Posters

LSH-MoE accelerates Mixture-of-Experts training by 1.28x-2.2x via Locality-Sensitive Hashing, significantly reducing communication costs.

LRM-Zero: Training large reconstruction models solely on synthetic data, achieving quality comparable to real-data trained models.

LP-3DGS learns to optimally prune 3D Gaussian splatting, achieving significant efficiency gains without compromising rendering quality via a trainable binary mask and the Gumbel-Sigmoid method.

This paper establishes tight lower bounds for the uniform stability of gradient-based bilevel programming algorithms used for hyperparameter optimization, resolving a key open problem regarding the ti…

First optimal algorithms matching lower bounds for non-smooth convex decentralized optimization over time-varying networks are presented, substantially improving theoretical performance.

FRLC: a novel algorithm for low-rank optimal transport using latent coupling, enabling faster computation and better interpretability for diverse applications.

Low-precision local training, surprisingly, is sufficient for accurate federated learning, significantly reducing communication and computation costs.

Low-degree polynomials fail to efficiently infer roots in broadcasting tree problems below the Kesten-Stigum bound.

LOVA³ enhances MLLMs by teaching them to ask and assess image-based questions, improving their multimodal understanding and performance on various benchmarks.

LoTLIP boosts language-image pre-training for superior long text understanding by cleverly integrating corner tokens and utilizing a massive dataset of 100M long-caption images.

Deep learning optimization is revolutionized by a new function class, enabling convergence guarantees without over-parameterization and accommodating saddle points.

Lorentz Geometric Algebra Transformer (L-GATr): A novel, scalable architecture for high-energy physics, achieving high-precision, data-efficient learning and outperforming existing methods on regressi…

LoRA-GA: A novel initialization method dramatically speeds up low-rank adaptation (LoRA) for LLMs, achieving convergence rates comparable to full fine-tuning while improving performance.

LoQT enables efficient large language model training on consumer hardware via quantized weights and low-rank weight updates, overcoming memory limitations.

Generative image restoration models face a critical trade-off: higher perceptual quality often leads to increased hallucinations (unreliable predictions).

LookHere: Vision Transformers excel at high-resolution image classification by using 2D attention masks to direct attention heads, improving generalization and extrapolation.

This paper enhances prophet inequalities by allowing lookback, improving competitive ratios and providing algorithms for diverse observation orders, thereby bridging theory and real-world online selec…

New dataset MUSIC-AVQA-R and a multi-faceted cycle collaborative debiasing strategy significantly improve audio-visual question answering robustness.

AdaptOD: a novel approach for robust OOD detection in long-tailed recognition, dynamically adapting outlier distributions to true OOD distributions using a dual-normalized energy loss for improved acc…

Dynamic Rebalancing Contrastive Learning with Dual Reconstruction (2DRCL) pre-training significantly boosts object detection accuracy, especially for underrepresented classes.