lenet.txt

.. _lenet:

Convolutional Neural Networks (LeNet)
=====================================

.. note::
    This section assumes the reader has already read through :doc:`logreg` and
    :doc:`mlp`. Additionally, it uses the following new Theano functions and concepts:
    `T.tanh`_, `shared variables`_, `basic arithmetic ops`_, `T.grad`_,
    `floatX`_, `downsample`_ , `conv2d`_, `dimshuffle`_. If you intend to run the
    code on GPU also read `GPU`_.

    To run this example on a GPU, you need a good GPU. It needs
    at least 1GB of GPU RAM.  More may be required if your monitor is
    connected to the GPU.
    
    When the GPU is connected to the monitor, there is a limit
    of a few seconds for each GPU function call. This is needed as
    current GPUs can't be used for the monitor while doing
    computation. Without this limit, the screen would freeze
    for too long and make it look as if the computer froze. 
    This example hits this limit with medium-quality GPUs. When the
    GPU isn't connected to a monitor, there is no time limit. You can
    lower the batch size to fix the time out problem.

.. _T.tanh: http://deeplearning.net/software/theano/tutorial/examples.html?highlight=tanh

.. _shared variables: http://deeplearning.net/software/theano/tutorial/examples.html#using-shared-variables

.. _basic arithmetic ops: http://deeplearning.net/software/theano/tutorial/adding.html#adding-two-scalars

.. _T.grad: http://deeplearning.net/software/theano/tutorial/examples.html#computing-gradients

.. _floatX: http://deeplearning.net/software/theano/library/config.html#config.floatX

.. _GPU: http://deeplearning.net/software/theano/tutorial/using_gpu.html

.. _downsample: http://deeplearning.net/software/theano/library/tensor/signal/downsample.html

.. _conv2d: http://deeplearning.net/software/theano/library/tensor/signal/conv.html#module-conv

.. _dimshuffle: http://deeplearning.net/software/theano/library/tensor/basic.html#tensor._tensor_py_operators.dimshuffle

.. note::
    The code for this section is available for download `here`_ and the `3wolfmoon image`_

.. _here: http://deeplearning.net/tutorial/code/convolutional_mlp.py

.. _3wolfmoon image: https://raw.githubusercontent.com/lisa-lab/DeepLearningTutorials/master/doc/images/3wolfmoon.jpg


Motivation
++++++++++

Convolutional Neural Networks (CNN) are biologically-inspired variants of MLPs.
From Hubel and Wiesel's early work on the cat's visual cortex [Hubel68]_, we
know the visual cortex contains a complex arrangement of cells. These cells are
sensitive to small sub-regions of the visual field, called a *receptive
field*. The sub-regions are tiled to cover the entire visual field. These
cells act as local filters over the input space and are well-suited to exploit
the strong spatially local correlation present in natural images.

Additionally, two basic cell types have been identified: Simple cells respond
maximally to specific edge-like patterns within their receptive field. Complex
cells have larger receptive fields and are locally invariant to the exact
position of the pattern.

The animal visual cortex being the most powerful visual processing system in
existence, it seems natural to emulate its behavior. Hence, many
neurally-inspired models can be found in the literature. To name a few: the
NeoCognitron [Fukushima]_, HMAX [Serre07]_ and LeNet-5 [LeCun98]_, which will
be the focus of this tutorial.


Sparse Connectivity
+++++++++++++++++++

CNNs exploit spatially-local correlation by enforcing a local connectivity
pattern between neurons of adjacent layers. In other words, the inputs of
hidden units in layer **m** are from a subset of units in layer **m-1**, units
that have spatially contiguous receptive fields. We can illustrate this
graphically as follows:

.. figure:: images/sparse_1D_nn.png
    :align: center

Imagine that layer **m-1** is the input retina. In the above figure, units in
layer **m** have receptive fields of width 3 in the input retina and are thus
only connected to 3 adjacent neurons in the retina layer. Units in layer
**m+1** have a similar connectivity with the layer below. We say that their
receptive field with respect to the layer below is also 3, but their receptive
field with respect to the input is larger (5). Each unit is unresponsive to
variations outside of its receptive field with respect to the retina. The
architecture thus ensures that the learnt "filters" produce the strongest
response to a spatially local input pattern.

