课时1：Why do we need machine learning

课时2：What are neural networks

课时3：Some simple models of neurons

课时4： A simple example of learning

课时5：Three types of learning

课时6：An overview of the main types of network architecture

课时7：Perceptrons

课时8：A geometrical view of perceptrons

课时9：Why the learning works

课时10：What perceptrons can not do

课时11：Learning the weights of a linear neuron

课时12：The error surface for a linear neuron

课时13：Learning the weights of a logistic output neuron

课时14：The backpropagation algorithm

课时15：How to use the derivatives computed by the backpropagation algorithm

课时16：Learning to predict the next word

课时17：A brief diversion into cognitive science

课时18：Another diversion_The softmax output function

课时19：Neuro-probabilistic language models

课时20：ways to deal with large number of possible outputs

课时21：Why object recognition is difficult

课时22：Ways to achieve viewpoint invariance

课时23：Convolutional neural networks for hand-written digit recognition

课时24：Convolutional neural networks for object recognition

课时25：Overview of mini-batch gradient descent

课时26：A bag of tricks for mini-batch descent

课时27：The momentum method

课时28：A separate, adaptive learning rate for each connection

课时29：rmsprop_divide the gradient

课时30：Modeling sequences_brief overview

课时31：Training RNNs with backpropagation

课时32：A toy example of training an RNN

课时33：Why it is difficul to train an RNN

课时34：Long term short term memory

课时35：Modeling character strings with multiplicative connections

课时36：Learning to predict the next character using HF

课时37：Echo state networks

课时38：Overview of ways to improve generalization

课时39：Limiting size of the weights

课时40：Using noise as a regularizer

课时41：Introduction to the bayesian approach

课时42：The bayesian interpretation of weight decay

课时43：MacKays quick and dirty method of fixing weight costs

课时44：Why it helps to combine models

课时45：Mixtures of experts

课时46：The idea of full bayesian learning

课时47：Making full bayesian learning practical

课时48：Dropout an efficient way to combine neural nets

课时49：Hopfield Nets

课时50：Dealing with spurious minima in hopfield nets

课时51：Hopfields Nets with hidden units

课时52：Using stochastic units to improve search

课时53：How a boltzmann machine models data

课时54：The boltzmann machine learning algorithm

课时55：More efficient ways to get the statistics

课时56：Restricted boltzmann machines

课时57：An example of contrastive divergence learning

课时58：RBMs for collaborative filtering

课时59：The ups and downs of backpropagation

课时60：Belief nets

课时61：The wake-sleep algorithm

课时62：Learning layers of features by stacking RBMs

课时63：Discriminative fine-tuning for DBNs

课时64：What happens during discriminative fine-tuning

课时65：Modeling real-valued data with an RBM

课时66：RBMs are infinite sigmoid belief nets

课时67：From principal components analysis to autoencoders

课时68：Deep Autoencoders

课时69：Deep autoencoders for document retrieval and visualization

课时70：Semantic hashing

课时71：Learning binary codes for image retrieval

课时72：Shallow autoencoders for pre-training

课时73：Learning a joint model of images and captions

课时74：Hierarchical coordinate frames

课时75：Bayesian optimization of neural network hyperparameters