Pith. sign in

REVIEW 4 major objections 5 minor 43 references

GeoAda: Efficiently Finetune Geometric Diffusion Models with Equivariant Adapters

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read GeoAda shows that a frozen geometric diffusion model can be steered to new conditioning tasks by training only small SE(3)-equivariant adapter modules, which preserve the model's symmetries and its original task performance.

desk verdict A solid, well-empirically-supported equivariant adapter method for geometric diffusion models that deserves peer review, with a sketchy proof and presentation issues that are fixable. read the letter →

arxiv 2507.02085 v1 pith:ZI5A5VT3 submitted 2025-07-02 cs.LG cs.AI

Wanjia Zhao , Jiaqi Han , Siyi Gu , Mingjian Jiang , James Zou , Stefano Ermon This is my paper
classification cs.LGcs.AI
keywords geometricdiffusionmodelsSE(3)-equivariantadaptersparameter-efficientfine-tuningcontrolcatastrophicforgettingequivariantzero-convolutionmoleculegenerationtrajectoryprediction
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

GeoAda is a recipe for adapting a pretrained geometric diffusion model to downstream tasks with new conditioning signals while leaving the original network frozen. The paper argues that coupling operators can inject three kinds of geometric control — global type vectors, subgraphs such as protein pockets, and extra observed frames of a trajectory — into a trainable copy of selected denoiser layers, and that an equivariant zero-initialized convolution can merge the result back into the frozen score. Because only the lightweight adapter blocks are trained, the model keeps its geometric inductive bias, avoids overfitting on small downstream datasets, and does not forget the task it was pretrained on. The authors prove the composition stays $\mathrm{SE}(3)$-equivariant and show across particle dynamics, molecular dynamics, human motion, and molecule generation that the adapted model matches or beats full fine-tuning on the new task while retaining original-task accuracy.

What carries the argument

The carrying object is the equivariant adapter block, a four-stage module: a coupling operator $f$ that injects the control into the noised input without breaking its geometric structure, a trainable copy $\epsilon_{\theta'}$ of a subset of the frozen denoiser layers (picking the first layer of every $K$ consecutive layers proved best), a decoupling operator $g$ that selects the predicted components belonging to the original input, and an equivariant zero-convolution $z_\phi(\{x_i\},\{h_i\}) = (\phi_x (x_i - \bar{x}), \phi_h h_i)$ whose learnable weights start at zero. The zero initialization makes the adapter output vanish exactly at initialization, so the pretrained score is untouched at the start of fine-tuning while the gradients through $\phi_x$ and $\phi_h$ are nonzero, letting the control path learn smoothly; the center-of-mass subtraction $\bar{x}$ makes the output translation-invariant, and rotation equivariance is carried by the geometric feature transformations.

What would settle it

Fine-tune GeoAda on CHARGED PARTICLES with the conditioning length raised from 10 to 50 frames instead of 10 to 15: if the frozen backbone errors on the longer input, ignores the added frames, or shows no gain over the 10-frame baseline, then the claim that frame control adapts without architectural modification is falsified.

Watch

Extended reading notes

Core claim

The paper's central claim is that a pretrained geometric diffusion model can be redirected to new controlled-generation tasks without touching its weights: the frozen denoiser $\epsilon_\theta$ is supplemented by an equivariant adapter $s_{\theta',\phi}$ whose output is added to the original score at sampling time. The adapter encodes the control $C$ through a coupling operator $f$ — node-wise addition of a lifted global vector for global type control, graph union for subgraph control, and frame concatenation for frame control — processes it with a trainable copy $\epsilon_{\theta'}$ of selected pretrained layers, extracts the relevant nodes or frames with a decoupling operator $g$, and scales the result with an equivariant zero-initialized convolution $z_\phi$. Proposition 4.1 states that $g \circ \epsilon_\theta \circ f$ is $\mathrm{SE}(3)$-equivariant for every control type, so the adapted score inherits the invariant marginal of Eq. (1). On CHARGED PARTICLES, five MD17 molecules, five CMU Mocap actions, and QM9/GEOM-Drugs to CrossDocked2020 pocket-conditioned generation, GeoAda matches or outperforms full fine-tuning on the downstream metric while the pretraining-task metrics stay at their original values, where full fine-tuning, partial fine-tuning, and prompt baselines degrade or collapse.

