본문 바로가기

전체 글

[논문] The Imperfective Paradox in Large Language Models https://aclanthology.org/2026.acl-long.689/ The Imperfective Paradox in Large Language ModelsBolei Ma, Yusuke Miyao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.aclanthology.org해당 논문을 보고 작성하였습니다. AbstractLLM은 사건의 compositional semantic을 실제로 이해하는가, 아니면 표면적인 heuristic에 의존하는가? 본 연구에서는 imperfective paradox를 조사한다. 이는 과거진행형이 활동 .. 연구실 공부 2026. 10. 9.
[논문] Preserving LLM Capabilities through Calibration Data Curation: From Analysis to Optimization https://arxiv.org/abs/2510.10618 Preserving LLM Capabilities through Calibration Data Curation: From Analysis to OptimizationPost-training compression has been a widely employed approach to scale down large language model (LLM) and facilitate efficient inference. In various proposed compression methods, including pruning and quantization, calibration data plays a vital role by iarxiv.org해당 논문을 보.. 연구실 공부 2026. 10. 8.
[논문] EchoingPixels: Aliasing-Resistant Joint Token Reduction for Audio-Visual LLMs https://openreview.net/forum?id=e8JTAKZYt6 Verifying your browser | OpenReviewVerifying your browser Complete the check below to continue to OpenReview Please complete the verification above. Have an OpenReview account? Sign in to skip this check.openreview.net해당 논문을 보고 작성하였습니다. AbstractAudio-Visual LLM (AV-LLM)은 많고 중복되는 audio-visual token을 처리할 때 사용되는 많은 연산량 문제가 존재한다. 기존 unimodal compression tec.. 연구실 공부 2026. 8. 31.
[논문] SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics https://arxiv.org/abs/2506.01844 SmolVLA: A Vision-Language-Action Model for Affordable and Efficient RoboticsVision-language models (VLMs) pretrained on large-scale multimodal datasets encode rich visual and linguistic knowledge, making them a strong foundation for robotics. Rather than training robotic policies from scratch, recent approaches adapt VLMs into visarxiv.org해당 논문을 보고 작성하였습니다. Abst.. 연구실 공부 2026. 8. 22.
[논문] EgoVLA: Learning Vision-Language-Action Models from Egocentric Human Videos https://arxiv.org/abs/2507.12440 EgoVLA: Learning Vision-Language-Action Models from Egocentric Human VideosReal robot data collection for imitation learning has led to significant advancements in robotic manipulation. However, the requirement for robot hardware in the process fundamentally constrains the scale of the data. In this paper, we explore training Visarxiv.org해당 논문을 보고 작성하였습니다. Abstra.. 연구실 공부 2026. 8. 22.
[논문] OmniSIFT: Modality-Asymmetric Token Compression for Efficient Omni-modal Large Language Models https://arxiv.org/abs/2602.04804 OmniSIFT: Modality-Asymmetric Token Compression for Efficient Omni-modal Large Language ModelsOmni-modal Large Language Models (Omni-LLMs) have demonstrated strong capabilities in audio-video understanding tasks. However, their reliance on long multimodal token sequences leads to substantial computational overhead. Despite this challenge, token comarxiv.org해당 논문을.. 연구실 공부 2026. 8. 19.
[논문] OmniFocus: Query-Guided Modality-Balanced Token Compression for Omni-Modal Large Language Models https://arxiv.org/abs/2607.03050 OmniFocus: Query-Guided Modality-Balanced Token Compression for Omni-Modal Large Language ModelsOmni modal large language models (OmniLLMs) have attracted wide attention for their ability to jointly process audio and video, but they generate large token sequences under audio-visual inputs, leading to substantial inference cost. Existing audio-visualarxiv.org해당 논문.. 연구실 공부 2026. 8. 19.
[논문] Omni-Prune: Query-Aware Unified Token Pruning for Efficient Omnimodal Large Language Models https://arxiv.org/abs/2607.23445 Omni-Prune: Query-Aware Unified Token Pruning for Efficient Omnimodal Large Language ModelsOmnimodal large language models (OmniLLMs) are rapidly extending multimodal reasoning to cover synchronized audio and video. However, the resulting audio-video token sequences are long, leading to high prefill latency and GPU memory usage at inference timearxiv.org해당 논문을 보고.. 연구실 공부 2026. 8. 18.
728x90

티스토리툴바