Spotlight Others

Reverse Transition Kernel (RTK) framework accelerates diffusion inference by enabling balanced subproblem decomposition, achieving superior convergence rates with RTK-MALA and RTK-ULD algorithms.

L1-norm based 3D convolution achieves competitive performance with lower energy consumption and latency compared to traditional methods, as proven through universal approximation theorem and experimen…

ResAD, a novel framework, tackles class-generalizable anomaly detection by learning residual feature distributions, achieving remarkable results on diverse datasets without retraining.

Deep learning models for neural activity lack reproducibility; this paper introduces adaptive regularization and iterative feature pruning to improve embedding consistency and predictive performance.

Bayesian neural networks often underfit due to their lack of reparameterization invariance; this paper introduces a Riemannian diffusion process to improve posterior sampling and enhance predictive pe…

Recursive PAC-Bayes: A frequentist method enabling sequential prior updates without information loss, resulting in significantly tighter generalization bounds.

CordsNet: a hybrid CNN-RNN architecture enabling biologically realistic, robust image recognition through continuous-time recurrent dynamics.

Reconstruct & Match (REMA) enhances deep learning’s out-of-distribution robustness by leveraging object’s topological homogeneity, outperforming state-of-the-art methods.

CORUN-Colabator: a novel cooperative unfolding network and coherence-based label generator achieves state-of-the-art real-world image dehazing by effectively integrating physical knowledge and generat…

QKFormer: A groundbreaking spiking transformer achieving 85.65% ImageNet accuracy using a linear-complexity, energy-efficient Q-K attention mechanism.

CutMix and Cutout data augmentation methods provably improve feature learning by enabling the network to learn rarer features and noise vectors more effectively.

PeskaVLP: Hierarchical knowledge augmentation boosts surgical video-language pretraining!

Spherical DYffusion: a novel AI model generates accurate, physically consistent global climate ensemble simulations, surpassing existing methods in efficiency and skill.

Graph-EFM: a novel probabilistic weather forecasting model using hierarchical graph neural networks that efficiently generates large ensembles for improved accuracy and uncertainty quantification.

Pre-trained text-to-image diffusion models create highly effective, versatile representations for embodied AI control, surpassing previous methods.

Poisson Variational Autoencoder (P-VAE) improves deep learning by encoding inputs as discrete spike counts, enhancing biological realism and interpretability while avoiding posterior collapse and achi…

Single image to physically compatible 3D objects: A new framework ensures 3D models maintain stability and mirror real-world equilibrium states, advancing realism in dynamic simulations and 3D printi…

Peri-midFormer uses a novel periodic pyramid transformer to effectively model complex periodic variations in time series, achieving state-of-the-art results in forecasting, imputation, classification,…

PCP-MAE enhances point cloud self-supervised learning by cleverly predicting masked patch centers, leading to superior 3D object classification and scene segmentation.

Particle Variational Inference (PVI) revolutionizes semi-implicit variational inference by directly optimizing the ELBO using a novel particle approximation, improving efficiency and expressiveness ov…