universal-approximation-theorem

Scripts and notebooks to accompany the book Data-Driven Methods for Dynamic Systems

Codebase for Master's dissertation in Mathematics at Durham University. Topic: applying neural networks to differential equations. Grade: 85/100.

A Machine Learning based system to detect semantics in the form of semantic label and suggest optimized alternatives for Python and C++ function-based code snippet.Semantic Detection is language agnostic with accuracy above 95% while optimization with accuracy above 85% for both Python and C++.

A mechanistic interpretability framework visualizing the Universal Approximation Theorem. It deconstructs Neural Networks into weighted ReLU basis functions to reveal how models construct complex non-linear topologies from piecewise linear segments.

This interactive web application demonstrates how neural networks can approximate virtually any continuous function, showcasing the Universal Approximation Theorem through real-time visualization.

🔍 Explore how Multi-Layer Perceptrons work by visualizing function approximation through Neural Basis Decomposition and Mechanistic Interpretability techniques.

A Biologically-Grounded Architecture for Artificial Intelligence

Implementasi forward pass Multi-Layer Perceptron (MLP) 1 hidden layer dari nol menggunakan numpy murni (tanpa framework deep learning).