Shallow Water Equations: Numerical Schemes, PINNs, and Friction Parameterization
Coming Soon
A comparative study of solving the shallow water equations using classical numerical schemes such as Lax and MacCormack versus Physics-Informed Neural Networks (PINNs). The discussion also explores parameterizing the friction term using conventional approaches (e.g., Manning’s equation, Darcy–Weisbach, Chezy’s formula) compared to machine learning-based parameterizations, highlighting trade-offs in accuracy, interpretability, and computational efficiency.
Read More →FutureWarning: Democracy Deprecation Notice
Published on October 3, 2025
Democracy has long been treated as the most "stable release" of political systems which is tested through centuries of trial and error, refined through revolutions, and widely adopted because it worked better than the alternatives. But like any system, its adoption was never purely about ideals. It was, above all, a practical solution to control the problem.
Read More →Solving an ODE: Symbolic, Numerical, and PINNs
Published on October 1, 2025
A comprehensive comparison of three approaches to solving ordinary differential equations: analytical solutions with SymPy, numerical approximation using Euler's method, and physics-informed neural networks. We solve dy/dt = 2t - y and compare their insights and results.
Read More →The Trade-off Between Model Expressivity and Inductive Bias in Machine Learning
Published on September 24, 2025
Understanding the balance between model complexity and built-in assumptions is fundamental to successful machine learning. This post explores the critical trade-off between model expressivity and inductive bias, illustrated through simple examples and real-world applications like CNNs vs. Vision Transformers.
Read More →