Multi Stage
Multi-stage approaches are increasingly prevalent in machine learning, aiming to improve model performance and efficiency by breaking down complex tasks into sequential sub-tasks. Current research focuses on leveraging this strategy across diverse applications, employing architectures like two-stage neural networks, incorporating techniques such as retrieval augmentation and contrastive learning, and utilizing pre-trained models for transfer learning. This methodology offers significant advantages in handling high-dimensional data, mitigating computational costs, and enhancing accuracy, particularly in resource-constrained or low-data scenarios, with impacts spanning fields from image processing and natural language processing to medical diagnosis and robotic manipulation.
Papers
Two-layer retrieval augmented generation framework for low-resource medical question-answering: proof of concept using Reddit data
Sudeshna Das, Yao Ge, Yuting Guo, Swati Rajwal, JaMor Hairston, Jeanne Powell, Drew Walker, Snigdha Peddireddy, Sahithi Lakamana, Selen Bozkurt, Matthew Reyna, Reza Sameni, Yunyu Xiao, Sangmi Kim, Rasheeta Chandler, Natalie Hernandez, Danielle Mowery, Rachel Wightman, Jennifer Love, Anthony Spadaro, Jeanmarie Perrone, Abeed Sarker
Deep Grokking: Would Deep Neural Networks Generalize Better?
Simin Fan, Razvan Pascanu, Martin Jaggi