Posters

Boost time-series model accuracy with Segment, Shuffle, and Stitch (S3)! This simple layer shuffles data segments to enhance representation learning, improving classification, forecasting, and anomaly…

Unsupervised SAM (UnSAM) achieves competitive image segmentation results without human annotation, surpassing previous unsupervised methods and even improving supervised SAM’s accuracy.

AnyChange achieves zero-shot image change detection by adapting the Segment Anything Model (SAM) via a training-free bitemporal latent matching method, significantly outperforming previous state-of-th…

SEEV framework efficiently verifies ReLU neural barrier functions by reducing activation regions and using tight over-approximations, significantly improving verification efficiency without sacrificin…

Dissected Dynamics Modeling (DDM) excels at multi-modal visual reinforcement learning by cleverly separating and integrating common and unique features across different sensory inputs for more accurat…

Boosting vision-language model performance, Contrastive ALignment (CAL) prioritizes visually correlated text tokens during training via a simple, computationally efficient re-weighting strategy, signi…

TokenCLIPose leverages language priors to predict human keypoints beyond bounding boxes, improving pose estimation accuracy significantly on ice hockey, lacrosse and CrowdPose datasets.

AI agents can secretly collude using steganography, hiding their interactions from oversight. This research formalizes this threat, analyzes LLMs’ capabilities, and proposes mitigation strategies.

SearchLVLMs: A plug-and-play framework efficiently augments large vision-language models with up-to-date internet knowledge via hierarchical filtering, significantly improving accuracy on visual quest…

Revolutionizing large neural networks, this paper introduces a continuous parameterization of structured matrices, discovering that full-rank structures without parameter sharing achieve optimal scali…

Training-free architecture search finds optimal subnets in LLMs, boosting inference speed and slashing memory needs without retraining.

SE(3)-bi-equivariant Transformers (BITR) revolutionizes point cloud assembly by guaranteeing robust alignment even with non-overlapping clouds, thanks to its unique equivariance properties.

SDP4Bit achieves up to 4.08x speedup in LLM training by quantizing weight differences and gradients to ~4 bits, maintaining accuracy.

SCube: Instant large-scale 3D scene reconstruction from sparse images using VoxSplats, a novel 3D Gaussian splat representation.

SCOREQ: a novel triplet loss contrastive regression approach for superior speech quality prediction, addressing generalization issues in no-reference metrics.

Researchers developed a novel algorithm to automatically find optimal schedules for denoising diffusion models (DDMs), significantly improving sample quality and efficiency without manual parameter tu…

Score-based generative models are provably robust to multiple error sources, as shown via a novel Wasserstein uncertainty propagation theorem.

FuncMol: A new neural field model generates 3D molecules efficiently, outperforming existing methods by achieving an order of magnitude faster sampling speed.

Researchers improved 3D shape generation from 2D diffusion models by showing that existing Score Distillation Sampling is a reparameterized version of DDIM and fixing its high-variance noise issue via…

Schur Nets boost higher-order GNNs by efficiently exploiting local graph structure for automorphism equivariance, achieving improved performance without the computational burden of traditional methods…