Moontae Lee

Director@LG Corporation·韩国·OpenReview

研究方向

Code Generation/Refinement · Reasoning · Planning · Alignment/Self-evolving from Human/AI Feedback · Generative AI: Diffusion/Consistency Models and Machine Unlearning · Retrieval-Augmented Generation/Evaluation · Large Language Models (Text LLMs and Code LLMs)

Moontae Lee

SafeDPO: A Simple Approach to Direct Preference Optimization with Enhanced Safety

The CoT Encyclopedia: Analyzing, Predicting, and Controlling how a Reasoning Model will Think

Hybrid Deep Searcher: Scalable Parallel and Sequential Search Reasoning

SPRIG: Improving Large Language Model Performance by System Prompt Optimization

Contrastive Thinking Decoding can Improve Answers for Reasoning Models

Process Reward Models That Think

Training-free Detection of AI-generated images via Cropping Robustness

SafeDPO: A Simple Approach to Direct Preference Optimization with Enhanced Safety

Counterfactual Voting Adjustment for Quality Assessment and Fairer Voting in Online Platforms with Helpfulness Evaluation

Mean-field Continuous Sequence Predictors

Towards Robust and Parameter-Efficient Knowledge Unlearning for LLMs

Training-free Detection of AI-generated Images via High-frequency Influence

Stochastically Capturing Partial Relationship among Features for Multivariate Forecasting