. The reported gains on Chignolin and the transferable Robin model would indicate practical utility for peptide sampling and transferable generative models in statistical physics.

major comments (2)
  1. [Abstract] Abstract: The claim that autoregressive proposals q(x) supply exact, order-independent likelihoods whose importance weights p(x)/q(x) recover unbiased samples from the target Boltzmann distribution is load-bearing for all performance assertions. The provided text gives no derivation or empirical test showing that a fixed coordinate ordering avoids long-range correlation artifacts that cannot be corrected by reweighting alone.
  2. [Abstract] Abstract: The reported improvements (including the >60% reduction in E-W₂ for Robin on 8-residue systems and gains on 10-residue Chignolin) are presented without reference to any independent reference sampler, effective sample size diagnostics, or explicit verification that the reweighted distribution matches the target beyond energy-based metrics; this leaves the central unbiased-sampling claim unsupported.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their comments, which highlight important aspects of the theoretical and empirical support for ArBG. We address each major comment below and indicate planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that autoregressive proposals q(x) supply exact, order-independent likelihoods whose importance weights p(x)/q(x) recover unbiased samples from the target Boltzmann distribution is load-bearing for all performance assertions. The provided text gives no derivation or empirical test showing that a fixed coordinate ordering avoids long-range correlation artifacts that cannot be corrected by reweighting alone.

    Authors: Autoregressive models define an exact joint density by the chain rule: q(x) = q(x_1) ∏_i q(x_i | x_<i). This factorization yields an exact likelihood for any fixed ordering, independent of the target Boltzmann density p. Importance sampling with weights w(x) = p(x)/q(x) is unbiased for expectations under p by the standard importance-sampling identity, regardless of ordering. Ordering influences the variance of the weights (i.e., sampling efficiency) but does not introduce bias that reweighting cannot correct. We will insert a short derivation of this fact in Section 2 and a brief discussion of ordering effects on long-range correlations. revision: yes

  2. Referee: [Abstract] Abstract: The reported improvements (including the >60% reduction in E-W_{2} for Robin on 8-residue systems and gains on 10-residue Chignolin) are presented without reference to any independent reference sampler, effective sample size diagnostics, or explicit verification that the reweighted distribution matches the target beyond energy-based metrics; this leaves the central unbiased-sampling claim unsupported.

    Authors: We agree that explicit diagnostics strengthen the empirical case. In the revised manuscript we will report effective sample sizes for the reweighted ensembles, include comparisons against independent reference samplers (e.g., long MD trajectories) on the smaller systems, and add a direct check that the reweighted marginals on selected observables align with the reference. The energy-based E-W_{2} metric remains central because it directly quantifies deviation from the Boltzmann distribution, but the additional diagnostics will be included. revision: yes

Circularity Check

0 steps flagged

No circularity: ArBG claims rest on empirical benchmarks and architectural departure from flows, not self-referential fits or derivations.

full rationale

The abstract and provided text introduce ArBG as an autoregressive alternative to normalizing flows for Boltzmann generators, citing empirical gains on peptides and a large Robin model. No equations, fitted parameters renamed as predictions, or self-citation chains appear in the given material. Performance claims (e.g., E-W2 reduction) are presented as measured outcomes on external benchmarks rather than quantities forced by construction from the model's own training data or prior self-references. The central premise—that autoregressive factorizations yield tractable exact densities correctable by importance sampling—is stated as a modeling choice with no reduction to tautology shown.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The framework implicitly assumes that autoregressive factorization plus importance sampling can target the Boltzmann measure, but no details are supplied.

pith-pipeline@v0.9.1-grok · 5777 in / 1073 out tokens · 27435 ms · 2026-06-26T04:48:54.604135+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

300 extracted references · 12 canonical work pages

  1. [1]

    2025 , journal=

    MD-LLM-1: A Large Language Model for Molecular Dynamics , author=. 2025 , journal=

    2025

  2. [2]

    2025 , eprint=

    FARMER: Flow AutoRegressive Transformer over Pixels , author=. 2025 , eprint=

    2025

  3. [3]

    The Continuous Language of Protein Structure , elocation-id =

    Billera, Lukas and Oresten, Anton and St. The Continuous Language of Protein Structure , elocation-id =. 2024 , doi =

    2024

  4. [4]

    2025 , eprint=

    NextStep-1: Toward Autoregressive Image Generation with Continuous Tokens at Scale , author=. 2025 , eprint=

    2025

  5. [5]

    2016 , booktitle=

    Conditional Image Generation with PixelCNN Decoders , author=. 2016 , booktitle=

    2016

  6. [6]

    2025 , booktitle=

    HollowFlow: Efficient Sample Likelihood Evaluation using Hollow Message Passing , author=. 2025 , booktitle=

