Hinton Machine Learning and Neural Networks

Geoffrey Hinton, one of the founders of machine learning, gave a lecture and recording in 2012. The content of the course is distinguished by its profound wisdom and insight. The course provides an in-depth introduction to the neural network method in Machine Learning, the application of artificial neural networks in speech recognition and object recognition, image segmentation, modeling language and human movement, and its role in machine learning.

• Chapter 1：Why do we need machine learning (13 minutes and 14 seconds)

• Chapter 2：What are neural networks (8 minutes and 30 seconds)

• Chapter 3：Some simple models of neurons (8 minutes and 23 seconds)

• Chapter 4： A simple example of learning (5 minutes and 38 seconds)

• Chapter 5：Three types of learning (7 minutes and 37 seconds)

• Chapter 6：An overview of the main types of network architecture (7 minutes and 8 seconds)

• Chapter 7：Perceptrons (7 minutes and 56 seconds)

• Chapter 8：A geometrical view of perceptrons (6 minutes and 24 seconds)

• Chapter 9：Why the learning works (6 minutes and 2 seconds)

• Chapter 10：What perceptrons can not do (4 minutes and 38 seconds)

• Chapter 11：Learning the weights of a linear neuron (11 minutes and 55 seconds)

• Chapter 12：The error surface for a linear neuron (5 minutes and 3 seconds)

• Chapter 13：Learning the weights of a logistic output neuron (3 minutes and 56 seconds)

• Chapter 14：The backpropagation algorithm (11 minutes and 51 seconds)

• Chapter 15：How to use the derivatives computed by the backpropagation algorithm (9 minutes and 49 seconds)

• Chapter 16：Learning to predict the next word (12 minutes and 33 seconds)

• Chapter 17：A brief diversion into cognitive science (4 minutes and 26 seconds)

• Chapter 18：Another diversion_The softmax output function (7 minutes and 20 seconds)

• Chapter 19：Neuro-probabilistic language models (7 minutes and 52 seconds)

• Chapter 20：ways to deal with large number of possible outputs (12 minutes and 16 seconds)

• Chapter 21：Why object recognition is difficult (4 minutes and 40 seconds)

• Chapter 22：Ways to achieve viewpoint invariance (5 minutes and 58 seconds)

• Chapter 23：Convolutional neural networks for hand-written digit recognition (16 minutes and 1 seconds)

• Chapter 24：Convolutional neural networks for object recognition (17 minutes and 44 seconds)

• Chapter 25：Overview of mini-batch gradient descent (8 minutes and 22 seconds)

• Chapter 26：A bag of tricks for mini-batch descent (13 minutes and 15 seconds)

• Chapter 27：The momentum method (8 minutes and 42 seconds)

• Chapter 28：A separate, adaptive learning rate for each connection (5 minutes and 44 seconds)

• Chapter 29：rmsprop_divide the gradient (11 minutes and 38 seconds)

• Chapter 30：Modeling sequences_brief overview (17 minutes and 23 seconds)

• Chapter 31：Training RNNs with backpropagation (6 minutes and 23 seconds)

• Chapter 32：A toy example of training an RNN (6 minutes and 14 seconds)

• Chapter 33：Why it is difficul to train an RNN (7 minutes and 43 seconds)

• Chapter 34：Long term short term memory (9 minutes and 15 seconds)

• Chapter 35：Modeling character strings with multiplicative connections (14 minutes and 35 seconds)

• Chapter 36：Learning to predict the next character using HF (12 minutes and 24 seconds)

• Chapter 37：Echo state networks (9 minutes and 37 seconds)

• Chapter 38：Overview of ways to improve generalization (11 minutes and 44 seconds)

• Chapter 39：Limiting size of the weights (6 minutes and 22 seconds)

• Chapter 40：Using noise as a regularizer (7 minutes and 31 seconds)

• Chapter 41：Introduction to the bayesian approach (10 minutes and 49 seconds)

• Chapter 42：The bayesian interpretation of weight decay (10 minutes and 52 seconds)

• Chapter 43：MacKays quick and dirty method of fixing weight costs (3 minutes and 31 seconds)

• Chapter 44：Why it helps to combine models (13 minutes and 10 seconds)

• Chapter 45：Mixtures of experts (13 minutes and 15 seconds)

• Chapter 46：The idea of full bayesian learning (7 minutes and 27 seconds)

• Chapter 47：Making full bayesian learning practical (6 minutes and 44 seconds)

• Chapter 48：Dropout an efficient way to combine neural nets (8 minutes and 35 seconds)

• Chapter 49：Hopfield Nets (13 minutes and 1 seconds)

• Chapter 50：Dealing with spurious minima in hopfield nets (11 minutes and 2 seconds)

• Chapter 51：Hopfields Nets with hidden units (9 minutes and 39 seconds)

• Chapter 52：Using stochastic units to improve search (10 minutes and 24 seconds)

• Chapter 53：How a boltzmann machine models data (11 minutes and 44 seconds)

• Chapter 54：The boltzmann machine learning algorithm (12 minutes and 15 seconds)

• Chapter 55：More efficient ways to get the statistics (14 minutes and 48 seconds)

• Chapter 56：Restricted boltzmann machines (10 minutes and 54 seconds)

• Chapter 57：An example of contrastive divergence learning (7 minutes and 14 seconds)

• Chapter 58：RBMs for collaborative filtering (8 minutes and 16 seconds)

• Chapter 59：The ups and downs of backpropagation (9 minutes and 53 seconds)

• Chapter 60：Belief nets (12 minutes and 35 seconds)

• Chapter 61：The wake-sleep algorithm (13 minutes and 14 seconds)

• Chapter 62：Learning layers of features by stacking RBMs (17 minutes and 34 seconds)

• Chapter 63：Discriminative fine-tuning for DBNs (9 minutes and 40 seconds)

• Chapter 64：What happens during discriminative fine-tuning (8 minutes and 39 seconds)

• Chapter 65：Modeling real-valued data with an RBM (9 minutes and 56 seconds)

• Chapter 66：RBMs are infinite sigmoid belief nets (17 minutes and 11 seconds)

• Chapter 67：From principal components analysis to autoencoders (7 minutes and 57 seconds)

• Chapter 68：Deep autoencoders (4 minutes and 10 seconds)

• Chapter 69：Deep autoencoders for document retrieval and visualization (8 minutes and 19 seconds)

• Chapter 70：Semantic hashing (8 minutes and 50 seconds)

• Chapter 71：Learning binary codes for image retrieval (9 minutes and 37 seconds)

• Chapter 72：Shallow autoencoders for pre-training (7 minutes and 2 seconds)

• Chapter 73：Learning a joint model of images and captions (9 minutes and 5 seconds)

• Chapter 74：Hierarchical coordinate frames (9 minutes and 40 seconds)

• Chapter 75：Bayesian optimization of neural network hyperparameters (13 minutes and 29 seconds)