We are proud to share that Ayman Chaouki, PhD candidate at ORAILIX, has had his paper "Branches: Efficiently Seeking Optimal Sparse Decision Trees via AO*"* accepted at the 42nd International Conference on Machine Learning (ICML) 2025.
This work, conducted under the supervision of Professors Jesse Read and Albert Bifet, introduces a novel algorithm called Branches, which efficiently searches for optimal sparse decision trees. The method leverages an AND/OR search framework and solves it using an AO* strategy. This work demonstrates significantly faster convergence than existing methods, both in terms of iteration count and runtime. The algorithm also comes with theoretical guarantees, providing an upper bound on the number of branches evaluated before termination. To deepen your understanding of AND/OR search and AO*, Ayman recommends two essential readings: Principles of Artificial Intelligence by Nils J. Nilsson and Heuristics: Intelligent Search Strategies for Computer Problem Solving by Judea Pearl.The code is available on his GitHub: Branches
The paper can be read on arXiv: Branches: Efficiently Seeking Optimal Sparse Decision Trees with AO* This research reflects ORAILIX’s ongoing commitment to developing AI that is both rigorous and impactful.