    2025

  7. [7]

    2024 , booktitle=

    Autoregressive Image Generation without Vector Quantization , author=. 2024 , booktitle=

    2024

  8. [8]

    2026 , booktitle=

    FALCON: Few-step Accurate Likelihoods for Continuous Flows , author=. 2026 , booktitle=

    2026

  9. [9]

    2024 , booktitle=

    GIVT: Generative Infinite-Vocabulary Transformers , author=. 2024 , booktitle=

    2024

  10. [10]

    2026 , author =

    Generative molecular dynamics , journal =. 2026 , author =

    2026

  11. [11]

    Physical review , volume=

    Correlations in the motion of atoms in liquid argon , author=. Physical review , volume=. 1964 , publisher=

    1964

  12. [12]

    Studies in molecular dynamics. I. General method , author=. The Journal of Chemical Physics , volume=. 1959 , publisher=

    1959

  13. [13]

    arXiv preprint arXiv:2202.04164 , year=

    Enhanced sampling methods for molecular dynamics simulations , author=. arXiv preprint arXiv:2202.04164 , year=

    arXiv

  14. [14]

    2025 , eprint=

    Scalable Autoregressive 3D Molecule Generation , author=. 2025 , eprint=

    2025

  15. [15]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =

    Syed, Saifuddin and Bouchard-Côté, Alexandre and Deligiannidis, George and Doucet, Arnaud , title =. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =

  16. [16]

    2026 , booktitle=

    MarS-FM: Generative Modeling of Molecular Dynamics via Markov State Models , author=. 2026 , booktitle=

    2026

  17. [17]

    2024 , eprint=

    Force-Guided Bridge Matching for Full-Atom Time-Coarsened Dynamics of Peptides , author=. 2024 , eprint=

    2024

  18. [18]

    2024 , eprint=

    EquiJump: Protein Dynamics Simulation via SO(3)-Equivariant Stochastic Interpolants , author=. 2024 , eprint=

    2024

  19. [19]

    2024 , eprint=

    Scaling Diffusion Language Models via Adaptation from Autoregressive Models , author=. 2024 , eprint=

    2024

  20. [20]

    2024 , eprint=

    DPLM-2: A Multimodal Diffusion Protein Language Model , author=. 2024 , eprint=

    2024

  21. [21]

    2024 , eprint=

    Scaling up Masked Diffusion Models on Text , author=. 2024 , eprint=

    2024

  22. [22]

    Nature , pages=

    Accurate structure prediction of biomolecular interactions with AlphaFold 3 , author=. Nature , pages=. 2024 , publisher=

    2024

  23. [23]

    Algorithms for molecular biology , volume=

    ViennaRNA Package 2.0 , author=. Algorithms for molecular biology , volume=. 2011 , publisher=

    2011

  24. [24]

    Bioinformatics , volume=

    Forna (force-directed RNA): Simple and effective online RNA secondary structure diagrams , author=. Bioinformatics , volume=. 2015 , publisher=

    2015

  25. [25]

    Nucleic acids research , volume=

    RNAcentral 2021: secondary structure integration, improved sequence search and new member databases , author=. Nucleic acids research , volume=. 2021 , publisher=

    2021

  26. [26]

    Nature Methods , pages=

    Accurate RNA 3D structure prediction using a language model-based deep learning approach , author=. Nature Methods , pages=. 2024 , publisher=

    2024

  27. [27]

    , title =

    Chang, Huiwen and Zhang, Han and Jiang, Lu and Liu, Ce and Freeman, William T. , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2022 , pages =

    2022

  28. [28]

    arXiv preprint arXiv:2403.00043 , year=

    Rinalmo: General-purpose rna language models can generalize well on structure prediction tasks , author=. arXiv preprint arXiv:2403.00043 , year=

    arXiv

  29. [29]

    ArXiv , year=

    Generative Flows on Discrete State-Spaces: Enabling Multimodal Flows with Applications to Protein Co-Design , author=. ArXiv , year=

  30. [30]

    ArXiv , year=

    DPLM-2: A Multimodal Diffusion Protein Language Model , author=. ArXiv , year=

  31. [31]

    ArXiv , year=

    Diffusion Language Models Are Versatile Protein Learners , author=. ArXiv , year=

  32. [32]

    ArXiv , year=

    A Reparameterized Discrete Diffusion Model for Text Generation , author=. ArXiv , year=

  33. [33]

    ArXiv , year=

    Think While You Generate: Discrete Diffusion with Planned Denoising , author=. ArXiv , year=

  34. [34]

    arXiv preprint arXiv:2406.04329 , year=

    Simplified and Generalized Masked Diffusion for Discrete Data , author=. arXiv preprint arXiv:2406.04329 , year=

    arXiv

  35. [35]

