🏢 Tencent AI Lab

Weight Diffusion (W-Diff) masters evolving domain generalization by using conditional diffusion models to learn classifier weight evolution patterns, enabling superior generalization to unseen future …

VLORA: Boosting Multimodal LLMs efficiency by merging visual features into model weights instead of extending input sequences.

VFIMamba uses state-space models for efficient and dynamic video frame interpolation, achieving state-of-the-art results by introducing a novel Mixed-SSM Block and curriculum learning.

ALPHALLM boosts LLM performance in complex reasoning tasks by using imagination, search, and criticism to create a self-improving loop, eliminating the need for extra training data.

Researchers found that superior OOD detection performance comes at the cost of reduced generalization. Their novel Decoupled Uncertainty Learning (DUL) algorithm harmonizes OOD detection and generali…

StrategyLLM uses four LLM agents to generate consistent, generalizable few-shot prompts, significantly improving LLM problem-solving performance across various tasks.

Self-play adversarial language game boosts LLM reasoning!

SAFE, a novel parameter-efficient tuning framework, boosts pre-trained model performance in continual learning by balancing model stability and plasticity through slow and fast learning stages, signif…

RobIR: Robust inverse rendering in high-illumination scenes using ACES tone mapping and regularized visibility estimation for accurate BRDF reconstruction.

RLE: A novel data augmentation strategy unifying cross-spectral re-ID, significantly boosting model performance by mimicking local linear transformations.

RG-SAN achieves state-of-the-art 3D referring expression segmentation by leveraging spatial awareness and rule-guided weak supervision, significantly improving accuracy and handling of ambiguous descr…

Opponent Modeling with In-context Search (OMIS) leverages in-context learning and decision-time search for stable and effective opponent adaptation in multi-agent environments.

LLMs’ performance drastically varies depending on prompt phrasing; this paper introduces ROBUSTAL-PACAEVAL to evaluate lower-bound performance via worst-case prompt analysis, revealing model inconsist…

M³GPT, a novel multimodal framework, achieves superior motion comprehension and generation by integrating text, music, and motion data into a unified LLM representation.

SignCL, a novel contrastive learning strategy, significantly boosts gloss-free sign language translation by mitigating representation density, achieving substantial performance gains.

IDGen synthesizes LLM evaluation prompts using Item Discrimination theory, creating a more challenging and discriminative dataset than previous methods.

TAGI, a novel method, generates task-specific adapters from instructions, enhancing LLM cross-task generalization by using knowledge distillation and a two-stage hypernetwork training process.

Boost multi-task reinforcement learning with Cross-Task Policy Guidance (CTPG)! CTPG cleverly uses policies from already mastered tasks to guide the learning of new tasks, significantly improving effi…

DiffusionFake enhances deepfake detection by cleverly reversing the image generation process, enabling detectors to learn more robust features and significantly improve cross-domain generalization.

Diffusion-of-Thought (DoT) boosts reasoning in diffusion language models by enabling parallel reasoning steps, outperforming larger autoregressive models in speed and accuracy.