Seoul, South Korea
Building Pegasus, the Video-first AI.
Pretraining efficient foundational model architectures.
Explored interaction between knowledge graph and language models; implemented graph neural network with BERT embeddings to improve SOTA performance on Commonsense Knowledge Graph Population task; EMNLP 2021
Explore Active Learning techniques to efficiently finetune multilingual models for Machine Translation task.
Explore causal intervention based model editing methods to understand and control language model’s output and reduce toxicity.