🏢 National University of Singapore

LogQuant: 2-bit quantization for KV cache, superior accuracy!

MedAgent-Pro: An evidence-based reasoning agentic system for reliable multi-modal medical diagnosis.

URA: Ultra-resolution adaptation made easy! Uses synthetic data & minor weight tuning for efficient, high-res text-to-image diffusion models.

Coarse-to-Fine Token Prediction improves autoregressive image generation by assigning the same coarse label for similar tokens, balancing generation quality and computational efficiency.

4DGS-1K: Achieves 1000+ FPS for dynamic scene rendering via a compact, memory-efficient framework, offering a 41x storage reduction and 9x faster speed.

Impossible videos expose AI limits!

TPDiff accelerates video diffusion by progressively increasing frame rates during diffusion, optimizing computational efficiency with a novel stage-wise training strategy.

PE3R: Achieves fast and accurate 3D scene reconstruction from 2D images by enhanced perception and efficiency.

VLMs often disproportionately trust text over visual data, leading to performance drops and safety concerns.

EDGS: Achieves faster, high-quality dynamic scene rendering by sparse time-variant attribute modeling and intelligent static area filtering.

PhotoDoodle: Mimicking artistic image editing with personalized decorative elements through learning from few-shot pairwise data.

InterFeedback: LMMs need better human feedback to enhance AI assistants!

LongPO: Self-evolve LLMs to excel in long contexts via short-to-long preference optimization, boosting performance without sacrificing short-context skills.

NExT-Mol: Combines 1D language models with 3D diffusion for molecule generation, achieving state-of-the-art performance and validity.

CoT-Valve dynamically adjusts reasoning chain lengths based on task difficulty, significantly reducing inference costs in large language models without substantial accuracy loss.

Enhance-A-Video boosts video generation quality without retraining, by enhancing cross-frame correlations in diffusion transformers, resulting in improved coherence and visual fidelity.

GuardReasoner enhances LLM safety with reasoning-based guardrails, improving performance, explainability, and generalization on various benchmarks.

CLEAR: Conv-Like Linearization boosts pre-trained Diffusion Transformers, achieving 6.3x faster 8K image generation with minimal quality loss.

TinyFusion, a novel learnable depth pruning method, crafts efficient shallow diffusion transformers with superior post-fine-tuning performance, achieving a 2x speedup with less than 7% of the original…

Collaborative Decoding (CoDe) dramatically boosts visual auto-regressive model efficiency.