Neural Networks with RubyFANN
summary
Uses the fann C library via the RubyFANN gem to build a spam detector and tic-tac-toe agents. Spam model ingests 82 inputs from csmining.org and trains to near‑random performance, exposing mislabels or bad feature encodings. Tic‑tac‑toe compares a board-state-only network (trained on ~27 examples, weak) versus a board-state-plus-move network trained by self-play that wins consistently, showing that data volume, encoding and activation choices actually determine success.
AI-written summary · about
chapters (28)
- 0:00Introduction
- 0:23About Ethan
- 0:50Agenda
- 2:04What are Neural Networks
- 2:23Identity Function
- 3:15Experimentation
- 5:43Input in Training
- 7:41Bias
- 8:54Inference Engine
- 10:15Specific Problems
- 10:39RubyFANN
- 11:48Experiments
- 13:10Inputs
- 14:15Train Data
- 15:00Create Network
- 15:45Train Network
- 16:41Error Numbers
- 17:36Testing
- 18:13Neural Networks vs Bayesian Networks
- 19:47Should Neural Networks Train
- 20:35Neural Connections
- 21:11TicTacToe
- 22:33csv
- 23:17Results
- 24:04Tournament Results
- 25:50Activation Functions
- 26:44Other Applications
- 27:15Input Grid
related talks
42mDōmo arigatō, Mr. Roboto: Machine Learning with Ruby
29mTest Driven Neural Networks with Ruby
44mFast Neural Networks… a no brainer!
39mArtificial Intelligence vs. Natural Stupidity
51mMachine Learning with Ruby
35mIntelligent Ruby: Getting Started with Machine Learning
13mLet’s run a tiny chess neural network by hand!
20mFANCI : Feature-based Automated NXDomain Classification and Intelligence