    The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

    Simple and Effective Masked Diffusion Language Models , author=. The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

  36. [36]

    Sofroniew, Deniz Oktay, Zeming Lin, Robert Verkuil, Vincent Q

    Thomas Hayes and Roshan Rao and Halil Akin and Nicholas J. Sofroniew and Deniz Oktay and Zeming Lin and Robert Verkuil and Vincent Q. Tran and Jonathan Deaton and Marius Wiggert and Rohil Badkundri and Irhum Shafkat and Jun Gong and Alexander Derry and Raul S. Molina and Neil Thomas and Yousuf A. Khan and Chetan Mishra and Carolyn Kim and Liam J. Bartie a...

    work page doi:10.1126/science.ads0018 2025

  37. [37]

    Evolutionary-scale prediction of atomic-level protein structure with a language model , volume =

    Zeming Lin and Halil Akin and Roshan Rao and Brian Hie and Zhongkai Zhu and Wenting Lu and Nikita Smetanin and Robert Verkuil and Ori Kabeli and Yaniv Shmueli and Allan dos Santos Costa and Maryam Fazel-Zarandi and Tom Sercu and Salvatore Candido and Alexander Rives , title =. Science , volume =. 2023 , doi =. https://www.science.org/doi/pdf/10.1126/scien...

    work page doi:10.1126/science.ade2574 2023

  38. [38]

    ArXiv , year=

    The LAMBADA dataset: Word prediction requiring a broad discourse context , author=. ArXiv , year=

  39. [39]

    ArXiv , year=

    Masked Diffusion Models are Secretly Time-Agnostic Masked Models and Exploit Inaccurate Categorical Sampling , author=. ArXiv , year=

  40. [40]

    International Conference on Machine Learning , year=

    Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution , author=. International Conference on Machine Learning , year=

  41. [41]

    In: Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)

    Joshi, Mandar and Choi, Eunsol and Weld, Daniel and Zettlemoyer, Luke. T rivia QA : A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehension. Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2017. doi:10.18653/v1/P17-1147

    work page doi:10.18653/v1/p17-1147 2017

  42. [42]

    ArXiv , year=

    Training Verifiers to Solve Math Word Problems , author=. ArXiv , year=

  43. [43]

    ArXiv , year=

    A Corpus and Cloze Evaluation for Deeper Understanding of Commonsense Stories , author=. ArXiv , year=

  44. [44]

    ArXiv , year=

    Efficient Training of Language Models to Fill in the Middle , author=. ArXiv , year=

  45. [45]

    ROUGE : A Package for Automatic Evaluation of Summaries

    Lin, Chin-Yew. ROUGE : A Package for Automatic Evaluation of Summaries. Text Summarization Branches Out. 2004

    2004

  46. [46]

    ArXiv , year=

    Lessons from the Trenches on Reproducible Evaluation of Language Models , author=. ArXiv , year=

  47. [47]

    ArXiv , year=

    Likelihood-Based Diffusion Language Models , author=. ArXiv , year=

  48. [48]

    ArXiv , year=

    TinyLlama: An Open-Source Small Language Model , author=. ArXiv , year=

  49. [49]

    arXiv preprint arXiv:2311.07468 , year=

    Are we falling in a middle-intelligence trap? an analysis and mitigation of the reversal curse , author=. arXiv preprint arXiv:2311.07468 , year=

    arXiv

  50. [50]

    Proceedings of the 40th annual meeting of the Association for Computational Linguistics , pages=

    Bleu: a method for automatic evaluation of machine translation , author=. Proceedings of the 40th annual meeting of the Association for Computational Linguistics , pages=

  51. [51]

    a is b" fail to learn

    The reversal curse: Llms trained on" a is b" fail to learn" b is a" , author=. arXiv preprint arXiv:2309.12288 , year=

    arXiv

  52. [52]

    ArXiv , year=

    Llama 2: Open Foundation and Fine-Tuned Chat Models , author=. ArXiv , year=

  53. [53]

    2019 , journal=

    Language Models are Unsupervised Multitask Learners , author=. 2019 , journal=

    2019

  54. [54]

    Cell systems , year=

    ProGen2: Exploring the Boundaries of Protein Language Models , author=. Cell systems , year=

  55. [55]

    bioRxiv , year=

    Protein generation with evolutionary diffusion: sequence is all you need , author=. bioRxiv , year=

  56. [56]

    2024 , eprint=

    Informed Correctors for Discrete Diffusion Models , author=. 2024 , eprint=

    2024

  57. [57]

