Deep learning----------Deep Belief Networks

原创 2013年12月04日 21:04:25


中文简介: 

DBN是2006年提出的一种概率生成模型, 由多个限制玻尔兹曼机(RBM)[3]堆栈而成:

由Hinton在奋斗

  在训练时, Hinton采用了逐层无监督的方法来学习参数。首先把数据向量x和第一层隐藏层作为一个RBM, 训练出这个RBM的参数(连接x和h1的权重, x和h1各个节点的偏置等等), 然后固定这个RBM的参数, 把h1视作可见向量, 把h2视作隐藏向量, 训练第二个RBM, 得到其参数, 然后固定这些参数, 训练h2和h3构成的RBM, 具体的训练算法如下:

  

  上图最右边就是最终训练得到的生成模型:

  

  用公式表示为:

  

3. 利用DBN进行有监督学习

  在使用上述的逐层无监督方法学得节点之间的权重以及节点的偏置之后(亦即初始化), 可以在DBN的最顶层再加一层, 来表示我们希望得到的输出, 然后计算模型得到的输出和希望得到的输出之间的误差, 利用后向反馈的方法来进一步优化之前设置的初始权重。因为我们已经使用逐层无监督方法来初始化了权重值, 使其比较接近最优值, 解决了之前多层神经网络训练时存在的问题, 能够得到很好的效果。


The explanation in English and the corresponding code

Deep Belief Networks

[Hinton06] showed that RBMs can be stacked and trained in a greedy manner to form so-called Deep Belief Networks (DBN). DBNs are graphical models which learn to extract a deep hierarchical representation of the training data. They model the joint distribution between observed vector x and the \ell hidden layers h^k as follows:

(1)P(x, h^1, \ldots, h^{\ell}) = \left(\prod_{k=0}^{\ell-2} P(h^k|h^{k+1})\right) P(h^{\ell-1},h^{\ell})

where x=h^0P(h^{k-1} | h^k) is a conditional distribution for the visible units conditioned on the hidden units of the RBM at level k, and P(h^{\ell-1}, h^{\ell}) is the visible-hidden joint distribution in the top-level RBM. This is illustrated in the figure below.

_images/DBN3.png

The principle of greedy layer-wise unsupervised training can be applied to DBNs with RBMs as the building blocks for each layer[Hinton06][Bengio07]. The process is as follows:

1. Train the first layer as an RBM that models the raw input x =h^{(0)} as its visible layer.

2. Use that first layer to obtain a representation of the input that will be used as data for the second layer. Two common solutions exist. This representation can be chosen as being the mean activations p(h^{(1)}=1|h^{(0)}) or samples of p(h^{(1)}|h^{(0)}).

3. Train the second layer as an RBM, taking the transformed data (samples or mean activations) as training examples (for the visible layer of that RBM).

4. Iterate (2 and 3) for the desired number of layers, each time propagating upward either samples or mean values.

5. Fine-tune all the parameters of this deep architecture with respect to a proxy for the DBN log- likelihood, or with respect to a supervised training criterion (after adding extra learning machinery to convert the learned representation into supervised predictions, e.g. a linear classifier).

In this tutorial, we focus on fine-tuning via supervised gradient descent. Specifically, we use a logistic regression classifier to classify the input x based on the output of the last hidden layer h^{(l)} of the DBN. Fine-tuning is then performed via supervised gradient descent of the negative log-likelihood cost function. Since the supervised gradient is only non-null for the weights and hidden layer biases of each layer (i.e. null for the visible biases of each RBM), this procedure is equivalent to initializing the parameters of a deep MLP with the weights and hidden layer biases obtained with the unsupervised training strategy.

Justifying Greedy-Layer Wise Pre-Training

Why does such an algorithm work ? Taking as example a 2-layer DBN with hidden layers h^{(1)} and h^{(2)} (with respective weight parameters W^{(1)} and W^{(2)}), [Hinton06] established (see also Bengio09]_ for a detailed derivation) that \logp(x) can be rewritten as,

(2)\log p(x) = &KL(Q(h^{(1)}|x)||p(h^{(1)}|x)) + H_{Q(h^{(1)}|x)} + \\            &\sum_h Q(h^{(1)}|x)(\log p(h^{(1)}) + \log p(x|h^{(1)})).

