Pith. sign in

REVIEW 4 major objections 5 minor 33 references

A frozen generic image model reading 2D molecular drawings beats quantum-descriptor baselines for reaction yield prediction, and cross-attention fusion improves it further.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-05 00:17 UTC pith:X4LKORWT

load-bearing objection Vision-over-quantum claim for yield prediction is plausible but under-validated; the single random split and missing baselines make the headline RMSE numbers provisional. the 4 major comments →

arxiv 2608.00776 v1 pith:X4LKORWT submitted 2026-08-01 cs.LG physics.chem-ph

Generic Vision and Cross-Attention for Reaction Yield Prediction

Qiwei Han , Chi Zhou This is my paper
classification cs.LG physics.chem-ph
keywords reaction yield predictioncross-attention2D molecular imagesquantum descriptorsResNet-18multimodal fusionC–N cross-couplingsteric effects
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims that a generic, ImageNet-pretrained vision network—never exposed to chemistry—can extract enough steric information from simple 2D skeletal drawings of reactants to outperform a baseline MLP built on expensive quantum-mechanical descriptors (5.60% vs. 6.79% RMSE). It then claims that fusing the visual topology with tabular physical-organic descriptors through a cross-attention mechanism, where the tabular features query the images, reaches 5.27% RMSE and outperforms both single-modality models and naive concatenation. The authors use permutation feature importance and attention-weight extraction to argue that the vision pathway is not a passive feature pool: it actively localizes steric bulk, especially around the aryl halide, while a residual skip connection protects non-spatial electronic descriptors from being attenuated by the attention bottleneck. If correct, the work offers a computationally lighter route to spatially aware yield prediction that does not require domain-specific chemical pretraining or explicit 3D structure.

Core claim

The central claim is that 2D reaction topology alone, processed by an off-the-shelf frozen ResNet-18, carries predictive signal for reaction yield that explicit DFT-derived 1D descriptors miss, and that the optimal way to combine the two modalities is descriptor-guided cross-attention rather than vector concatenation. The champion architecture reaches 5.27% RMSE and R2=0.961 on the held-out test set. Mechanistically, the paper shows that cross-attention shifts feature reliance away from molecular weight as a steric proxy and toward explicit volume, surface area, and ovality queries that the visual pathway grounds in the images, while raw attention scores reveal sparse, binary-like routing th

What carries the argument

The central object is a multi-head cross-attention fusion block in which the tabular quantum descriptors act as the Query and the four component-level visual embeddings from the frozen ResNet-18 act as Keys and Values, forcing the explicit physics to actively interrogate the 2D topology. A residual skip connection concatenates the raw tabular features past the attention block, protecting non-spatial electronic parameters from what the paper calls the Residual Bottleneck. The argument is carried by comparing this architecture against a mean-pooled concatenation baseline, a vision-only model, and a tabular-only MLP, and by probing attention softmax outputs and permutation feature importance.

Load-bearing premise

The reported test set comes from a random 70/30 split with a locked but unreported seed, and the paper does not verify that the same aryl halide, ligand, additive, or base never appears in both training and test; if that independence fails, the frozen vision model can partly recognize memorized drawings and its advantage over tabular descriptors is inflated.

What would settle it

Re-run the same four architectures with a grouped split that keeps every reaction of a given aryl halide (or ligand) entirely in either training or test; if the vision-only model's RMSE advantage over the tabular MLP shrinks or reverses under this split, the headline comparison is driven by molecule-level leakage rather than visual steric learning.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Reaction yield models that currently pay the computational cost of DFT descriptors could instead derive macroscopic steric information from cheap 2D drawings, reserving quantum descriptors for electronic effects.
  • Cross-attention fusion, by making tabular features query images, avoids the modality-dominance failure mode of simple concatenation, so multimodal chemical models should adopt guided fusion rather than naive feature stacking.
  • The attention maps give a per-component chemical hierarchy automatically learned from data, which could be used to identify which reactant's steric shape most controls a reaction family before any mechanistic assumptions are made.
  • The residual-skip design principle generalizes: any fusion that routes all tabular features through a spatial attention bottleneck will degrade purely non-spatial features, so bypassing the bottleneck is necessary for mixed steric/electronic descriptor sets.
  • A frozen generic backbone performing at this level challenges the necessity of chemistry-specific pretraining for yield prediction, suggesting that simpler visual encoders can serve as scalable feature extractors in synthesis optimization pipelines.

