NeurIPS 2025 Submission: [Enhancing Neural Function Approximation: The XNet Outperforming KAN]

This repository contains the source code and experiments for our NeurIPS 2025 paper:

We provide implementations for function approximation benchmarks, Physics-Informed Neural Networks (PINNs), and PPO-based reinforcement learning baselines.

Note: This repository provides code for experimental reproduction. Some code components reference the original paper [Li et al, 2025] and utilize complex-valued implementations for mathematical equivalence, ensuring consistent results with the theoretical framework presented in our work.

---
Project Structure

├── Comparison_function_approximation/   # Function approximation experiments
├── Comparison-PPO-main/                 # PPO-based policy learning
├── PINN_Heat_equation/                  # Solving the heat equation in PINN framework
├── PINN_Poisson_equation/               # Solving the Poisson equation in PINN framework
├── Language_translation/               # German-English translation benchmarks
├── Requirements.txt                     # Python dependencies
├── README.md                            # This file