However, as shown above, stacking many such layers leads to (non-linear)
"filters" that become increasingly "global" (i.e. responsive to a larger region
of pixel space). For example, the unit in hidden layer **m+1** can encode a
non-linear feature of width 5 (in terms of pixel space).


Shared Weights
++++++++++++++

In addition, in CNNs, each filter :math:`h_i` is replicated across the entire
visual field. These replicated units share the same parameterization (weight
vector and bias) and form a *feature map*.

.. figure:: images/conv_1D_nn.png
    :align: center

In the above figure, we show 3 hidden units belonging to the same feature map.
Weights of the same color are shared---constrained to be identical. Gradient
descent can still be used to learn such shared parameters, with only a small
change to the original algorithm. The gradient of a shared weight is simply the
sum of the gradients of the parameters being shared.

Replicating units in this way allows for features to be detected *regardless
of their position in the visual field.* Additionally, weight sharing increases
learning efficiency by greatly reducing the number of free parameters being
learnt. The constraints on the model enable CNNs to achieve better
generalization on vision problems.


Details and Notation
++++++++++++++++++++

A feature map is obtained by repeated application of a function across
sub-regions of the entire image, in other words, by *convolution* of the
input image with a linear filter, adding a bias term and then applying a
non-linear function. If we denote the k-th feature map at a given layer as
:math:`h^k`, whose filters are determined by the weights :math:`W^k` and bias
:math:`b_k`, then the feature map :math:`h^k` is obtained as follows (for
:math:`tanh` non-linearities):

.. math::
    h^k_{ij} = \tanh ( (W^k * x)_{ij} + b_k ).

.. Note::
    Recall the following definition of convolution for a 1D signal.
    :math:`o[n] = f[n]*g[n] = \sum_{u=-\infty}^{\infty} f[u] g[n-u] = \sum_{u=-\infty}^{\infty} f[n-u] g[u]`.

    This can be extended to 2D as follows:
    :math:`o[m,n] = f[m,n]*g[m,n] = \sum_{u=-\infty}^{\infty} \sum_{v=-\infty}^{\infty} f[u,v] g[m-u,n-v]`.

To form a richer representation of the data, each hidden layer is composed of
*multiple* feature maps, :math:`\{h^{(k)}, k=0..K\}`. The weights :math:`W` of
a hidden layer can be represented in a 4D tensor containing elements for every
combination of destination feature map, source feature map, source vertical
position, and source horizontal position. The biases :math:`b` can be
represented as a vector containing one element for every destination feature
map. We illustrate this graphically as follows:

.. figure:: images/cnn_explained.png
    :align: center

    **Figure 1**: example of a convolutional layer

The figure shows two layers of a CNN. **Layer m-1** contains four feature maps.
**Hidden layer m** contains two feature maps (:math:`h^0` and :math:`h^1`).
Pixels (neuron outputs) in :math:`h^0` and :math:`h^1` (outlined as blue and
red squares) are computed from pixels of layer (m-1) which fall within their
2x2 receptive field in the layer below (shown as colored rectangles). Notice
how the receptive field spans all four input feature maps. The weights
:math:`W^0` and :math:`W^1` of :math:`h^0` and :math:`h^1` are thus 3D weight
tensors. The leading dimension indexes the input feature maps, while the other
two refer to the pixel coordinates.

Putting it all together, :math:`W^{kl}_{ij}` denotes the weight connecting
each pixel of the k-th feature map at layer m, with the pixel at coordinates
(i,j) of the l-th feature map of layer (m-1).


