-
A Hands-On Machine Learning Workflow: scikit-learn, XGBoost, TensorFlow/Keras and PyTorch on Benchmark Datasets
Five runnable tutorials on public datasets with scikit-learn, XGBoost, TensorFlow, Keras and PyTorch, from baselines to training curves and error analysis.
-
Matching Model Complexity to Data: A Controlled Comparison of scikit-learn, XGBoost, Keras and PyTorch
Four ML libraries on one tabular dataset, one split and fixed seeds, showing when a 31-parameter logistic regression beats ensembles and neural networks.
-
Physics-Informed Neural Networks versus Adaptive Runge-Kutta: Accuracy and Cost on a Forced Linear ODE
A PINN and an adaptive RK45 solver are compared on a forced linear ODE with a known exact solution, measuring accuracy and wall-clock cost.
-
Polynomial Chaos Surrogates for Uncertainty Quantification: Legendre Projection versus Monte Carlo
A Legendre polynomial chaos surrogate for Runge's function, built by Gauss quadrature, matches million-sample Monte Carlo accuracy with 20 evaluations.
-
Stiffness and Absolute Stability: Explicit and Implicit Solvers on the Van der Pol Oscillator
Why stiffness forces explicit ODE solvers to take tiny steps, shown through Dahlquist stability regions and an RK45 versus Radau benchmark on Van der Pol.