Posters

SAMPa: Parallelizing gradient computations in Sharpness-Aware Minimization (SAM) achieves a 2x speedup and superior generalization.

Samba: a novel severity-aware recurrent model, tackles cross-domain medical image grading by sequentially encoding image patches and recalibrating states using EM, significantly improving accuracy.

This paper introduces SAM-guided masked token prediction, a novel framework for 3D scene understanding that leverages foundation models to significantly improve 3D object detection and semantic segmen…

SAFEWORLD: a new benchmark reveals and fixes LLMs’ struggle with diverse safety standards.

Reinforcement Learning from Safety Feedback (RLSF) efficiently infers cost functions from trajectory-level feedback, enabling safe policy learning in complex environments.

SAFE, a novel parameter-efficient tuning framework, boosts pre-trained model performance in continual learning by balancing model stability and plasticity through slow and fast learning stages, signif…

TVSAFEOPT: Safe time-varying optimization using spatio-temporal kernels ensures safety while tracking time-varying reward and safety functions, providing optimality guarantees in stationary settings.

This paper characterizes the fundamental tradeoff between trusting and distrusting learned type beliefs in games, establishing upper and lower bounds for optimal strategies in both normal-form and sto…

A novel safe & sparse Newton method (SSNS) for entropic-regularized optimal transport boasts strict error control, avoids singularity, needs no hyperparameter tuning, and offers rigorous convergence a…

A novel primal-dual method boosts offline safe reinforcement learning efficiency for convex CMDPs by using uncertainty parameters and achieving a sample complexity of O(1/(1-γ)√n).

S2HPruner bridges the discretization gap in neural network pruning via a novel soft-to-hard distillation framework, achieving superior performance across various benchmarks without fine-tuning.

S2FT: Structured Sparse Fine-Tuning achieves state-of-the-art LLM fine-tuning performance, training efficiency, and inference scalability by selecting sparsely and computing densely.

S-STE achieves efficient 2:4 sparse pre-training by introducing a novel continuous pruning function, overcoming the limitations of previous methods and leading to improved accuracy and speed.

S-SOS: A new algorithm solves complex, parameterized polynomial problems with provable convergence, enabling efficient solutions for high-dimensional applications like sensor network localization.

S-MolSearch: a novel semi-supervised framework using 3D molecular data and contrastive learning achieves state-of-the-art in bioactive molecule search, outperforming existing methods.

Rule-Based Rewards (RBRs) enhance LLM safety by using AI feedback and a few-shot prompt-based approach, achieving higher safety-behavior accuracy with less human annotation than existing methods.

RTify: A novel framework aligns deep neural networks’ dynamics with human reaction times for improved visual decision-making models.

RSA: Language unlocks metric depth from single images!

RouterDC: A query-based router trained via dual contrastive learning assembles multiple LLMs, significantly outperforming individual LLMs and existing routing methods on both in- and out-of-distributi…

Rough Transformers: A lightweight continuous-time sequence modeling approach using path signatures to significantly reduce computational costs, improving efficiency and accuracy, particularly for long…