5 papers
FALCON: False-Negative Aware Learning of Contrastive Negatives in Vision-Language Alignment
Myunsoo Kim, Seongwoong Shim, Byung-Jun Lee
False negatives pose a critical challenge in vision-language pretraining (VLP) due to the many-to-many correspondence between images and texts in large-scale datasets. These false…
Beyond RAG vs. Long-Context: Learning Distraction-Aware Retrieval for Efficient Knowledge Grounding
Seongwoong Shim, Myunsoo Kim, Jae Hyeon Cho +1
Retrieval-Augmented Generation (RAG) is a framework for grounding Large Language Models (LLMs) in external, up-to-date information. However, recent advancements in context window s…
Adaptive Non-uniform Timestep Sampling for Accelerating Diffusion Model Training
Myunsoo Kim, Donghyeon Ki, Seong-Woong Shim +1
As a highly expressive generative model, diffusion models have demonstrated exceptional success across various domains, including image generation, natural language processing, and…
Rethinking DPO: The Role of Rejected Responses in Preference Misalignment
Jay Hyeon Cho, JunHyeok Oh, Myunsoo Kim +1
Direct Preference Optimization (DPO) is a simple and efficient framework that has attracted substantial attention. However, it often struggles to meet its primary objectives -- inc…
NBDI: A Simple and Effective Termination Condition for Skill Extraction from Task-Agnostic Demonstrations
Myunsoo Kim, Hayeong Lee, Seong-Woong Shim +2
Intelligent agents are able to make decisions based on different levels of granularity and duration. Recent advances in skill learning enabled the agent to solve complex, long-hori…