Let Your Imagination Flow!🎨 Flow Matching and Rectified Flows

Welcome to the Madrid ML Meetup! 🤗 LLMs with agentic loops are all the rage right now, but another family of methods plays an important role in image generation: flow models. These models learn a process that transforms samples from a simple random distribution into samples from a complex, unknown distribution represented by a large dataset. This enables the generation of synthetic data, including images, videos, and tabular data, in a way that scales effectively to large training datasets. 🧠 🎨 In this talk, we will provide an intuitive introduction to two popular flow methods: Flow Matching and Rectified Flows. Starting with simple distributions, we will explore their basic concepts before demonstrating an application using a simple image dataset. For those who want to go the extra mile, a public implementation is available on GitHub. 💻 🎤 Our speaker, Álvaro Barbero, holds a PhD in machine learning from the Universidad Autónoma de Madrid. His research interests include open-source large language models and image generation methods. He lectures on the master’s programme in Data Science and AI at Afi Global Education and previously served as Director of AI at the Instituto de Ingeniería del Conocimiento. He currently leads the AI Lab at Lynx Tech, a Spanish scale-up that uses the latest artificial intelligence techniques to combat fraud and money laundering in the banking sector. 🛡️ The event is organized by Machine Learning Circle, Universidad Politécnica de Madrid, Universidad Rey Juan Carlos, and ARQUIMEA Research Center, in collaboration with the Madrid City Council. 🏦