🏢 Hong Kong Polytechnic University

Voxel Mamba: a group-free 3D object detection method using state space models, achieving higher accuracy and efficiency by overcoming limitations of serialization-based Transformers.

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…

Causal Representation Editing (CRE) improves safe image generation by precisely removing unsafe concepts from diffusion models, enhancing efficiency and flexibility.

Noise Regularization rescues Deep Canonical Correlation Analysis from model collapse!

Orthogonal regularization prevents dimensional collapse in self-supervised learning, significantly boosting model performance across diverse benchmarks.

LM-WEATHER uses pre-trained language models to create highly accurate, personalized weather models directly on resource-constrained devices, achieving state-of-the-art results with significantly reduc…

OwMatch: a novel framework conquering open-world semi-supervised learning challenges by combining conditional self-labeling and consistency for substantially enhanced accuracy across known and unknown…

OSEDiff: One-step diffusion network for real-world image super-resolution, achieving comparable or better results than multi-step methods with significantly reduced computational cost and improved ima…

MetaLA: Unified optimal linear approximation to softmax attention map, achieving linear complexity and surpassing existing models in various benchmarks.

KnowGPT: A novel framework boosts Large Language Model accuracy by intelligently integrating knowledge graphs, significantly reducing factual errors and achieving near-human performance on benchmark d…

LLM4EA: A novel framework efficiently merges knowledge graphs using LLMs, overcoming noisy annotations and high costs via active learning and unsupervised label refinement, boosting accuracy and effic…

DEMO framework enhances text-to-video generation by decomposing text encoding and conditioning into content and motion components, resulting in videos with significantly improved motion dynamics.

Cross-Modal Representation Flattening (CMRF) improves multi-modal domain generalization by creating consistent flat loss regions and enhancing knowledge transfer between modalities, outperforming exis…

Can AI understand humor? A new benchmark, YESBUT, reveals that even state-of-the-art models struggle with the nuanced humor of juxtaposed comics, highlighting the need for improved AI in understandin…

Coke: A cost-efficient KBQA strategy using LLMs and KGMs, maximizing accuracy while minimizing GPT-4 fees by up to 20.89%

DDL-CXR dynamically generates up-to-date chest X-ray image representations using latent diffusion models, effectively addressing asynchronous multimodal clinical data for improved prediction.

AdaNeg dynamically generates negative proxies during testing to improve vision-language model OOD detection, significantly outperforming existing methods on ImageNet.