Where Pith is reading between the lines

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

  • If the split is grouped by reagent identity rather than random molecules, the vision advantage may shrink or vanish; the paper's split choice is testable and the reported seed is not disclosed, so the headline comparison should be re-run under a scaffold or grouped split before relying on it.
  • The learned attention hierarchy suggests a new diagnostic tool for high-throughput screening: one could use the per-component attention budget as a cheap proxy for which reactant's substitution pattern most destabilizes the transition state, potentially prioritizing experimental design before full DFT optimization.
  • The framework could be extended to reaction classes beyond C–N coupling by replacing the fixed four-component image sequence with an arbitrary reagent set and letting the cross-attention mechanism learn how many visual slots to attend to.
  • Because the paper shows volume estimates work better as directional queries than as scalar answers when vision is available, other noisy computed descriptors might similarly be repurposed as search keys in multimodal architectures rather than being discarded for their conformer sensitivity.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a dual-modal architecture for reaction yield prediction on the Doyle rxnpredict HTE dataset. A frozen ImageNet-pretrained ResNet-18 encodes four RDKit-drawn molecular components (additive, aryl halide, base, ligand) as separate 224x224 images, while an MLP encodes tabular DFT-derived descriptors. The two modalities are fused either by simple concatenation after mean pooling or by an 8-head cross-attention mechanism in which the tabular features act as queries and the visual features as keys/values, with a concatenated bypass of the raw tabular features. The authors report RMSE values of 6.80% (tabular MLP), 5.60% (vision-only), 5.78% (simple concat), and 5.27% (cross-attention, R2=0.961). They also present permutation feature importance and extracted attention weights to argue that cross-attention offloads steric identification to the visual pathway, learns a chemical hierarchy favoring the aryl halide, and requires a residual bypass to protect non-spatial electronic descriptors.

Significance. If the reported accuracy advantage is real, the finding is significant: a generic, frozen computer-vision backbone on simple 2D drawings could outperform an expensive DFT-descriptor tabular baseline, and cross-attention fusion could further improve accuracy while providing interpretable attention maps. The architecture is clearly specified, the equations are explicit, and the use of a public benchmark is a strength. However, the empirical claims currently rest on a single random split with no error bars, no reported seed, no grouped-split analysis, and no direct comparison to published baselines. These omissions are load-bearing for the central claim that generic vision and cross-attention 'significantly outperform' existing methods. The mechanistic analyses are also based on a single fitted model and are interpreted as if the attention weights and feature-importance rankings were stable, which is not established.

major comments (4)
  1. [Section 3.1, Fig. 2] The headline comparison (6.80 vs. 5.60 vs. 5.27 RMSE) is reported from a single 70/30 split with a locked but unreported seed and no error bars. The 0.33 percentage-point advantage of cross-attention over vision-only is within the range of seed-to-seed noise typically seen on HTE datasets. Please report the seed, run multiple seeds, and provide confidence intervals or paired significance tests. Without this, the claim of 'superior predictive accuracy' is not established.
  2. [Section 2.1, dataset partitioning] A random split is not adequate for this combinatorial HTE dataset. The same aryl halide, ligand, additive, and base are reused across many reactions, so identical or nearly identical RDKit drawings will appear in both training and test sets. For the vision modality, whose input is a deterministic image of the exact molecule, this allows template matching and can inflate the visual advantage over tabular descriptors. Please quantify train/test overlap (e.g., fraction of test reactions whose components appear in training), perform a grouped or scaffold split, and report performance on reactions with no component overlap.
  3. [Results, Sections 3.1 and 5] The paper claims that the proposed models 'significantly outperform' previous state-of-the-art methods, but no published baseline is evaluated on the same split. The original Random Forest benchmark [1] is mentioned but its RMSE is not reported, and no comparison is made to DRFP [7], GNN-based yield predictors, or other recent methods. The abstract and conclusion's claim of superiority over 'traditional methodologies' therefore lacks direct evidence. Please add baseline experiments with the same split and metrics.
  4. [Sections 3.2 and 3.3, Figs. 3 and 4] The PFI and attention-weight analyses are based on a single fitted model with no uncertainty quantification. Differences such as the +1.25 vs. +0.40 PFI penalties in Fig. 3C, or the attention means in Fig. 4C, could be within noise. The claim that molecular weight 'completely vanishes' and that the attention hierarchy is robust requires repeated runs or statistical testing. Moreover, interpreting attention weights as causal importance is descriptive; a perturbation test (e.g., occluding the high-attention aryl halide image) would be needed to support the mechanistic conclusions.
minor comments (5)
  1. [Section 2.1] Typos: 'T abular Descriptors' and '2D Visual T opologies' should read 'Tabular' and 'Topologies'.
  2. [Eq. (8)] The operation is described as a 'residual skip connection,' but Eq. (8) is a concatenation, not an additive residual. Consider calling it a 'bypass connection' to avoid confusion with ResNet-style additive residuals.
  3. [Eqs. (2), (3), and (8)] The notation X_tab is used for both the raw tabular descriptors and the dimensionally aligned tabular embedding. In Eq. (2) and (8), X_tab appears to be raw, while in Eq. (3) it is projected by W_Q,i. Clarify which quantity is used and specify its dimension so that the Q/K/V dimensions in Eq. (6) are unambiguous.
  4. [Section 3.2, last paragraph] The paragraph discussing behavior when explicit geometric descriptors are omitted is speculative and is not backed by an experiment. Either remove it or add an ablation study that drops volume/surface-area descriptors.
  5. [General] No code, seed value, or reproducible training setup is provided. Releasing these would strengthen reproducibility, especially because the reported results depend on a particular split and set of hyperparameters.