Load-bearing premise

The frame-control setup assumes the frozen pretrained denoiser can accept a trajectory whose number of conditioned frames changes from 10 at pretraining to 15 at fine-tuning, since the coupling operator just concatenates the extra frames and the paper never shows how the backbone's time embeddings or attention layers handle the new input dimension.

Editorial extensions

If this is right

  • Adapting to a new control with GeoAda costs about half the parameters of full fine-tuning (roughly 2.7 MB tunable on the tested datasets), which makes transfer practical for small downstream datasets.
  • The pretraining task survives adaptation: GeoAda keeps pretrain metrics at their original values on CHARGED PARTICLES and MD17, while full fine-tuning, partial fine-tuning, and prompt baselines degrade or collapse.
  • Training is punctuated by a 'sudden convergence': loss and validation MSE drop abruptly around steps 4500–4700, with ADE and FDE falling by 68.3% and 73.4%, so early-stopping rules based on gradual improvement can be misleading.
  • The choice of which pretrained layers to copy matters: selecting the first layer of every $K$ consecutive layers outperforms choosing only early or only late layers under the same parameter budget, making layer selection itself a tunable design choice.

Reading between the lines

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

  • The adapter recipe should port to other equivariant generative paradigms such as score-based SDEs or equivariant flow matching, since the proof only relies on the frozen network being $\mathrm{SE}(3)$-equivariant; the paper does not claim this.
  • If implicit regularization is the mechanism, the gap between GeoAda and full fine-tuning should widen as downstream data shrink; a controlled data-amount sweep would test this directly.
  • The three handcrafted coupling operators are the least general component: controls with high-dimensional or structured semantics (for example, text embeddings) would likely need learned coupling, a direction the paper's limitation section leaves open.
  • Frame control is the fragile case: increasing the conditioning length from 10 to 15 frames changes the backbone's input dimension, and the paper never verifies that the frozen architecture truly handles variable-length inputs.
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 paper proposes GeoAda, a parameter-efficient adapter framework for fine-tuning frozen SE(3)-equivariant geometric diffusion models. Control signals of three types (global type, subgraph, frame) are injected through coupling operators, processed by trainable copies of selected pretrained layers, and projected back through decoupling operators followed by equivariant zero-initialized convolutions. The authors claim that this design preserves SE(3)-equivariance, avoids architectural modification of the pretrained backbone, mitigates overfitting and catastrophic forgetting, and achieves state-of-the-art downstream performance across particle dynamics (CHARGED PARTICLES), molecular dynamics (MD17), human motion prediction (CMU Mocap), and pocket-conditioned molecule generation (CrossDocked2020). A proof of equivariance is given in the appendix, and ablations study the number of adapter blocks, zero convolution, and trainable copies.

Significance. If the technical gaps identified below are resolved, GeoAda would be a useful contribution to geometric diffusion fine-tuning. The framework adapts the ControlNet recipe to SE(3)-equivariant diffusion models, and the zero-convolution device is a sensible way to protect the pretrained score at initialization. The empirical evaluation is broad, includes standard deviations, reports parameter counts, and contains ablations that support the importance of the zero-convolution and trainable-copy design choices. The paper also documents an interesting sudden-convergence phenomenon. However, the central 'plug-and-play, no architectural modification' claim is currently under-supported for two of the three control types, and the equivariance proof is more asserted than derived.

