Evolutionary Learning

Machine Learning Pong

A Pong experiment where neural-network agents improve through selection, mutation, and repeated matches.

JavaScript

p5.js

Machine learning

Why It Is Useful

The project is useful as a small, inspectable example of training behavior without directly programming the final strategy.

Problem It Solves

A paddle controller can be hand-coded, but the more interesting problem is whether a simple agent can discover useful behavior from repeated evaluation.

Real-World Applications

Optimization of controllers where a perfect rule set is not obvious.
Game AI experiments and automated playtesting.
Introductory reinforcement-learning and evolutionary-search demonstrations.

Concepts Used

Neural-network inputs and outputs
Fitness scoring
Mutation and selection
Agent evaluation loops

Solution Used

Agents observe game state and choose paddle movement through a small network.
Better performers are selected across generations.
Random mutation explores new strategies while repeated matches filter out weak behavior.