Posters

Recommendation algorithms on UGC platforms face a critical trade-off: prioritizing user satisfaction reduces creator engagement, jeopardizing long-term content diversity. This research introduces a ga…

ConBench: Unveiling Inconsistency in Large Vision-Language Models

A new, nuisance-free Distributionally Robust Metric (DRM) is proposed for selecting robust Conditional Average Treatment Effect (CATE) estimators, improving the reliability of personalized decision-ma…

Neural network efficiency is improved by analyzing weight norm variance across channels to identify optimal layer widths, resulting in reduced parameters and boosted performance.

MapUnveiler: a novel paradigm for online vectorized HD map construction that leverages clip-level token interaction and propagation to unveil hidden map elements and achieve state-of-the-art performan…

Self-attention, a key component of transformers, is revealed to be a projection of query vectors onto the principal components of the key matrix, derived from kernel PCA. This novel perspective leads…

SalientLoRA unveils optimal LoRA ranks by analyzing rank salience via time-series analysis, improving fine-tuning efficiency and performance significantly.

Transformers learn complex tasks surprisingly well through in-context learning, but the mechanism remains unclear. This paper proves that a two-layer transformer trained on n-gram Markov chain data co…

LLMs struggle with genuine causal reasoning; new benchmark CausalProbe-2024 reveals limitations, and G2-Reasoner method improves causal reasoning by integrating general knowledge and goal-oriented pro…

Two-Stage Backdoor Defense (TSBD) unveils and mitigates backdoor vulnerabilities by cleverly unlearning weight changes and suppressing backdoor neuron activeness, significantly improving the robustnes…

Vision Transformers (ViTs) generalize surprisingly well, even when overfitting training data; this work provides the first theoretical explanation by characterizing the optimization dynamics of ViTs a…

Untrained neural networks revolutionize snapshot compressive imaging (SCI) by enabling high-dimensional data recovery from a single 2D measurement, achieving state-of-the-art results without needing e…

First unsupervised object detection method with theoretical guarantees to recover true object positions, up to quantifiable small shifts!

Modality Adaptation with Diffusion Models (MADM) achieves state-of-the-art semantic segmentation by using pre-trained text-to-image diffusion models to enhance cross-modality capabilities and generate…

AltO: a novel unsupervised learning framework for accurately estimating homography from multimodal image pairs, achieving performance comparable to supervised methods.

This study introduces a novel unsupervised hierarchy-agnostic image segmentation method achieving detailed and unbiased parsing of semantic image structures across various datasets.

AI automates mathematical constant formula discovery by analyzing convergence dynamics, revealing known and novel formulas for π, ln(2), and other constants.

ImAD: An end-to-end unsupervised anomaly detection method conquering missing data’s challenge by integrating imputation and detection in a unified framework, achieving superior accuracy!

UnSeg: One universal unlearnable example generator protects images from image segmentation model training.

Fast disease progression inference is achieved via optimal transport, enabling high-dimensional, interpretable models and offering broad clinical applications.