major comments (4)
  1. [§4.2, Eq. (5); §5.1.1] The frame-control coupling in Eq. (5) concatenates the noised trajectory x_i^[T] with the control frames x_i^[\tilde T] along the temporal axis, producing an input with T+\tilde T frames. Section 5.1.1 changes the number of conditioned frames from 10 at pretraining to 15 at finetuning (and from 0 to 10 in the unconditional setting). The paper never states whether the EGTN backbone and its copied layers are variable-length in the temporal dimension. If the first linear layer or temporal attention has weights tied to T, the frozen model cannot consume the concatenated input, and the claim of adaptation 'without modifying the original model architecture' fails for this control type. Please specify the backbone's length handling, or describe any projection or resizing applied to the copied layers; the equivariance proof in Appendix 7 does not address this domain-compatibility issue.
  2. [§5.3, Table 5] The QM9-pretrained GeoAda model is trained with atom types (H, C, N, O, F), while CrossDocked2020 is processed with eight atom types (H, C, N, O, F, P, S, Cl). The paper does not explain how a frozen denoiser whose node-feature vocabulary was learned over five types can consume eight-dimensional one-hot vectors. If a new embedding or projection is introduced at finetuning, that is an architectural modification that must be described and justified; if the backbone is type-agnostic, that property should be stated explicitly. As written, the subgraph-control experiments do not establish the plug-and-play claim for the QM9 checkpoint.
  3. [Appendix 7, Prop. 4.1] The proof of Proposition 4.1 is a sequence of assertions rather than a derivation. It does not define an SE(3) action on each control type (notably the global-type vector \tilde c, which is invariant under the group action), and it states without proof that the coupling operators 'respect the SE(3) structure.' A rigorous proof should verify f(h·G, h·C)=h·f(G,C) and g(h·G'')=h·g(G'') for each of the three controls, and should also specify how the equivariant zero-convolution in Eq. (6) behaves under translations in the centered representation used by the backbone.
  4. [§5.3, Tables 5–7] The abstract and §4.4 claim that GeoAda 'consistently matches or outperforms full fine-tuning baselines' and avoids catastrophic forgetting. For subgraph control, however, the comparison in Tables 5–7 is against task-specific methods trained from scratch on CrossDocked2020 (liGAN, GraphBP, AR, Pocket2Mol, TargetDiff); no Full FT, PARTIAL-k, prompt, or adapter ablation is reported for this control type. The subgraph experiment therefore supports competitiveness against specialized generators but not the paper's central fine-tuning claim, and should be supplemented or the claim narrowed.
minor comments (5)
  1. [Abstract and Section 1] There are several typos and grammar issues, including 'Frame Fontrol', 'freezed', and 'sythetic accessibility'; the manuscript should be proofread.
  2. [Table 1] Table 1 contains malformed entries such as '3924 3950' and isolated 'nan' cells; please reformat the table so that missing values and standard deviations are presented consistently.
  3. [Table 7] The ring-size row for size 7 reports '23.5' without a percent sign, unlike the other entries in the table.
  4. [Appendix 8.4.1] The notation for W_{t,s} and γ is confusing: W_{t,s} is said to be in R^N while γ is a vector over T, and the construction of the weights w(t,s) should be defined with explicit dimensions.
  5. [Section 5, Baselines] The PARTIAL-k baseline is cited as [10, 14, 43], but reference [43] appears to be a colorization paper rather than a partial fine-tuning method; please check the citation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: GeoAda's equivariance guarantee is a direct compositional proof and its empirical claims are benchmarked externally.

full rationale

GeoAda's central theoretical claim (Prop. 4.1) is that g ∘ ϵθ ∘ f inherits SE(3)-equivariance from an equivariant frozen denoiser; the appendix proof verifies the transformation behavior of each explicit coupling/decoupling operator (Eqs. 3–5) and of the zero-convolution (Eq. 6), so the conclusion follows by composition rather than being assumed as the conclusion. The adapter parameters are optimized with the standard denoising objective (Eq. 7) on external datasets (Charged Particles, MD17, CMU Mocap, CrossDocked2020), so the reported state-of-the-art numbers are measured against held-out benchmarks, not encoded by construction. The method does rely on the authors' own GeoTDM/EGTN backbone and cites several prior works from the same group, but those citations supply the pretrained model and its equivariance as an input assumption; they do not by themselves establish that GeoAda works, and no uniqueness claim is imported from them. The frame-control concern (Eq. 5 concatenating T + T̃ frames while pretraining used 10 and finetuning uses 15 frames) is a legitimate correctness/verification gap about backbone variable-length support, but it is not a circularity: the downstream error is still computed against ground-truth trajectories rather than derived from the adapter's own definition. No fitted parameter is renamed as a prediction, and no load-bearing premise reduces to a self-citation chain.

