논문해외HF Daily Papers
Where Do Vision Language Models Fail at 3D Reasoning?
Vision-language models can get individual judgments right while struggling to recover and consistently use the exact properties of a 3D object.
실시간
국내외 뉴스 · 기업 공식 발표 · 논문 · 영상 · 커뮤니티 소식을 한국시간 기준으로 제공합니다. 같은 소식은 하나로 묶고, 기본은 핵심만 보여 드립니다.
모든 수집 소식(구글 뉴스로 들어온 단건 보도 포함).
Vision-language models can get individual judgments right while struggling to recover and consistently use the exact properties of a 3D object.
Predicting future physical states stands as a central challenge in embodied intelligence.
Effective computer-use agents (CUAs) should seamlessly combine visual GUI interaction with code and terminal execution within a single policy.
Trajectory-level GRPO is a common approach for training multi-step GUI agents, but repeatedly rolling out complete trajectories from the task root can be…
Coding agents increasingly rely on automated approvers to decide whether a proposed shell command may run.
On-policy Self-distillation (OPSD) has proven effective for post-training large language models by constructing a teacher conditioned on privileged references.
Safety alignment for Large Language Models is heavily focused on English with limited multilingual representation, particularly damaging low-resource language…
The pursuit of fair and equitable AI in human-centric tasks is critically hindered by the inherent biases and privacy risks of real-world datasets, which offer…
DyCraft: An Agentic System for Long-Horizon 4D Scene Generation
기존 코딩 에이전트는 고품질 3D 장면을 만들지만 물리적으로 불가능한 결과물을 내는 등 장기 애니메이션 생성에 어려움을 겪는다.
Incentivizing Self-Learning in Large Language Models via Reinforcement Learning
대규모 언어 모델(LLM)은 지식 집약적 과제 해결에 파라미터 지식과 RAG를 활용한다. 그러나 생소한 분야의 문제 해결에는 단순 사실 검색을 넘어 지식 습득과 추론의 긴밀한 결합이 필요하다.
Scaling Mathematical Proof Supervision from Structured Textbooks
언어 모델용 수학 감독은 통상 단일 문제-해답 쌍으로 단순화되어 출처의 구조와 증거를 버린다. 이에 대규모 코퍼스 TOME-Math를 도입한다.
Music audio generation spans instrumental clips, full songs with vocals, edits of existing recordings, and systems that play alongside a musician in real time.
Jev enables efficient decision-making with large language models (LLMs) through typed querying primitives, allowing the same semantic state to be expressed in…
LCM-Promo: An LLM-Based Uplift Model for Promotion Allocation
상거래 데이터로 파인튜닝된 LLM 기반 모델 LCM-Promo를 통해 사용자 의도와 프로모션 반응을 예측한다.
AutoVend: Benchmarking LLM Agents on Long-Horizon Autonomous Vending Operations
대형 언어 모델(LLM) 에이전트는 소비자의 쇼핑부터 판매자의 비즈니스 운영까지 웹 기반 상거래를 확장하고 있다.
Existing multi-agent collaboration (MAC) methods have primarily explored a group thinking collaboration topology, where agents independently generate solutions…
When the sources retrieved for a query disagree, the right response depends on why they disagree: a question with several valid answers, a contested topic…
Many open mixture-of-experts (MoE) language models train with Muon instead of AdamW.
Classification accuracy and true-positive and false-positive rates alone do not determine whether restarting an AI agent improves success.
Large language models (LLMs) have shown increasing promise in research-level problem solving, yet a comprehensive collection of such problems is still missing…
We present PREEN (Preset Regression via an Exemplar-based Editor Network), which casts color style transfer as regression onto the 29 parameters of a…
Reinforcement learning (RL) has shown promise for training multimodal agents to reason and use tools.
Visual agents need to coordinate reasoning and action while their environment evolves during inference.
Jailbreak benchmarks test one checkpoint and treat which jailbreaks break it as a property of its pretraining data and safety training.
Solving long-horizon agentic tasks requires language agents to interact with their environments over multiple steps, and reliable performance often comes from…
Low-rank SVD-based compression offers a powerful strategy to reduce the computational costs of LLMs.
A robot generalist must interpret a scene, act on it, and satisfy every condition of a task.
Face image quality assessment (FIQA) estimates the utility of a face image for automated face recognition and has been proposed as a morphing attack detection…
We introduce SIMIT, a test-time self-improvement framework in which a vision-language model creates its own query-specific training data before answering an…