Circularity Check

1 steps flagged

No construction-level circularity; the only self-citation ([25]) is minor and not load-bearing, so the empirical accuracy claims stand on the paper's own held-out evaluation.

specific steps
  1. other [Section 1, Introduction, paragraph citing [23,24,25]; References item [25]]
    "To maximize this multimodal synergy, previous research has extensively explored cross-attention mechanisms, demonstrating them to be a highly promising direction for fusing disparate data types [23, 24, 25]."

    Reference [25] (Chemfusion, arXiv:2607.17033) is by the present paper's first author, so this is a self-citation. However, it is only used as background motivation for choosing cross-attention; the paper's central claims are established by its own held-out evaluation (Vision-Only 5.60% vs Tabular 6.80% RMSE; Cross-Attention 5.27% RMSE, Fig. 2) and by its own architecture equations (Eqs. 3-9) and ablations (Fig. 4A residual bottleneck: 6.17% vs 5.27%). No constants, equations, or uniqueness claims are imported from [25]. Accordingly, this self-citation is not load-bearing and does not force any result by construction; it is noted only for transparency (rubric score 2).

full rationale

The paper is an empirical ML study rather than a formal derivation, and I find no equation-level circularity. The prediction head (Eq. 9) maps the fused feature vector to a continuous yield y-hat, while the target yield is separate experimental data, so outputs are never defined in terms of inputs that already contain the answer. The tabular queries, visual keys/values, and attention weights (Eqs. 3-8) are architectural definitions; the empirical claims (frozen ResNet beats tabular MLP; cross-attention beats concat) are evaluated on a held-out 70/30 test split and are not the same quantities used to fit the models. The PFI and raw attention-weight analyses are post-hoc descriptions of the fitted model, not fitted parameters renamed as predictions, so they do not constitute circularity. The only flagged item is the self-citation [25] used as motivational context; it does not carry the derivational weight of the paper. The random-split/overlap concern in the reader's take is a data-leakage and statistical-validity threat, not a construction-level circularity step, since the paper never defines test predictions in terms of the training fit. Score 2 reflects one minor, non-load-bearing self-citation amid an otherwise self-contained central evaluation.

Axiom & Free-Parameter Ledger

5 free parameters · 6 axioms · 0 invented entities

The paper's central numbers depend on a handful of hand-chosen architectural constants (d_k, heads, seed) and several domain assumptions about image fidelity and split independence. No new physical entities are introduced. The Residual Bottleneck and chemical hierarchy are interpretations of ablations and attention weights, not independent entities.