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

No new physical entities (particles, mediators, conserved quantities) are introduced. The adapter is a trainable module, not a postulated entity with independent evidence. The central claims rest on the stated equivariance assumptions and the empirical benchmarks.

free parameters (3)
  • Number of adapter blocks B = 3
    Set to 3 in all experiments (Section 5); ablation in Table 20 shows performance depends on B, so it is a hand-tuned hyperparameter affecting the reported results.
  • Layer-selection stride K = not specified
    Section 4.3 states choosing 'the first layer for every K consecutive layers' performs best, but K is never reported, so the exact adapter configuration is not reproducible.
  • Zero-convolution scalars phi_x, phi_h = initialized to 0, learned
    Eq. (6) defines these as learnable scalars; they are part of the trained model rather than ad hoc fits, but they are still parameters tuned on downstream data.
assumptions (4)
  • domain assumption The pretrained denoiser is SE(3)-equivariant
    Invoked in Proposition 4.1 and Appendix §7 to prove the adapter preserves equivariance; standard for geometric diffusion models.
  • ad hoc to paper The coupling and decoupling operators are SE(3)-equivariant for global type, subgraph, and frame controls
    Appendix §7 states this without a detailed proof; the main equivariance guarantee rests on it.
  • domain assumption The frozen backbone can process variable-length frame inputs
    Frame-control experiments change the number of conditioned frames from 10 to 15 (Section 5.1); the paper does not justify this for the backbone.
  • domain assumption A subset of the pretrained layers, when copied and stacked with zero convolution, remains equivariant and composable
    Section 4.3 uses trainable layer copies; this assumes each selected layer is equivariant and that the stack preserves the group action.

how reviews work

0 comments
Cite this review

Pith. "Pith review of GeoAda: Efficiently Finetune Geometric Diffusion Models with Equivariant Adapters." pith.science (2026). https://pith.science/paper/ZI5A5VT3

@misc{pith2026250702085,
  author       = {Pith},
  title        = {Pith review of: GeoAda: Efficiently Finetune Geometric Diffusion Models with Equivariant Adapters},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZI5A5VT3}},
  note         = {Machine review of arXiv:2507.02085}
}
read the original abstract

Geometric diffusion models have shown remarkable success in molecular dynamics and structure generation. However, efficiently fine-tuning them for downstream tasks with varying geometric controls remains underexplored. In this work, we propose an SE(3)-equivariant adapter framework ( GeoAda) that enables flexible and parameter-efficient fine-tuning for controlled generative tasks without modifying the original model architecture. GeoAda introduces a structured adapter design: control signals are first encoded through coupling operators, then processed by a trainable copy of selected pretrained model layers, and finally projected back via decoupling operators followed by an equivariant zero-initialized convolution. By fine-tuning only these lightweight adapter modules, GeoAda preserves the model's geometric consistency while mitigating overfitting and catastrophic forgetting. We theoretically prove that the proposed adapters maintain SE(3)-equivariance, ensuring that the geometric inductive biases of the pretrained diffusion model remain intact during adaptation. We demonstrate the wide applicability of GeoAda across diverse geometric control types, including frame control, global control, subgraph control, and a broad range of application domains such as particle dynamics, molecular dynamics, human motion prediction, and molecule generation. Empirical results show that GeoAda achieves state-of-the-art fine-tuning performance while preserving original task accuracy, whereas other baselines experience significant performance degradation due to overfitting and catastrophic forgetting.

Figures

Figures reproduced from arXiv: 2507.02085 by the authors.