    2024 , eprint=

    Masked Diffusion Models are Secretly Time-Agnostic Masked Models and Exploit Inaccurate Categorical Sampling , author=. 2024 , eprint=

    2024

  58. [58]

    bioRxiv , year=

    Efficient evolution of human antibodies from general protein language models and sequence information alone , author=. bioRxiv , year=

  59. [59]

    ArXiv , year=

    ALBERT: A Lite BERT for Self-supervised Learning of Language Representations , author=. ArXiv , year=

  60. [60]

    ArXiv , year=

    RoBERTa: A Robustly Optimized BERT Pretraining Approach , author=. ArXiv , year=

  61. [61]

    North American Chapter of the Association for Computational Linguistics , year=

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding , author=. North American Chapter of the Association for Computational Linguistics , year=

  62. [62]

    2025 , eprint=

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning , author=. 2025 , eprint=

    2025

  63. [63]

    2024 , eprint=

    How Discrete and Continuous Diffusion Meet: Comprehensive Analysis of Discrete Diffusion Models via a Stochastic Integral Framework , author=. 2024 , eprint=

    2024

  64. [64]

    2013 , publisher =

    Limit theorems for stochastic processes , author =. 2013 , publisher =

    2013

  65. [65]

    ArXiv , year=

    Score-based Continuous-time Discrete Diffusion Models , author=. ArXiv , year=

  66. [66]

    2024 , eprint=

    Discrete Flow Matching , author=. 2024 , eprint=

    2024

  67. [67]

    Johnson and Jonathan Ho and Daniel Tarlow and Rianne van den Berg , title =

    Jacob Austin and Daniel D. Johnson and Jonathan Ho and Daniel Tarlow and Rianne van den Berg , title =. CoRR , volume =. 2021 , url =. 2107.03006 , timestamp =

    arXiv 2021

  68. [68]

    2022 , eprint=

    A Continuous Time Framework for Discrete Denoising Models , author=. 2022 , eprint=

    2022

  69. [69]

    George and Zhang, Qing , year =

    Yin, G. George and Zhang, Qing , year =. Continuous-. doi:10.1007/978-1-4614-4346-9 , keywords =

    work page doi:10.1007/978-1-4614-4346-9

  70. [70]

    2024 , eprint=

    Simple Guidance Mechanisms for Discrete Diffusion Models , author=. 2024 , eprint=

    2024

  71. [71]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =

    Benton, Joe and Shi, Yuyang and De Bortoli, Valentin and Deligiannidis, George and Doucet, Arnaud , title =. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume =. 2024 , month =. doi:10.1093/jrsssb/qkae005 , url =

    work page doi:10.1093/jrsssb/qkae005 2024

  72. [72]

    2024 , eprint=

    Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data , author=. 2024 , eprint=

    2024

  73. [73]

    Proceedings of the 31th International Conference on Machine Learning , year =

    Benigno Uria and Iain Murray and Hugo Larochelle , title =. Proceedings of the 31th International Conference on Machine Learning , year =

  74. [74]

    Gritsenko and Jasmijn Bastings and Ben Poole and Rianne van den Berg and Tim Salimans , title =

    Emiel Hoogeboom and Alexey A. Gritsenko and Jasmijn Bastings and Ben Poole and Rianne van den Berg and Tim Salimans , title =. 10th International Conference on Learning Representations , year =

  75. [75]

    Analysis and

    Budhiraja, Amarjit and Dupuis, Paul , year =. Analysis and. doi:10.1007/978-1-4939-9579-0 , keywords =

    work page doi:10.1007/978-1-4939-9579-0

  76. [76]

    2022 , eprint=

    Training and Inference on Any-Order Autoregressive Models the Right Way , author=. 2022 , eprint=

    2022

  77. [77]

    2021 , eprint=

    Discovering Non-monotonic Autoregressive Orderings with Variational Inference , author=. 2021 , eprint=

    2021

  78. [78]

    1976 , issn =

    A general method for numerically simulating the stochastic time evolution of coupled chemical reactions , journal =. 1976 , issn =. doi:https://doi.org/10.1016/0021-9991(76)90041-3 , url =

    work page doi:10.1016/0021-9991(76)90041-3 1976

  79. [79]

    , author=

    GFN2-xTB-An Accurate and Broadly Parametrized Self-Consistent Tight-Binding Quantum Chemical Method with Multipole Electrostatics and Density-Dependent Dispersion Contributions. , author=. Journal of chemical theory and computation , year=

  80. [80]

    The Journal of Physical Chemistry , author =

    Exact stochastic simulation of coupled chemical reactions , volume =. The Journal of Physical Chemistry , author =. 1977 , note =. doi:10.1021/j100540a008 , number =

    work page doi:10.1021/j100540a008 1977

Showing first 80 references.