# Cross-Attention

8 related articles

Subscribe to updates

arXiv Nov 12, 2026
AFM-IR hyperspectral imaging with just 5% of spectra: CACNP rebuilds the full chemical cube in under 0.1 s with per-voxel uncertainty
Nano-IR hyperspectral imaging collects a spectrum point by point at every pixel, so a complete cube can take hours. CACNP feeds sparsely sampled spectra together with a few single-wavenumber structure images into a convolutional-attentive conditional neural process trained on synthetic cubes, returning the full cube and a per-voxel uncertainty. On experimental data of polystyrene and poly(methyl methacrylate) particles in epoxy, 5% of the spectra suffices, inference takes under 0.1 s, and reconstruction error approaches the sample noise level.
Nano-IR hyperspectral imaging collects a spectrum point by point at every pixel, so a complete cube can take hours. CACNP feeds sparsely sampled spectra together with a few single-wavenumber structure images into a convolutional-attentive conditional neural process trained on synthetic cubes, returning the full cube and a per-voxel uncertainty. On experimental data of polystyrene and poly(methyl methacrylate) particles in epoxy, 5% of the spectra suffices, inference takes under 0.1 s, and reconstruction error approaches the sample noise level.
William F. Kent（First author）
Convolutional-Attentive Conditional Neural Process AFM-IR Hyperspectral Cube Reconstruction Uncertainty Calibration Cross-Attention Adaptive Sampling
bioRxiv Nov 3, 2026
Cellular perturbation prediction drops explicit variable values: RELAY reads shared-variable relations and reaches 0.495 PDCorr on PBMC-3D
Cellular responses to perturbation depend on several experimental variables at once, including perturbation identity, cell type, donor and dose, so each measured population occupies one coordinate of a multidimensional condition space. Earlier methods conditioned on explicit representations of those variable values and struggled when a value was absent from training. RELAY instead infers a target from measured control–perturbed pairs that share some variables with it, receiving only which variables are shared and never their values; across chemical, genetic and cytokine perturbation data it beats STACK, scGen and TabPFN on unseen variable values and supervised methods on unseen combinations, reaching 0.495 PDCorr on PBMC-3D against 0.225 for the strongest supervised baseline, STATE.
Cellular responses to perturbation depend on several experimental variables at once, including perturbation identity, cell type, donor and dose, so each measured population occupies one coordinate of a multidimensional condition space. Earlier methods conditioned on explicit representations of those variable values and struggled when a value was absent from training. RELAY instead infers a target from measured control–perturbed pairs that share some variables with it, receiving only which variables are shared and never their values; across chemical, genetic and cytokine perturbation data it beats STACK, scGen and TabPFN on unseen variable values and supervised methods on unseen combinations, reaching 0.495 PDCorr on PBMC-3D against 0.225 for the strongest supervised baseline, STATE.
Shanghai Jiao Tong University
Yuze Sun（First author）Xiaoman Zhang（Corresponding author）
RELAY Perturbation-Response Prediction In-Context Learning Cross-Attention
researchsquare.com Oct 19, 2026
One freehand sketch, one 3D mesh: keeping local stroke tokens lifts average IoU from 0.616 to 0.628
Reconstructing a 3D mesh from a single freehand sketch usually compresses sketch features into one global vector, which loses fine strokes and thin structures. Locality-Preserving Sketch Token Fusion instead keeps both a global shape representation and spatially arranged local tokens, and lets the global representation query those tokens through cross-attention during decoding. On ShapeNet-Synthetic with ground-truth viewpoints, average IoU rises from 0.616 for SS2M to 0.628; on the human-drawn ShapeNet-Sketch, average IoU is 0.498 and 0.487 under ground-truth and predicted viewpoints respectively.
Reconstructing a 3D mesh from a single freehand sketch usually compresses sketch features into one global vector, which loses fine strokes and thin structures. Locality-Preserving Sketch Token Fusion instead keeps both a global shape representation and spatially arranged local tokens, and lets the global representation query those tokens through cross-attention during decoding. On ShapeNet-Synthetic with ground-truth viewpoints, average IoU rises from 0.616 for SS2M to 0.628; on the human-drawn ShapeNet-Sketch, average IoU is 0.498 and 0.487 under ground-truth and predicted viewpoints respectively.
Guangdong University of Science and Technology Macau University of Science and Technology
Qinchuan Lei（First author）
Locality-Preserving Sketch Token Fusion Sketch-based 3D Reconstruction Cross-Attention View Disentanglement
Scientific Reports Oct 20, 2026
Rat sleep staging holds up when the drug changes: trained on saline alone, F1 falls less than 1% under psilocybin and MDL100907
On more than 1.1 million annotated rat EEG/EMG segments, a robustness-oriented deep learning pipeline reaches 94.3% F1 for four-class sleep staging, 7.3 and 6.4 points above SlumberNet's 87.0% and random forest's 87.9%. Most automated stagers were evaluated on random splits that preserve the training distribution, and degrade when subjects or drug conditions shift. The pipeline combines a global context vector, supervised contrastive learning and a dual-branch classifier to learn representations insensitive to subject and treatment; trained on saline recordings only and tested on unseen animals, every metric drops by less than 1%.