Figure 1
Figure 1. Overall framework of GeoAda. The model integrates diverse control signals, including frame, global type, and subgraph controls through lightweight equivariant adapters inserted into the frozen pretrained denoiser. To this end, we propose a general and efficient framework(GeoAda) that enables the transfer of geometric diffusion models across diverse downstream tasks with minimal computational overhead. Inspired by th… view at source ↗
Figure 2
Figure 2. Overall Framework With the control encoding in § 4.2, a straight￾forward approach to leverage a pretrained diffu￾sion model on downstream tasks is to perform supervised finetuning (SFT). However, SFT usu￾ally induces suboptimal empirical performance, since 1. SFT is parameter-inefficient since each gradient update is conducted on all parameters of the pretrained model; 2. the full-parameter finetuning is prone to ov… view at source ↗
Figure 3
Figure 3. Visualization of MD17 Experimental setup. We employ the MD17 [3] dataset, which contains the DFT-simulated molecular dynamics (MD) trajectories of 8 small molecules, with the number of atoms for each molecule ranging from 9 (Ethanol and Malonaldehyde) to 21 (Aspirin). For each molecule, we construct a training set of 5000 trajectories, and 1000/1000 for validation and testing, uniformly sampled along the time dimens… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Visualization of Running trajectory Results We report short-term and long-term motion prediction results on the CMU Mocap dataset in Ta￾bles 3 and 4. More results on jumping and soccer sce￾narios are in App. 9.2. GeoAda consistently achieves state-of-the-art performanc…
Figure 5
Figure 5. Figure 5: The sudden convergence phenomenon Parameter efficiency analysis As shown in Appendix 9.3.1, we explore the impact of varying the number of equivariant zero layers. Increasing the number of trainable copy layers generally improves performance, but introduces more parame…
Figure 6
Figure 6. Figure 6: GeoAda [PITH_FULL_IMAGE:figures/full_fig_p019_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 38 canonical work pages

  1. [1]

    Geom, energy-annotated molecular confor- mations for property prediction and molecular generation

    Simon Axelrod and Rafael Gomez-Bombarelli. Geom, energy-annotated molecular confor- mations for property prediction and molecular generation. Scientific Data, 9(1):185, 2022. 8

    work page 2022

  2. [2]

    Sdfusion: Multimodal 3d shape completion, reconstruction, and generation, 2023

    Yen-Chi Cheng, Hsin-Ying Lee, Sergey Tulyakov, Alexander Schwing, and Liangyan Gui. Sdfusion: Multimodal 3d shape completion, reconstruction, and generation, 2023. 2

    work page 2023

  3. [3]

    Machine learning of accurate energy-conserving molecular force fields

    Stefan Chmiela, Alexandre Tkatchenko, Huziel E Sauceda, Igor Poltavsky, Kristof T Schütt, and Klaus-Robert Müller. Machine learning of accurate energy-conserving molecular force fields. Science advances, 3(5):e1603015, 2017. 1, 7

    work page 2017

  4. [4]

    Equijump: Protein dynamics simulation via so(3)-equivariant stochastic interpolants, 2024

    Allan dos Santos Costa, Ilan Mitnikov, Franco Pellegrini, Ameya Daigavane, Mario Geiger, Zhonglin Cao, Karsten Kreis, Tess Smidt, Emine Kucukbenli, and Joseph Jacobson. Equijump: Protein dynamics simulation via so(3)-equivariant stochastic interpolants, 2024. 3

    work page 2024

  5. [5]

    Universal prompt tun- ing for graph neural networks

    Taoran Fang, Yunchao Zhang, Yang Yang, Chunping Wang, and Lei Chen. Universal prompt tun- ing for graph neural networks. Advances in Neural Information Processing Systems, 36:52464– 52489, 2023. 3, 6, 7, 8, 16, 17, 18

    work page 2023

  6. [6]

    Three-dimensional convolutional neural networks and a cross- docked data set for structure-based drug design

