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

My work tracks the evolution of data annotation, from its early days of manual labeling to LLM/MLLM-labeled data.

For specific deep learning tasks, such as object detection, text classification, and robotics VLA models, I believe that under the "data-cover" paradigm, re-labeling is a key method for improving accuracy. As for LLMs and MLLMs, I am currently learning how to improve their accuracy as measured by standard benchmarks.