San Francisco Bay Area
Build video understanding and moderation system with Vision Language Large Models (VLLM).
• Introduced methods for training deep learning models on unlabeled data based on biased models. • Conducted systematic analysis on understanding the advantages of using self-supervised models on noisy datasets. • Developed a method to reduce the effects of corrupted data on model training. • Invented an algorithm to find clean samples from noisy datasets with a few clean samples as reference.