    Paul G Francoeur, Tomohide Masuda, Jocelyn Sunseri, Andrew Jia, Richard B Iovanisci, Ian Snyder, and David R Koes. Three-dimensional convolutional neural networks and a cross- docked data set for structure-based drug design. Journal of chemical information and modeling, 60(9):4200–4215, 2020. 8

    work page 2020

  7. [7]

    3d equivariant diffusion for target-aware molecule generation and affinity prediction

    Jiaqi Guan, Wesley Wei Qian, Xingang Peng, Yufeng Su, Jian Peng, and Jianzhu Ma. 3d equivariant diffusion for target-aware molecule generation and affinity prediction. In The Eleventh International Conference on Learning Representations, 2023. 1, 4

    work page 2023

  8. [8]

    3d equivariant diffusion for target-aware molecule generation and affinity prediction.arXiv preprint arXiv:2303.03543, 2023

    Jiaqi Guan, Wesley Wei Qian, Xingang Peng, Yufeng Su, Jian Peng, and Jianzhu Ma. 3d equivariant diffusion for target-aware molecule generation and affinity prediction.arXiv preprint arXiv:2303.03543, 2023. 8, 9

    arXiv 2023

Show all 43 references
  1. [9]

    Geometric trajectory diffusion models

    Jiaqi Han, Minkai Xu, Aaron Lou, Haotian Ye, and Stefano Ermon. Geometric trajectory diffusion models. arXiv preprint arXiv:2410.13027, 2024. 1, 3

    2024 arXiv

  2. [10]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 16000–16009, 2022. 6, 7, 8, 16, 17, 18

    2022

  3. [11]

    Denoising diffusion probabilistic models

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models. Advances in neural information processing systems, 33:6840–6851, 2020. 3, 6, 13

    2020

  4. [12]

    Equivariant diffusion for molecule generation in 3d

    Emiel Hoogeboom, Vıctor Garcia Satorras, Clément Vignac, and Max Welling. Equivariant diffusion for molecule generation in 3d. In International conference on machine learning, pages 8867–8887. PMLR, 2022. 1, 3, 4

    2022

  5. [13]

    Physics-informed regularization for domain-agnostic dynamical system modeling

    Zijie Huang, Wanjia Zhao, Jingdong Gao, Ziniu Hu, Xiao Luo, Yadi Cao, Yuanzhou Chen, Yizhou Sun, and Wei Wang. Physics-informed regularization for domain-agnostic dynamical system modeling. arXiv preprint arXiv:2410.06366, 2024. 1

    arXiv

  6. [14]

    Visual prompt tuning

    Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. Visual prompt tuning. In European conference on computer vision, pages 709–727. Springer, 2022. 6

    2022

  7. [15]

    Neural relational inference for interacting systems

    Thomas Kipf, Ethan Fetaya, Kuan-Chieh Wang, Max Welling, and Richard Zemel. Neural relational inference for interacting systems. arXiv preprint arXiv:1802.04687, 2018. 1, 6

    2018 arXiv

  8. [16]

    Adaptergnn: Parameter-efficient fine-tuning improves generalization in gnns, 2023

    Shengrui Li, Xueting Han, and Jing Bai. Adaptergnn: Parameter-efficient fine-tuning improves generalization in gnns, 2023. 3

    2023

  9. [17]

    Generating 3d molecules for target protein binding

    Meng Liu, Youzhi Luo, Kanji Uchino, Koji Maruhashi, and Shuiwang Ji. Generating 3d molecules for target protein binding. In International Conference on Machine Learning, 2022. 9 10

    2022

  10. [18]

    Graphprompt: Unifying pre-training and downstream tasks for graph neural networks, 2023

    Zemin Liu, Xingtong Yu, Yuan Fang, and Xinming Zhang. Graphprompt: Unifying pre-training and downstream tasks for graph neural networks, 2023. 3

    2023

  11. [19]

    A 3d generative model for structure-based drug design

    Shitong Luo, Jiaqi Guan, Jianzhu Ma, and Jian Peng. A 3d generative model for structure-based drug design. Advances in Neural Information Processing Systems, 34, 2021. 9

