Spotlight Others

EMR-MERGING: A tuning-free model merging technique achieves high performance by electing a unified model and generating lightweight task-specific modulators, eliminating the need for additional data …

EigenVI: a novel score-based variational inference method using orthogonal function expansions, offers closed-form solutions by solving eigenvalue problems, outperforming existing Gaussian BBVI method…

4DGF, a novel neural scene representation, achieves interactive-speed novel view synthesis for large-scale dynamic urban areas by efficiently combining 3D Gaussians and neural fields.

Double-Ended Synthesis Planning (DESP) significantly boosts computer-aided synthesis planning by using a bidirectional search, outperforming existing methods on multiple benchmarks, especially when sp…

Researchers developed a sample-efficient variational approach for transition path sampling using Doob’s h-transform, significantly reducing computational costs while accurately capturing transition pa…

Run-Length Tokenization (RLT) dramatically speeds up video transformer training and inference by efficiently removing redundant video tokens, matching baseline model performance with significant time …

Boosting dataset distillation, a new method, Diversity-Driven Synthesis, uses directed weight adjustment to create diverse, representative synthetic datasets, improving model performance while reducin…

DRAGON: A novel GNN framework using distributed-order fractional calculus surpasses traditional methods by capturing complex graph dynamics with enhanced flexibility and performance.

Vision transformers’ self-attention mechanism is dissected revealing how early layers focus on similar features for perceptual grouping while later layers integrate dissimilar features for contextuali…

New Cell-Type Dynamical Systems (CTDS) model disentangles neural population dynamics by incorporating distinct cell types, improving prediction accuracy and biological interpretability.

Adaptive likelihood estimation and MAP inference during reverse diffusion tackles real-world image noise.

MuLAN, a novel learned diffusion process, achieves state-of-the-art density estimation by adaptively adding multivariate Gaussian noise at varying rates across an image, significantly reducing trainin…

Diffusion models with cross-attention: a powerful inductive bias for effortless disentanglement!

DiffSF boosts scene flow estimation accuracy and reliability by cleverly combining transformer networks with denoising diffusion models, offering state-of-the-art results and uncertainty quantificatio…

DiffLight: a novel conditional diffusion model for traffic signal control effectively addresses data-missing scenarios by unifying traffic data imputation and decision-making, demonstrating superior p…

This paper introduces a novel differentiable framework for learning task graphs from video demonstrations of procedural activities. By directly optimizing the weights of a task graph’s edges, the mod…

DeSparsify: A stealthy adversarial attack exhausts vision transformer resources by exploiting token sparsification mechanisms’ dynamic nature, highlighting the need for improved resource management i…

Deep Submodular Peripteral Networks (DSPNs) learn submodular functions efficiently using graded pairwise comparisons, surpassing traditional methods and demonstrating superiority in experimental desig…

Deep learning tackles Markov chain convergence rate analysis! Deep Contractive Drift Calculator (DCDC) provides sample-based bounds in Wasserstein distance, surpassing traditional methods’ limitations…

CycleNet enhances long-term time series forecasting by explicitly modeling inherent periodic patterns using a novel Residual Cycle Forecasting technique, achieving state-of-the-art accuracy and effici…