My research focuses on scalable methods to achieve 99% accuracy in deep learning models, applicable to NLP, CV, VLA, and LLM/MLLM.

For example, Re-Label and VLA share the same closed-loop paradigm: Predict → Error → Correction → Re-Predict.