Introduction to Neural Networks

on YouTube ↗25m

" summary

Builds a working Python neural net implemented from scratch using sigmoid neurons, quadratic cost and stochastic gradient descent with backpropagation; trains on MNIST (40k train / 10k test) with a 784-30-10 architecture over ~30 epochs and reports 9,330/10,000 correct (~93% accuracy). Explains why sigmoids enable gradient-based learning, shows the backprop equations and pseudo-code, and contrasts the toy implementation with libraries like TensorFlow and Keras that reach ~98%.

AI-written summary · about

" chapters (13)

1. 0:00Intro
2. 1:31PERCEPTRON
3. 4:06NETWORK LAYERS
4. 5:11SIGMOID NEURON
5. 5:50CLASSIFYING THE MNIST DATA SET
6. 6:49MNIST NETWORK DESIGN
7. 8:11TRAINING THE NETWORK
8. 10:26LEARNING - MINIMISING COST USING
9. 13:56NOTATION
10. 14:54BACKPROPAGATION EQUATIONS
11. 16:11BACKPROPAGATION PSEUDO
12. 17:57STOCHASTIC GRADIENT DESCENT
13. 22:48NN LIBRARIES FOR PYTHON

" related talks

• 43mA Gentle Introduction to Neural NetworksTariq Rashid · Agile on the Beach 2018
• 01h 25mBuilding & Optimization of Neural Networks from Scratch with NumpyM. Marjanovic · DSC Europe 23
• 01h 50mNeural Networks: WalkthroughKatharine Beaumont · Devoxx Belgium 2017
• 07mNeural Networks Demystified: Part 3MLconf SF 2015
• 01h 04mWriting a Neural Net from ScratchJoe Albahari · NDC Sydney 2018
• 44mMachine Learning ExposedJames Weaver · Codemotion Amsterdam 2017
• 58mNeural Networks: A PrimerRishal Hurbans · NDC Sydney 2018
• 33mLearn Neural Networks With Go - Not Math!Ellen Körbes · DevDays Europe 2019