Posters

Deep Equilibrium Models excel on imbalanced data due to feature convergence and self-duality properties, unlike explicit models, as shown through Neural Collapse analysis.

PartCLIPSeg, a novel framework, leverages generalized parts and object-level contexts to achieve significant improvements in open-vocabulary part segmentation, outperforming state-of-the-art methods.

Larger machine learning models don’t always mean better performance; strategic interactions can reverse this trend, as this research shows, prompting a new paradigm for model selection in games.

Linear probing then fine-tuning (LP-FT) significantly improves language model fine-tuning; this paper uses Neural Tangent Kernel (NTK) theory to explain why.

Researchers unveil how multi-modal LLMs process information, revealing that early layers are key for storage, and introduce MULTEDIT, a model-editing algorithm for correcting errors and inserting new …

Diffusion models generate unrealistic images by smoothly interpolating between data modes; this paper identifies this ‘mode interpolation’ failure and proposes a metric to detect and reduce it.

Diffusion models’ surprising generalizability stems from an inductive bias towards learning Gaussian data structures, a finding that reshapes our understanding of their training and generalization.

Language model emergent abilities aren’t exclusive to large models; they emerge when pre-training loss falls below a threshold, irrespective of model or data size.

Researchers unveil a novel framework to dissect bias in large-scale visual datasets, identifying unique visual attributes and leveraging language models for detailed analysis, paving the way for creat…

New methods minimize outlier features in transformer training, improving quantization and efficiency without sacrificing convergence speed.

Training-free guidance revolutionizes diffusion models by enabling zero-shot conditional generation, but suffers from misaligned gradients and slow convergence. This paper provides theoretical analysi…

PamaCF, a novel personalized adversarial collaborative filtering technique, significantly improves recommendation robustness and accuracy against poisoning attacks by dynamically adjusting perturbatio…

Graph self-supervised learning models surprisingly exhibit high redundancy, allowing for significant parameter reduction without performance loss. A novel framework, SLIDE, leverages this discovery f…

Researchers developed SCAV, a novel framework to effectively reveal safety risks in LLMs by accurately interpreting their safety mechanisms. SCAV-guided attacks significantly improve attack success r…

Biologically-inspired ORGANICs neural circuit achieves dynamic divisive normalization, ensuring unconditional stability and seamless backpropagation training for high-dimensional recurrent networks.

Self-Contrast Mixture-of-Experts (SCMoE) boosts MoE model reasoning by cleverly using ‘unchosen’ experts during inference. This training-free method contrasts outputs from strong and weak expert acti…

TRACER, a novel robust offline RL algorithm, uses Bayesian inference to handle uncertainty from diverse data corruptions, significantly outperforming existing methods.

Uncertainty of Thoughts (UoT) algorithm significantly boosts LLMs’ information-seeking abilities, leading to substantial performance gains across diverse tasks.

UMFC: Unsupervised Multi-domain Feature Calibration improves vision-language model transferability by mitigating inherent model biases via a novel, training-free feature calibration method.

UMB: A novel model enhances open-world object detection by understanding model behavior, surpassing state-of-the-art with a 5.3 mAP gain for unknown classes.