Dive into the world of neural networks with this comprehensive conference talk that guides you through coding a neural net for image recognition from scratch using C#. Learn about gradient descent, activation functions, and backpropagation as you build a functional neural network without relying on external libraries. Explore traditional programming concepts, understand the importance of training neural networks, and discover key components such as perceptrons, NAND gates, and logistic sigmoid functions. Gain insights into error calculation, loss minimization, and gradient descent techniques. Examine neural network architecture, including convolutional networks, and determine optimal learning rates. Work entirely in LINQPad with practical, hands-on examples that you can keep and reference later.
Overview
Syllabus
Introduction
Traditional Programming
Why not use a library
Training to neural networks
Perceptron
State
Transient State
NAND Gate
Basic NAND Gate
Logistic Sigmoid
Runoff
Error
Loss
Minimize
Gradient Descent
Summary
Neural Network Architecture
Convolutional Network
Optimal Learning Rate
Taught by
NDC Conferences