free parameters (5)
  • Random split seed = not reported
    70/30 split and all stochastic training depend on a locked but unspecified seed; different seeds may change RMSE deltas.
  • Latent dimension d_k = 256
    Cross-attention embedding dimension chosen by hand (Eq. 3-7).
  • Attention heads h = 8
    Number of heads chosen by hand.
  • Training hyperparameters = lr=1e-3, wd=1e-4, batch=32, 500 epochs
    Optimization settings chosen without reported sensitivity analysis.
  • ImageNet normalization parameters = standard ImageNet mean/std
    Applied to molecular drawings; choice affects visual encodings.
axioms (6)
  • domain assumption rxnpredict tabular descriptors are accurate physical-organic parameters
    Section 2.1 uses DFT-derived HOMO/LUMO, dipole, and Sterimol values as ground-truth inputs.
  • domain assumption RDKit 2D skeletal images faithfully capture steric/topological features governing yield
    Section 2.1 generates 224x224 drawings; the whole vision claim assumes these images carry the relevant spatial signal.
  • domain assumption Random 70/30 split creates independent train/test molecules
    Section 2.1; no scaffold or grouping by substrate; likely violated for shared aryl halides and ligands.
  • domain assumption Frozen ImageNet features transfer to molecular line drawings
    Section 2.2 freezes ResNet-18; no chemical pretraining; visual results depend on transfer.
  • domain assumption Softmax attention weights reflect causal chemical importance
    Section 3.3 maps attention to chemical hierarchy; attention is a correlation, not necessarily causal.
  • standard math Standard linear algebra and softmax attention definitions
    Equations 1-8 rely on matrix products, softmax, and mean pooling.

pith-pipeline@v1.3.0-alltime-deepseek · 10248 in / 14158 out tokens · 131499 ms · 2026-08-05T00:17:32.272336+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Generic Vision and Cross-Attention for Reaction Yield Prediction." pith.science (2026). https://pith.science/paper/X4LKORWT

@misc{pith2026260800776,
  author       = {Pith},
  title        = {Pith review of: Generic Vision and Cross-Attention for Reaction Yield Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/X4LKORWT}},
  note         = {Machine review of arXiv:2608.00776}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Traditional reaction yield prediction is constrained by 1D quantum descriptors that lack explicit spatial information. To address this gap, a dual-modal Vision Cross-Attention architecture is proposed, fusing tabular physical-organic data with 2D molecular topologies. Notably, it is demonstrated that a generic computer vision backbone processing simple 2D skeletal structures independently outperforms purely quantum-based baselines. By synergizing both modalities, superior predictive accuracy compared to traditional methodologies is achieved by the optimal cross-attention framework (Test RMSE = 5.27%). Through mechanistic probing, active, descriptor-guided spatial querying is observed, effectively offloading macroscopic steric identification to the visual pathway. Furthermore, a dynamic chemical hierarchy is learned by the network to heavily prioritize critical steric bottlenecks, such as the aryl halide. Concurrently, residual skip connections are utilized to protect non-spatial electronic parameters from destructive attenuation during fusion. Collectively, a scalable and highly interpretable blueprint is provided for augmenting physical chemistry with deep visual learning.

Figures

Figures reproduced from arXiv: 2608.00776 by Chi Zhou, Qiwei Han.

Figure 1
Figure 1. Figure 1: The Dual-Modal Vision Cross-Attention Architecture. 2D spatial topologies extracted via a generic ResNet-18 backbone are actively queried by 1D physical-organic descrip￾tors to dynamically contextualize the macroscopic reaction space. A residual skip connection explicitly bypasses the spatial attention bottleneck to preserve strict quantum-chemical rigor prior to the final yield prediction. 1D physical-org… view at source ↗
Figure 2
Figure 2. Figure 2: Evaluating Architectural Fusion and Modality Synergy. Predictive perfor￾mance (RMSE and R2 ) across baseline and fused architectures. The dual-modal Cross-Attention framework achieves the lowest error, seamlessly synergizing spatial visual topology with explicit tabular quantum descriptors. 0 2 4 6 8 10 +RMSE Penalty (%) Additive O1 Electrostatic Charge Additive N1 Electrostatic Charge Aryl Halide C2 Nmr S… view at source ↗
Figure 3
Figure 3. Figure 3: Mechanistic shifts in feature reliance via Permutation Feature Importance (PFI). (A) Tabular Baseline: The MLP relies heavily on aryl halide molecular weight as a mathematical proxy for molecular size. (B) Simple Concat: Passive mean pooling blurs spatial details, forcing the network to take a computational shortcut, falling back on molecular weight and purely electronic descriptors. (C) Cross-Attention: T… view at source ↗
Figure 4
Figure 4. Figure 4: Architectural ablation and visual verification of dynamic spatial special￾ization. (A) The Residual Bottleneck: Forcing non-spatial electronic descriptors entirely through the spatial attention block severely compromises predictive accuracy (6.17% RMSE). Implementing a residual skip connection protects these quantum parameters from destructive attenuation, restoring superior predictive accuracy (5.27%). (B… view at source ↗