KL(Q(h^{(1)}|x) || p(h^{(1)}|x)) represents the KL divergence between the posterior Q(h^{(1)}|x) of the first RBM if it were standalone, and the probability p(h^{(1)}|x) for the same layer but defined by the entire DBN (i.e. taking into account the prior p(h^{(1)},h^{(2)}) defined by the top-level RBM). H_{Q(h^{(1)}|x)} is the entropy of the distribution Q(h^{(1)}|x).

It can be shown that if we initialize both hidden layers such that W^{(2)}={W^{(1)}}^TQ(h^{(1)}|x)=p(h^{(1)}|x) and the KL divergence term is null. If we learn the first level RBM and then keep its parameters W^{(1)} fixed, optimizing Eq. (2) with respect to W^{(2)} can thus only increase the likelihood p(x).

Also, notice that if we isolate the terms which depend only on W^{(2)}, we get:

\sum_h Q(h^{(1)}|x)p(h^{(1)})

Optimizing this with respect to W^{(2)} amounts to training a second-stage RBM, using the output of Q(h^{(1)}|x) as the training distribution, when x is sampled from the training distribution for the first RBM.

Implementation

To implement DBNs in Theano, we will use the class defined in the Restricted Boltzmann Machines (RBM) tutorial. One can also observe that the code for the DBN is very similar with the one for SdA, because both involve the principle of unsupervised layer-wise pre-training followed by supervised fine-tuning as a deep MLP. The main difference is that we use the RBM class instead of the dA class.

We start off by defining the DBN class which will store the layers of the MLP, along with their associated RBMs. Since we take the viewpoint of using the RBMs to initialize an MLP, the code will reflect this by seperating as much as possible the RBMs used to initialize the network and the MLP used for classification.

class DBN(object):

    def __init__(self, numpy_rng, theano_rng=None, n_ins=784,
                hidden_layers_sizes=[500, 500], n_outs=10):
        """This class is made to support a variable number of layers.

        :type numpy_rng: numpy.random.RandomState
        :param numpy_rng: numpy random number generator used to draw initial
                weights

        :type theano_rng: theano.tensor.shared_randomstreams.RandomStreams
        :param theano_rng: Theano random generator; if None is given one is
                       generated based on a seed drawn from `rng`

        :type n_ins: int
        :param n_ins: dimension of the input to the DBN

        :type n_layers_sizes: list of ints
        :param n_layers_sizes: intermediate layers size, must contain
                           at least one value

        :type n_outs: int
        :param n_outs: dimension of the output of the network
        """

        self.sigmoid_layers = []
        self.rbm_layers = []
        self.params = []
        self.n_layers = len(hidden_layers_sizes)

        assert self.n_layers > 0

        if not theano_rng:
            theano_rng = RandomStreams(numpy_rng.randint(2 ** 30))

        # allocate symbolic variables for the data
        self.x = T.matrix('x')  # the data is presented as rasterized images
        self.y = T.ivector('y')  # the labels are presented as 1D vector of
                                 # [int] labels

self.sigmoid_layers will store the feed-forward graphs which together form the MLP, while self.rbm_layers will store the RBMs used to pretrain each layer of the MLP.

