Learning from Data: The Two Cultures

Explore the intersection of statistical modeling and artificial intelligence in this thought-provoking ACM conference talk. Delve into the evolution of data analysis approaches, from Breiman's "Two Cultures" of statistical modeling to the emerging task modeling culture driving AI breakthroughs. Examine the strengths and limitations of various methodologies, including data modeling, algorithmic modeling, and task-first approaches. Gain insights from speaker Adji Bousso Dieng, a renowned computer scientist and statistician, as she discusses the importance of bridging these cultures to address challenges in AI development. Learn about innovative techniques such as deep exponential families, structured variation, and embedded topic models. Discover how principles from statistical modeling can enhance task modeling practices, including careful data selection, experimental design, and meaningful benchmarking. Reflect on the ethical considerations and potential pitfalls in AI research, such as data imbalance, carbon footprint, and controversial tasks. Equip yourself with a comprehensive understanding of current trends in data analysis and AI, and explore ways to integrate domain knowledge and exploratory tools for more effective learning from data.