A Deep Dive into Program Synthesis: Generating Logic from Examples
Not every problem needs an LLM. Real AI engineering means having a toolbox, not just a hammer. This session dives into a crucial tool for precision tasks: Program Synthesis, a powerful technique for generating reliable, logical code when LLMs won't cut it.
Join speaker Oren Matar at 42 Berlin for a practical, end-to-end walkthrough of building a Program Synthesis system. We'll cover everything from dataset representation to model architectures and tree search algorithms. Using the generation of 3D textures as a working example, you'll learn a powerful methodology for building models that succeed where other approaches fail.
Key Takeaways
↪ Understand Program Synthesis and its applications.
↪ Learn to structure a PS project from dataset to model.
↪ Analyze the trade-offs between different deep learning architectures for this task.
Who Should Attend?
This session is for ML practitioners, data scientists, and engineers with a working knowledge of deep learning (especially transformers and CNNs) who want to build robust, high-precision models.
Agenda
16:00: Doors Open & Registration
16:15: Opening from Ali Qeblawi - Host
16:20: Lecture Begins: A Deep Dive into Program Synthesis - Oren Matar
17:00 - 17:30: Q&A and Networking