elixir-berlin
A Deep Dive into Program Synthesis and reasoning: Generating Logic from Examples
Your name
Oren Matar
Your GitHub
No response
Your Social Link
https://www.linkedin.com/in/oren-matar-96476b139/
Your proposal type: talk, lightning talk or til (today I learned)
talk
Title of your presentation
A Deep Dive into Program Synthesis and reasoning: Generating Logic from Examples
Description
Program Synthesis (PS) is the task of automatically generating logical procedures or code from a small set of input-output examples. Often referred to as the "holy grail" of AI, PS generates models that are both highly generalizable and inherently interpretable. Its applications span a wide range of domains. While large language models and agents dominate today’s discussions, they often struggle with problems requiring precise reasoning or structured generalization. This session intends to highlight an alternative, underexplored, approach for solving reasoning problems. We will walk through the full lifecycle of developing a practical Program Synthesis system: from representing your dataset using a unique graph structure, to selecting the right transformer architecture for generating solutions, and integrating tree-search algorithms – similar to those used in AlphaGo. While the session is grounded in an example use case—generating procedural textures for 3D modeling—the approach is domain-agnostic and can be applied to a wide variety of synthesis problems.
Favourite Month you would like to present in
October