Next step, we construct n_layers sigmoid layers (we use the SigmoidalLayer class introduced in Multilayer Perceptron, with the only modification that we replaced the non-linearity from tanh to the logistic function s(x) = \frac{1}{1+e^{-x}}) and n_layers RBMs, where n_layersis the depth of our model. We link the sigmoid layers such that they form an MLP, and construct each RBM such that they share the weight matrix and the hidden bias with its corresponding sigmoid layer.

for i in xrange(self.n_layers):
    # construct the sigmoidal layer

    # the size of the input is either the number of hidden units of the
    # layer below or the input size if we are on the first layer
    if i == 0:
        input_size = n_ins
    else:
        input_size = hidden_layers_sizes[i - 1]

    # the input to this layer is either the activation of the hidden
    # layer below or the input of the DBN if you are on the first layer
    if i == 0:
        layer_input = self.x
    else:
        layer_input = self.sigmoid_layers[-1].output

    sigmoid_layer = HiddenLayer(rng=numpy_rng,
                                input=layer_input,
                                n_in=input_size,
                                n_out=hidden_layers_sizes[i],
                                activation=T.nnet.sigmoid)

    # add the layer to our list of layers
    self.sigmoid_layers.append(sigmoid_layer)

    # its arguably a philosophical question...  but we are going to only declare that
    # the parameters of the sigmoid_layers are parameters of the DBN. The visible
    # biases in the RBM are parameters of those RBMs, but not of the DBN.
    self.params.extend(sigmoid_layer.params)

    # Construct an RBM that shared weights with this layer
    rbm_layer = RBM(numpy_rng=numpy_rng,
                    theano_rng=theano_rng,
                    input=layer_input,
                    n_visible=input_size,
                    n_hidden=hidden_layers_sizes[i],
                    W=sigmoid_layer.W,
                    hbias=sigmoid_layer.b)
    self.rbm_layers.append(rbm_layer)

All that is left is to stack one last logistic regression layer in order to form an MLP. We will use the LogisticRegression class introduced in Classifying MNIST digits using Logistic Regression.

# We now need to add a logistic layer on top of the MLP
self.logLayer = LogisticRegression(
                 input=self.sigmoid_layers[-1].output,
                 n_in=hidden_layers_sizes[-1], n_out=n_outs)
self.params.extend(self.logLayer.params)

# construct a function that implements one step of fine-tuning compute
# the cost for second phase of training, defined as the negative log
# likelihood of the logistic regression (output) layer
self.finetune_cost = self.logLayer.negative_log_likelihood(self.y)

# compute the gradients with respect to the model parameters
# symbolic variable that points to the number of errors made on the
# minibatch given by self.x and self.y
self.errors = self.logLayer.errors(self.y)

The class also provides a method which generates training functions for each of the RBMs. They are returned as a list, where element i is a function which implements one step of training for the RBM at layer i.

def pretraining_functions(self, train_set_x, batch_size, k):
    ''' Generates a list of functions, for performing one step of gradient descent at a
    given layer. The function will require as input the minibatch index, and to train an
    RBM you just need to iterate, calling the corresponding function on all minibatch
    indexes.

    :type train_set_x: theano.tensor.TensorType
    :param train_set_x: Shared var. that contains all datapoints used for training the RBM
    :type batch_size: int
    :param batch_size: size of a [mini]batch
    :param k: number of Gibbs steps to do in CD-k / PCD-k
    '''

    # index to a [mini]batch
    index = T.lscalar('index')  # index to a minibatch

In order to be able to change the learning rate during training, we associate a Theano variable to it that has a default value.

learning_rate = T.scalar('lr')  # learning rate to use

# number of batches
n_batches = train_set_x.get_value(borrow=True).shape[0] / batch_size
# begining of a batch, given `index`
batch_begin = index * batch_size
# ending of a batch given `index`
batch_end = batch_begin + batch_size

pretrain_fns = []
for rbm in self.rbm_layers:

    # get the cost and the updates list
    # using CD-k here (persisent=None) for training each RBM.
    # TODO: change cost function to reconstruction error
    cost, updates = rbm.cd(learning_rate, persistent=None, k)

