Center for Functional Nanomaterials Seminar

Training machine learning models on tasks with limited or no training data is a crucial problem, typically limiting their generalizability and effectiveness. This poses several challenges for machine learning models, such as overfitting, model complexity, and data bias. Creating annotated data for training, on the other hand, is time-consuming, expensive, and requires linguistic knowledge. In this talk, I will focus on NLP/AI and talk about my work to overcome data scarcity problems by using data-augmentation, domain-adaptation, and transfer learning techniques for NER, grammar correction, and semantic sentence similarity tasks. Moreover, I will talk about other relevant topics, such as knowledge graph-based recommendation and persona-based open domain dialog systems. Finally, I will briefly discuss LLMs and their potential to demonstrate how we can apply them to develop experiment control assistants.