I was part of a small-scale research team that laid the foundation for the development of Samsung's Bixby.
Before the advent of TensorFlow and PyTorch, we utilized the Theano Framework to personally establish a multi-GPU training environment for constructing AI models tailored for NLP. Especially, I developed training and evaluation data designs, along with human annotator guidelines, to compile the data. Furthermore, prior to the release of BERT and ELMO, I adopted a multi-task learning approach that simultaneously trained the language model alongside the main model, enhancing the system's contextual understanding capabilities. This innovative methodology enabled the creation of a performance level capable of comprehending human language.
Subsequently, I continued to lead the NLP (Natural Language Processing) Research and Development for the AI Assistant Bixby until the year 2020.