    # compile the Theano function; check if k is also a Theano
    # variable, if so added to the inputs of the function
    if isinstance(k, theano.Variable):
        inputs = [index, theano.Param(learning_rate, default=0.1), k]
    else:
        inputs =  index, theano.Param(learning_rate, default=0.1)]
    fn = theano.function(inputs=inputs,
                         outputs=cost,
                         updates=updates,
                         givens={self.x: train_set_x[batch_begin:
                                                     batch_end]})
    # append `fn` to the list of functions
    pretrain_fns.append(fn)

return pretrain_fns

Now any function pretrain_fns[i] takes as arguments index and optionally lr – the learning rate. Note that the names of the parameters are the names given to the Theano variables (e.g. lr) when they are constructed and not the name of the python variables (e.g. learning_rate). Keep this in mind when working with Theano. Optionally, if you provide k (the number of Gibbs steps to perform in CD or PCD) this will also become an argument of your function.

In the same fashion, the DBN class includes a method for building the functions required for finetuning ( a train_model, avalidate_model and a test_model function).

def build_finetune_functions(self, datasets, batch_size, learning_rate):
    '''Generates a function `train` that implements one step of finetuning, a function
    `validate` that computes the error on a batch from the validation set, and a function
    `test` that computes the error on a batch from the testing set

    :type datasets: list of pairs of theano.tensor.TensorType
    :param datasets: It is a list that contain all the datasets;  the has to contain three
    pairs, `train`, `valid`, `test` in this order, where each pair is formed of two Theano
    variables, one for the datapoints, the other for the labels
    :type batch_size: int
    :param batch_size: size of a minibatch
    :type learning_rate: float
    :param learning_rate: learning rate used during finetune stage
    '''

    (train_set_x, train_set_y) = datasets[0]
    (valid_set_x, valid_set_y) = datasets[1]
    (test_set_x, test_set_y) = datasets[2]

    # compute number of minibatches for training, validation and testing
    n_valid_batches = valid_set_x.get_value(borrow=True).shape[0] / batch_size
    n_test_batches  = test_set_x.get_value(borrow=True).shape[0]  / batch_size

    index   = T.lscalar('index')    # index to a [mini]batch

    # compute the gradients with respect to the model parameters
    gparams = T.grad(self.finetune_cost, self.params)

    # compute list of fine-tuning updates
    updates = []
    for param, gparam in zip(self.params, gparams):
        updates.append((param, param - gparam * learning_rate))

    train_fn = theano.function(inputs=[index],
          outputs= self.finetune_cost,
          updates=updates,
          givens={
            self.x: train_set_x[index * batch_size: (index + 1) * batch_size],
            self.y: train_set_y[index * batch_size: (index + 1) * batch_size]})

    test_score_i = theano.function([index], self.errors,
             givens={
               self.x: test_set_x[index * batch_size: (index + 1) * batch_size],
               self.y: test_set_y[index * batch_size: (index + 1) * batch_size]})

    valid_score_i = theano.function([index], self.errors,
          givens={
             self.x: valid_set_x[index * batch_size: (index + 1) * batch_size],
             self.y: valid_set_y[index * batch_size: (index + 1) * batch_size]})

    # Create a function that scans the entire validation set
    def valid_score():
        return [valid_score_i(i) for i in xrange(n_valid_batches)]

    # Create a function that scans the entire test set
    def test_score():
        return [test_score_i(i) for i in xrange(n_test_batches)]

    return train_fn, valid_score, test_score

Note that the returned valid_score and test_score are not Theano functions, but rather Python functions. These loop over the entire validation set and the entire test set to produce a list of the losses obtained over these sets.

Putting it all together

The few lines of code below constructs the deep belief network :

numpy_rng = numpy.random.RandomState(123)
print '... building the model'
# construct the Deep Belief Network
dbn = DBN(numpy_rng=numpy_rng, n_ins=28 * 28,
          hidden_layers_sizes=[1000, 1000, 1000],
          n_outs=10)

There are two stages in training this network: (1) a layer-wise pre-training and (2) a fine-tuning stage.

For the pre-training stage, we loop over all the layers of the network. For each layer, we use the compiled theano function which determines the input to the i-th level RBM and performs one step of CD-k within this RBM. This function is applied to the training set for a fixed number of epochs given by pretraining_epochs.