    2021

  12. [20]

    Text-guided diffusion model for 3d molecule generation

    Yanchen Luo, Junfeng Fang, Sihang Li, Zhiyuan Liu, Jiancan Wu, An Zhang, Wenjie Du, and Xiang Wang. Text-guided diffusion model for 3d molecule generation. arXiv preprint arXiv:2410.03803, 2024. 4

    2024 arXiv

  13. [21]

    Geometry-complete diffusion for 3d molecule generation and optimization, 2024

    Alex Morehead and Jianlin Cheng. Geometry-complete diffusion for 3d molecule generation and optimization, 2024. 1, 2

    2024

  14. [22]

    Pocket2mol: Efficient molecular sampling based on 3d protein pockets

    Xingang Peng, Shitong Luo, Jiaqi Guan, Qi Xie, Jian Peng, and Jianzhu Ma. Pocket2mol: Efficient molecular sampling based on 3d protein pockets. arXiv preprint arXiv:2205.07249,

    arXiv

  15. [23]

    Generating 3D molecules condi- tional on receptor binding sites with deep generative models

    Matthew Ragoza, Tomohide Masuda, and David Ryan Koes. Generating 3D molecules condi- tional on receptor binding sites with deep generative models. Chem Sci, 13:2701–2713, Feb

  16. [24]

    Quantum chemistry structures and properties of 134 kilo molecules

    Raghunathan Ramakrishnan, Pavlo O Dral, Matthias Rupp, and O Anatole V on Lilienfeld. Quantum chemistry structures and properties of 134 kilo molecules. Scientific data, 1(1):1–7,

  17. [25]

    Diffusion-edfs: Bi-equivariant denoising generative modeling on se(3) for visual robotic manipulation, 2023

    Hyunwoo Ryu, Jiwoo Kim, Hyunseok An, Junwoo Chang, Joohwan Seo, Taehan Kim, Yubin Kim, Chaewon Hwang, Jongeun Choi, and Roberto Horowitz. Diffusion-edfs: Bi-equivariant denoising generative modeling on se(3) for visual robotic manipulation, 2023. 1, 2

    2023

  18. [26]

    E(n) equivariant graph neural networks

    Victor Garcia Satorras, Emiel Hoogeboom, and Max Welling. E(n) equivariant graph neural networks. arXiv preprint arXiv:2102.09844, 2021. 1, 6

    2021 arXiv

  19. [27]

    Equivariant message passing for the prediction of tensorial properties and molecular spectra

    Kristof Schütt, Oliver Unke, and Michael Gastegger. Equivariant message passing for the prediction of tensorial properties and molecular spectra. In International Conference on Machine Learning, pages 9377–9388. PMLR, 2021. 7

    2021

  20. [28]

    Learning gradient fields for molecular conformation generation

    Chence Shi, Shitong Luo, Minkai Xu, and Jian Tang. Learning gradient fields for molecular conformation generation. In International conference on machine learning, pages 9558–9568. PMLR, 2021. 7

    2021

  21. [29]

    Generative modeling by estimating gradients of the data distribution

    Yang Song and Stefano Ermon. Generative modeling by estimating gradients of the data distribution. Advances in neural information processing systems, 32, 2019. 3

    2019

  22. [30]

    Score-based generative modeling through stochastic differential equations

    Yang Song, Jascha Sohl-Dickstein, Diederik P Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. Score-based generative modeling through stochastic differential equations. In International Conference on Learning Representations, 2021. 3

    2021

  23. [31]

    A bound for the error in the normal approximation to the distribution of a sum of dependent random variables

    Charles Stein. A bound for the error in the normal approximation to the distribution of a sum of dependent random variables. In Proceedings of the sixth Berkeley symposium on mathematical statistics and probability, volume 2: Probability theory, volume 6, pages 583–603. Univer...

