The catastrophe of traditional statistical learning
Deep learning has seen slower adoption in the medical field than in other areas. One reason for that is, perhaps counterintuitively, the prevalence of predictive modeling in medicine. Many people work on clinical prediction models—and most of those people cut their teeth in the era when small specialized models were the norm. The journey from small specialized models to big general models has challenged age-old concepts in machine learning. Traditional notions—such as the importance of interpretability, the risk of overfitting with more parameters, and the necessity of task-specific training data—have been replaced with unconventional insights. More parameters often lead to better generalization, individual parameter interpretation has lost its emphasis, and big models can be trained for multiple tasks, unlocking new capabilities.
In this Endpoints webinar, join Charles Fisher, Founder, and CEO of Unlearn.AI, to unlearn old intuitions and discover how we can apply these insights to navigate the next frontier in machine learning and healthcare. This talk is for anyone interested in understanding why machine learning techniques have advanced so significantly in the last 15 years, why the machine learning revolution is here to stay, and how these changes will impact the future of medicine.