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

33 extracted references · 33 canonical work pages · 2 internal anchors

  1. [1]

    Predicting re- action performance in c–n cross-coupling using machine learning.Science, 360:186– 190, 2018

    Derek T Ahneman et al. Predicting re- action performance in c–n cross-coupling using machine learning.Science, 360:186– 190, 2018

    work page 2018

  2. [2]

    A platform for au- tomated nanomole-scale reaction screen- ing and micromole-scale synthesis in flow

    Damith Perera et al. A platform for au- tomated nanomole-scale reaction screen- ing and micromole-scale synthesis in flow. Science, 359:429–434, 2018

    work page 2018

  3. [3]

    Machine learning the ropes: principles, applica- 10 tions and directions in synthetic chem- istry.Chem

    Felix Strieth-Kalthoff et al. Machine learning the ropes: principles, applica- 10 tions and directions in synthetic chem- istry.Chem. Soc. Rev., 49:6154–6168, 2020

    work page 2020

  4. [4]

    Autonomous discovery in the chemical sciences part ii: Outlook.Angew

    Connor W Coley et al. Autonomous discovery in the chemical sciences part ii: Outlook.Angew. Chem. Int. Ed., 59:23414–23436, 2020

    work page 2020

  5. [5]

    Bayesian reac- tion optimization as a tool for chemical synthesis.Nature, 590:89–96, 2021

    Benjamin J Shields et al. Bayesian reac- tion optimization as a tool for chemical synthesis.Nature, 590:89–96, 2021

    work page 2021

  6. [6]

    Machine learning for chemical reactivity and reaction condi- tions.Curr

    J Schwallko et al. Machine learning for chemical reactivity and reaction condi- tions.Curr. Opin. Chem. Biol., 65:11–18, 2021

    work page 2021

  7. [7]

    Reaction classification and yield prediction using the differential reaction fingerprint drfp.Digit

    Daniel Probst et al. Reaction classification and yield prediction using the differential reaction fingerprint drfp.Digit. Discov., 1:91–97, 2022

    work page 2022

  8. [8]

    Prediction of higher-selectivity catalysts by computer- driven workflow and machine learning

    Andrew F Zahrt et al. Prediction of higher-selectivity catalysts by computer- driven workflow and machine learning. Science, 363:eaau5631, 2019

    work page 2019

  9. [9]

    Holistic predictive models of spatial and electronic require- ments for ligand-mediated selective catal- ysis.Nature, 571:343–348, 2019

    Jolene P Reid et al. Holistic predictive models of spatial and electronic require- ments for ligand-mediated selective catal- ysis.Nature, 571:343–348, 2019

    work page 2019

  10. [10]

    A structure-based platform for predicting chemical reactiv- ity.Chem, 6:1379–1390, 2020

    Frederik Sandfort et al. A structure-based platform for predicting chemical reactiv- ity.Chem, 6:1379–1390, 2020

    work page 2020

  11. [11]

    A comprehensive dis- covery platform for organophosphorus lig- ands for catalysis.J

    Tobias Gensch et al. A comprehensive dis- covery platform for organophosphorus lig- ands for catalysis.J. Am. Chem. Soc., 144:1205–1217, 2022

    work page 2022

  12. [12]

    Deep learning of activation energies.J

    Colin A Grambow et al. Deep learning of activation energies.J. Phys. Chem. Lett., 11:2992–2997, 2020

    work page 2020

  13. [13]

    Molecular graph convolutions: moving beyond fingerprints

    Steven Kearnes et al. Molecular graph convolutions: moving beyond fingerprints. J. Comput.-Aided Mol. Des., 30:595–608, 2016

    work page 2016

  14. [14]

    Moleculenet: a benchmark for molecular machine learn- ing.Chem

    Zhenqin Wu et al. Moleculenet: a benchmark for molecular machine learn- ing.Chem. Sci., 9:513–530, 2018

    work page 2018

  15. [15]

    Bigchem: Challenges and opportunities for big data analysis in chemistry.Mol

    Igor V Tetko et al. Bigchem: Challenges and opportunities for big data analysis in chemistry.Mol. Inform., 39:1900132, 2020

    work page 2020

  16. [16]