    1972

  24. [32]

    Gppt: Graph pre-training and prompt tuning to generalize graph neural networks

    Mingchen Sun, Kaixiong Zhou, Xin He, Ying Wang, and Xin Wang. Gppt: Graph pre-training and prompt tuning to generalize graph neural networks. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD ’22, page 1717–1727, New York, NY , USA, 2...

    2022

  25. [33]

    Fine-tuning of continuous-time diffusion models as entropy-regularized control, 2024

    Masatoshi Uehara, Yulai Zhao, Kevin Black, Ehsan Hajiramezanali, Gabriele Scalia, Nathaniel Lee Diamant, Alex M Tseng, Tommaso Biancalani, and Sergey Levine. Fine-tuning of continuous-time diffusion models as entropy-regularized control, 2024. 3

    2024

  26. [34]

    Se(3)-diffusionfields: Learning smooth cost functions for joint grasp and motion optimization through diffusion, 2023

    Julen Urain, Niklas Funk, Jan Peters, and Georgia Chalvatzaki. Se(3)-diffusionfields: Learning smooth cost functions for joint grasp and motion optimization through diffusion, 2023. 1, 2 11

    2023

  27. [35]

    Diffusion model alignment using direct preference optimization, 2023

    Bram Wallace, Meihua Dang, Rafael Rafailov, Linqi Zhou, Aaron Lou, Senthil Purushwalkam, Stefano Ermon, Caiming Xiong, Shafiq Joty, and Nikhil Naik. Diffusion model alignment using direct preference optimization, 2023. 3

    2023

  28. [36]

    In-context learning unlocked for diffusion models

    Zhendong Wang, Yifan Jiang, Yadong Lu, yelong shen, Pengcheng He, Weizhu Chen, Zhangyang Wang, and Mingyuan Zhou. In-context learning unlocked for diffusion models. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. 1, 3

    2023

  29. [37]

    Geometric- facilitated denoising diffusion model for 3d molecule generation

    Can Xu, Haosen Wang, Weigang Wang, Pengfei Zheng, and Hongyang Chen. Geometric- facilitated denoising diffusion model for 3d molecule generation. In Proceedings of the AAAI Conference on Artificial Intelligence, 2024. 1, 2

    2024

  30. [38]

    Eqmotion: Equivariant multi-agent motion prediction with invariant interaction reasoning

    Chenxin Xu, Robby T Tan, Yuhong Tan, Siheng Chen, Yu Guang Wang, Xinchao Wang, and Yanfeng Wang. Eqmotion: Equivariant multi-agent motion prediction with invariant interaction reasoning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, page...

    2023

  31. [39]

    Geometric latent diffusion models for 3d molecule generation

    Minkai Xu, Alexander Powers, Ron Dror, Stefano Ermon, and Jure Leskovec. Geometric latent diffusion models for 3d molecule generation. In International Conference on Machine Learning. PMLR, 2023. 1, 2

    2023

  32. [40]

    Geodiff: A geometric diffusion model for molecular conformation generation

    Minkai Xu, Lantao Yu, Yang Song, Chence Shi, Stefano Ermon, and Jian Tang. Geodiff: A geometric diffusion model for molecular conformation generation. In International Conference on Learning Representations, 2022. 1, 2, 3

    2022

  33. [41]

    Socialvae: Human trajectory prediction using timewise latents

    Pei Xu, Jean-Bernard Hayet, and Ioannis Karamouzas. Socialvae: Human trajectory prediction using timewise latents. In European Conference on Computer Vision, pages 511–528. Springer,

  34. [42]

    Adding conditional control to text-to-image diffusion models

    Lvmin Zhang, Anyi Rao, and Maneesh Agrawala. Adding conditional control to text-to-image diffusion models. In Proceedings of the IEEE/CVF international conference on computer vision, pages 3836–3847, 2023. 1, 2, 3, 5, 9

    2023

  35. [43]

    Colorful image colorization

    Richard Zhang, Phillip Isola, and Alexei A Efros. Colorful image colorization. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part III 14, pages 649–666. Springer, 2016. 5, 6 12 Appendix 7 Proof Below is th...

    2016

Pith tools

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