Show HN: Modernizing my old PhD work in an evening with little Qwen3.6 MoE (github.com)

A new JAX implementation of the Prioritized Grammar Enumeration (PGE) algorithm has been introduced for symbolic regression, showcased through the pge-jax library. This system autonomously discovers mathematical formulas from data by generating candidate expressions from a defined grammar, fitting their coefficients with JAX's Levenberg-Marquardt optimization, and selecting the top models via a multi-objective Pareto front using the NSGA-II algorithm. The significant advancement in this implementation lies in its fully JAX-native evaluation pipeline, which allows for GPU/TPU acceleration, JIT compilation, automatic differentiation, and eliminates the need for external ML libraries, streamlining the symbolic regression process. The key innovations of pge-jax include a systematic enumeration of expressions, enhanced filtering through grammar production rules, and efficient model evaluation through early termination and memoization techniques. This approach facilitates the creation of a Pareto front of trade-off solutions ranging from simple models to more complex, high-accuracy ones. As a result, users can select the best model interpretation tailored to their specific problem needs. This modernized framework not only simplifies the symbolic regression landscape but also promises to inspire further research and application in the AI/ML community, particularly in automating formula discovery and data analysis.