Next DiT

"Next" research focuses on improving prediction and generation capabilities across diverse domains, aiming to enhance efficiency and robustness. Current efforts involve developing novel transformer architectures (like DiT variants) and integrating graph neural networks to better handle complex relationships in data, such as spatio-temporal dependencies in mobility modeling or topological information in medical image segmentation. These advancements are impacting various fields, from personalized recommendations and anomaly detection to medical imaging and even fundamental physics research through improved AI-driven analysis of high-energy physics data.