Curing AI Issues at the Source: The Power of Data-Centric Learning
Yanjie Fu, Associate Professor, School of Computing and Augmented Intelligence, Arizona State University
Abstract: While modern machine learning models have achieved remarkable performance, they remain highly vulnerable to systemic issues such as poor generalization, bias, overfitting, domain shift, and adversarial attacks. Traditionally, the research community has focused heavily on model-centric improvements; however, practical deployment increasingly demands a shift toward a Data-Centric AI paradigm. Drawing inspiration from gene editing—where genetic codes are modified to cure diseases—this talk introduces the concept of "Data Reshaping." By using AI to systematically edit and reconstruct data into optimal, task-specific data shape, we can cure AI issues at their source and boost downstream predictive accuracy. In this talk, we will navigate the landscape of data-centric learning (problems, methods, emerging opportunities) and explore my journey from reinforcement data reshaping, to generative data reshaping, to LLM and agentic data reshaping.
Cost: free
Kisa Kowal
(847) 491-3974
Email