On more than 1.1 million annotated rat EEG/EMG segments, a robustness-oriented deep learning pipeline reaches 94.3% F1 for four-class sleep staging, 7.3 and 6.4 points above SlumberNet's 87.0% and random forest's 87.9%. Most automated stagers were evaluated on random splits that preserve the training distribution, and degrade when subjects or drug conditions shift. The pipeline combines a global context vector, supervised contrastive learning and a dual-branch classifier to learn representations insensitive to subject and treatment; trained on saline recordings only and tested on unseen animals, every metric drops by less than 1%.
Charles University
Vojtěch Brejtr（First author）Lenka Lhotská（Corresponding author）
Sleep Staging Supervised Contrastive Learning Cross-Attention Domain Adaptation
researchsquare.com Oct 26, 2026
Enzyme kinetics prediction learns pairwise differences: kcat R² of 0.71 under cold start, zero-shot PCC of 0.40 on external adenylate kinase data
Enzyme kinetics prediction has relied on absolute regression over heterogeneous data, where the same enzyme-substrate system can differ by orders of magnitude across studies and models drift toward the mean. DeltaKcat recasts the task as pairwise relative learning: within one study it fixes either the enzyme or the substrate and a siamese network predicts the kinetic difference between two enzyme-substrate complexes, after which an experimentally characterized reference complex restores absolute values. Under a cold-start split, absolute kcat prediction reaches R² 0.71 and PCC 0.85, against R² 0.14 to 0.48 for DLKcat, CataPro and UniKP; on external adenylate kinase data excluded from training, zero-shot PCC is 0.40.
Enzyme kinetics prediction has relied on absolute regression over heterogeneous data, where the same enzyme-substrate system can differ by orders of magnitude across studies and models drift toward the mean. DeltaKcat recasts the task as pairwise relative learning: within one study it fixes either the enzyme or the substrate and a siamese network predicts the kinetic difference between two enzyme-substrate complexes, after which an experimentally characterized reference complex restores absolute values. Under a cold-start split, absolute kcat prediction reaches R² 0.71 and PCC 0.85, against R² 0.14 to 0.48 for DLKcat, CataPro and UniKP; on external adenylate kinase data excluded from training, zero-shot PCC is 0.40.
Chinese Academy of Sciences Tsinghua University Shenzhen Institutes of Advanced Technology Boston University
Feiran Li（First author, Corresponding author）
DeltaKcat Pairwise relative learning Enzyme Kinetics Prediction Cross-Attention
arXiv Oct 12, 2026
Precipitation generation skips retraining the weather model: one diffusion generator on frozen backbones cuts day-7 Brier score from 0.245 to 0.218
A separately trained diffusion generator maps non-precipitation atmospheric states from a frozen weather model to observation-aligned 24-hour precipitation, improving daily accumulation and event detection on Pangu-Weather, GraphCast and HRES without backbone-specific retraining.
Guowen Li（First author）
Dynamics–Precipitation Decoupling Conditional Diffusion Model Precipitation Forecasting Cross-Attention
arXiv Sep 28, 2026
Why Video Diffusion Models Break Physics: Researchers Trace It to RoPE Spatial Anchoring and Fix It by Rescaling RoPE Frequency
The work presents an interpretability study of the "motion planning" process in text-to-video diffusion models, finding that excessive spatial attention decay induced by RoPE in self-attention makes early candidate regions lock prematurely onto physically implausible positions, and proposes a lightweight architectural change that merely rescales RoPE frequency across denoising steps, improving physical commonsense in both training-free and training-based experiments.
The work presents an interpretability study of the "motion planning" process in text-to-video diffusion models, finding that excessive spatial attention decay induced by RoPE in self-attention makes early candidate regions lock prematurely onto physically implausible positions, and proposes a lightweight architectural change that merely rescales RoPE frequency across denoising steps, improving physical commonsense in both training-free and training-based experiments.
Yueyan Li（First author）
Rotary Position Embedding Diffusion Models Self-Attention Cross-Attention Attribution Patching
arXiv Sep 24, 2026
AGGRNet splits medical image features into informative and non-informative via a learnable threshold, reaching 5.48% higher accuracy than HiFuse-Small on Kvasir
The work proposes AGGRNet, which embeds a Feature Extraction and Aggregation (FEA) module and a C2PCA block into a YOLOv11 classification backbone; spatial and channel attention plus a learnable threshold τ split feature maps into informative and non-informative parts that are then aggregated by cross-attention, yielding results above the compared SOTA models on five public datasets, including a 5.48% accuracy gain on Kvasir and a 2.2% gain on LIMUC.
The work proposes AGGRNet, which embeds a Feature Extraction and Aggregation (FEA) module and a C2PCA block into a YOLOv11 classification backbone; spatial and channel attention plus a learnable threshold τ split feature maps into informative and non-informative parts that are then aggregated by cross-attention, yielding results above the compared SOTA models on five public datasets, including a 5.48% accuracy gain on Kvasir and a 2.2% gain on LIMUC.
Indian Institute of Technology Kanpur Carnegie Mellon University
Ansh Makwe（First author）Priyanka Bagade（Corresponding author）
AGGRNet Feature Extraction and Aggregation Module C2PCA Block Medical Image Classification Cross-Attention

You’re all caught up