ML Pioneer John Platt Discusses AI for Science, Climate Tech, and Algorithm History
Machine learning pioneer John Platt featured on the Latent Space podcast to discuss the evolution of classical ML algorithms, Google's AI for Science projects, and how automated science can tackle global challenges like climate change. As artificial intelligence expands beyond generative text toward accelerating scientific discovery, insights from foundational figures like Platt highlight how ML can transform energy, materials science, and climate modeling. John Platt is best known in computer science for inventing Sequential Minimal Optimization (SMO) for SVM training and Platt scaling for probability calibration, both of which are core components in libraries like scikit-learn. Beyond his research contributions, he holds a Sci-Tech Academy Award, has two asteroids named after him, and currently leads AI research initiatives at Google.
## BACKGROUND
Sequential Minimal Optimization (SMO) dramatically simplified the training of Support Vector Machines (SVMs) by breaking down large quadratic programming problems into smaller sub-problems. Meanwhile, Platt scaling fits a logistic regression model to raw SVM outputs to produce calibrated probabilities, a technique still widely used in modern machine learning pipelines.