    Decimer: towards deep learning for chemical image recogni- tion.J

    Kohulan Rajan et al. Decimer: towards deep learning for chemical image recogni- tion.J. Cheminf., 13:1–9, 2021

    work page 2021

  17. [17]

    Chemception: A Deep Neural Network with Minimal Chemistry Knowledge Matches the Performance of Expert-developed QSAR/QSPR Models

    Garrett B Goh et al. Chemception: a deep neural network with minimal chemistry knowledge.arXiv preprint arXiv:1706.06689, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  18. [18]

    Deep learning in virtual screening: recent applications and developments.Int

    Thomas B Kimber et al. Deep learning in virtual screening: recent applications and developments.Int. J. Mol. Sci., 22:4435, 2021

    work page 2021

  19. [19]

    Drug discov- ery with explainable artificial intelligence

    Jos´ e Jim´ enez-Luna et al. Drug discov- ery with explainable artificial intelligence. Nat. Mach. Intell., 2:573–584, 2020

    work page 2020

  20. [20]

    A quantitative un- certainty metric controls error in neural network-driven chemical discovery.Chem

    Jon Paul Janet et al. A quantitative un- certainty metric controls error in neural network-driven chemical discovery.Chem. Sci., 10:7913–7922, 2019

    work page 2019

  21. [21]

    Comment on ”pre- dicting reaction performance in c–n cross- coupling using machine learning”.Sci- ence, 362:eaat8603, 2018

    Keiser V Chuang et al. Comment on ”pre- dicting reaction performance in c–n cross- coupling using machine learning”.Sci- ence, 362:eaat8603, 2018

    work page 2018

  22. [22]

    Multimodal deep learning for chemical applications.J

    Jian Guo et al. Multimodal deep learning for chemical applications.J. Chem. Inf. Model., 63:3561–3578, 2023

    work page 2023

  23. [23]

    Attention is all you need.Adv

    Ashish Vaswani et al. Attention is all you need.Adv. Neural Inf. Process. Syst., 30, 2017

    work page 2017

  24. [24]

    Perceiver: General perception with iterative attention

    Andrew Jaegle et al. Perceiver: General perception with iterative attention. In Int. Conf. Mach. Learn., pages 4651–4664, 2021

    work page 2021

  25. [25]

    ChemFusion: A Multimodal Cross-Attention Network for Reaction Yield Prediction

    Qiwei Han et al. Chemfusion: A mul- timodal cross-attention network for re- action yield prediction.arXiv preprint arXiv:2607.17033, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  26. [26]

    Rdkit: Open-source cheminformatics software, 2024

    Greg Landrum et al. Rdkit: Open-source cheminformatics software, 2024. 11

    work page 2024

  27. [27]

    Deep residual learn- ing for image recognition

    Kaiming He et al. Deep residual learn- ing for image recognition. InProc. IEEE Conf. Comput. Vis. Pattern Recognit., pages 770–778, 2016

    work page 2016

  28. [28]

    What makes train- ing multi-modal classification networks hard? InProc

    Weiyao Wang et al. What makes train- ing multi-modal classification networks hard? InProc. IEEE/CVF Conf. Com- put. Vis. Pattern Recognit., pages 12695– 12705, 2020

    work page 2020

  29. [29]

    Characterizing and over- coming the greedy nature of learning in multi-modal representations

    Nan Wu et al. Characterizing and over- coming the greedy nature of learning in multi-modal representations. InInt. Conf. Learn. Represent., 2022

    work page 2022

  30. [30]

    All models are wrong, but many are useful: Learning a variable’s importance.J

    Aaron Fisher et al. All models are wrong, but many are useful: Learning a variable’s importance.J. Mach. Learn. Res., 20:1– 81, 2019

    work page 2019

  31. [31]

    Freely available conformer generation methods: how good are they?J

    Jean-Paul Ebejer et al. Freely available conformer generation methods: how good are they?J. Chem. Inf. Model., 52:1146– 1158, 2012

    work page 2012

  32. [32]

    Geom, energy- annotated molecular conformations.Sci

    Simon Axelrod et al. Geom, energy- annotated molecular conformations.Sci. Data, 9:185, 2022

    work page 2022

  33. [33]

    End-to-end object detection with transformers

    Nicolas Carion et al. End-to-end object detection with transformers. InEur. Conf. Comput. Vis., pages 213–229, 2020. 12

    work page 2020