2025-04-01s

TeleAntiFraud-28k: A new audio-text dataset designed for telecom fraud detection, tackles data scarcity with innovative synthesis techniques and slow-thinking annotations.

RIG: Synergizes reasoning and imagination in an end-to-end generalist policy for embodied agents, improving sample efficiency and generalization.

Execution-guided SQL generation enhances accuracy in text-to-SQL tasks by using execution results to select the most semantically consistent query, improving performance and reducing inference costs.

Open-Reasoner-Zero pioneers scalable, accessible RL training for reasoning in LLMs, achieving superior performance with a minimalist approach.

KOFFVQA: Objectively evaluates Korean VLMs with a new free-form VQA benchmark, improving evaluation reliability via detailed grading criteria.

RL with Verifiable Rewards is now expanding to diverse domains like medicine!

EAST: Prioritizing uncertainty in self-training refines reasoning of Large Language Models.

Thinking Intervention offers a novel paradigm for controlling reasoning in LLMs, enabling fine-grained guidance and improvements in instruction-following and safety.

Easi3R: Training-free 4D reconstruction via attention disentanglement, enabling dynamic scene understanding from static 3D models.

TextCrafter: Precisely renders multiple texts in complex scenes, overcoming distortion and omission issues in existing visual text generation models.

MoCha: Movie-Grade Talking Character Synthesis!

MeshCraft: Efficient, controllable mesh generation using flow-based DiTs, outperforming auto-regressive methods in speed and user control.

Survey on efficient inference methods for Large Reasoning Models, focusing on mitigating token inefficiency while preserving quality.

Adapting Stable Diffusion for faster Text-to-Mesh Generation, PRD efficiently creates high-quality 3D models without needing extensive 3D training data.

TokenHSI: Unified Transformer for Physical Human-Scene Interactions through Task Tokenization.

LLMs generate Python heuristics for classical planning, outperforming traditional methods and challenging the state-of-the-art planning techniques.

DeLoRA: Decoupling angles and strength in low-rank adaptation for robust & efficient finetuning of large models!

UPME: Peer review for MLLMs, minus human bias!