#########################
# PRETRAINING THE MODEL #
#########################
print '... getting the pretraining functions'
# We are using CD-1 here
pretraining_fns = dbn.pretraining_functions(
        train_set_x=train_set_x,
        batch_size=batch_size,
        k=k)

print '... pre-training the model'
start_time = time.clock()
## Pre-train layer-wise
for i in xrange(dbn.n_layers):
    # go through pretraining epochs
    for epoch in xrange(pretraining_epochs):
        # go through the training set
        c = []
        for batch_index in xrange(n_train_batches):
            c.append(pretraining_fns[i](index=batch_index,
                                        lr=pretrain_lr))
        print 'Pre-training layer %i, epoch %d, cost '%(i,epoch),numpy.mean(c)

end_time = time.clock()

The fine-tuning loop is very similar to the one in the Multilayer Perceptron tutorial, the only difference being that we now use the functions given by build_finetune_functions.

Running the Code

The user can run the code by calling:

python code/DBN.py

With the default parameters, the code runs for 100 pre-training epochs with mini-batches of size 10. This corresponds to performing 500,000 unsupervised parameter updates. We use an unsupervised learning rate of 0.01, with a supervised learning rate of 0.1. The DBN itself consists of three hidden layers with 1000 units per layer. With early-stopping, this configuration achieved a minimal validation error of 1.27 with corresponding test error of 1.34 after 46 supervised epochs.

On an Intel(R) Xeon(R) CPU X5560 running at 2.80GHz, using a multi-threaded MKL library (running on 4 cores), pretraining took 615 minutes with an average of 2.05 mins/(layer * epoch). Fine-tuning took only 101 minutes or approximately 2.20 mins/epoch.

Hyper-parameters were selected by optimizing on the validation error. We tested unsupervised learning rates in \{10^{-1}, ..., 10^{-5}\}and supervised learning rates in \{10^{-1}, ..., 10^{-4}\}. We did not use any form of regularization besides early-stopping, nor did we optimize over the number of pretraining updates.

Tips and Tricks

One way to improve the running time of your code (given that you have sufficient memory available), is to compute the representation of the entire dataset at layer i in a single pass, once the weights of the i-1-th layers have been fixed. Namely, start by training your first layer RBM. Once it is trained, you can compute the hidden units values for every example in the dataset and store this as a new dataset which is used to train the 2nd layer RBM. Once you trained the RBM for layer 2, you compute, in a similar fashion, the dataset for layer 3 and so on. This avoids calculating the intermediate (hidden layer) representations, pretraining_epochstimes at the expense of increased memory usage.



相关文章推荐

Neural Networks and Deep Learning

  • 2017年11月10日 10:40
  • 3.46MB
  • 下载

Neural Networks and Deep Learning学习笔记ch6 - 深度学习

在上一章节中讨论了为什么深度神经网络网络会很难训练。这一节就谈谈是如何克服那些困难的,开始正式讨论深度学习(Deep Learning)。 希望通过这样一步一步地介绍能够对深度学习有一个比较清楚的脉...

deep learning 卷积神经网络的实现(Convolution Neural Networks)

本节将会讲到卷积神经网络的实现。说到卷积神经网络,在图像识别和目标检测方面已经取得了不错的效果,为什么要叫做卷积神经网络呢?主要是因为在特征提取的时候,输入图像会通过卷积核对原始图像进行特征抽取,然后...

Deep Learning读书笔记(一):Reducing the Dimensionality of Data with Neural Networks

这是发表在Science上的一篇文章,是Deep Learning的开山之作,同样也是我读的第一篇文章,我的第一篇读书笔记也从这开始吧。        文章的主要工作是数据的降维,等于说这里使用深度...
  • tuqinag
  • tuqinag
  • 2014年09月07日 20:16
  • 1933
内容举报
返回顶部
收藏助手
不良信息举报
您举报文章:Deep learning----------Deep Belief Networks
举报原因:
原因补充:

(最多只允许输入30个字)