The Convolution Operator
++++++++++++++++++++++++

ConvOp is the main workhorse for implementing a convolutional layer in Theano.
ConvOp is used by ``theano.tensor.signal.conv2d``, which takes two symbolic inputs:


* a 4D tensor corresponding to a mini-batch of input images. The shape of the
  tensor is as follows: [mini-batch size, number of input feature maps, image
  height, image width].

* a 4D tensor corresponding to the weight matrix :math:`W`. The shape of the
  tensor is: [number of feature maps at layer m, number of feature maps at
  layer m-1, filter height, filter width]


Below is the Theano code for implementing a convolutional layer similar to the
one of Figure 1. The input consists of 3 features maps (an RGB color image) of size
120x160. We use two convolutional filters with 9x9 receptive fields.

.. code-block:: python

        import theano
        from theano import tensor as T
        from theano.tensor.nnet import conv

        import numpy

        rng = numpy.random.RandomState(23455)

        # instantiate 4D tensor for input
        input = T.tensor4(name='input')

        # initialize shared variable for weights.
        w_shp = (2, 3, 9, 9)
        w_bound = numpy.sqrt(3 * 9 * 9)
        W = theano.shared( numpy.asarray(
                    rng.uniform(
                        low=-1.0 / w_bound,
                        high=1.0 / w_bound,
                        size=w_shp),
                    dtype=input.dtype), name ='W')

        # initialize shared variable for bias (1D tensor) with random values
        # IMPORTANT: biases are usually initialized to zero. However in this
        # particular application, we simply apply the convolutional layer to
        # an image without learning the parameters. We therefore initialize
        # them to random values to "simulate" learning.
        b_shp = (2,)
        b = theano.shared(numpy.asarray(
                    rng.uniform(low=-.5, high=.5, size=b_shp),
                    dtype=input.dtype), name ='b')

        # build symbolic expression that computes the convolution of input with filters in w
        conv_out = conv.conv2d(input, W)

        # build symbolic expression to add bias and apply activation function, i.e. produce neural net layer output
        # A few words on ``dimshuffle`` :
        #   ``dimshuffle`` is a powerful tool in reshaping a tensor;
        #   what it allows you to do is to shuffle dimension around
        #   but also to insert new ones along which the tensor will be
        #   broadcastable;
        #   dimshuffle('x', 2, 'x', 0, 1)
        #   This will work on 3d tensors with no broadcastable
        #   dimensions. The first dimension will be broadcastable,
        #   then we will have the third dimension of the input tensor as
        #   the second of the resulting tensor, etc. If the tensor has
        #   shape (20, 30, 40), the resulting tensor will have dimensions
        #   (1, 40, 1, 20, 30). (AxBxC tensor is mapped to 1xCx1xAxB tensor)
        #   More examples:
        #    dimshuffle('x') -> make a 0d (scalar) into a 1d vector
        #    dimshuffle(0, 1) -> identity
        #    dimshuffle(1, 0) -> inverts the first and second dimensions
        #    dimshuffle('x', 0) -> make a row out of a 1d vector (N to 1xN)
        #    dimshuffle(0, 'x') -> make a column out of a 1d vector (N to Nx1)
        #    dimshuffle(2, 0, 1) -> AxBxC to CxAxB
        #    dimshuffle(0, 'x', 1) -> AxB to Ax1xB
        #    dimshuffle(1, 'x', 0) -> AxB to Bx1xA
        output = T.nnet.sigmoid(conv_out + b.dimshuffle('x', 0, 'x', 'x'))

        # create theano function to compute filtered images
        f = theano.function([input], output)


Let's have a little bit of fun with this...

.. code-block:: python

        import numpy
        import pylab
        from PIL import Image

        # open random image of dimensions 639x516
        img = Image.open(open('doc/images/3wolfmoon.jpg'))
        # dimensions are (height, width, channel)
        img = numpy.asarray(img, dtype='float64') / 256.

        # put image in 4D tensor of shape (1, 3, height, width)
        img_ = img.transpose(2, 0, 1).reshape(1, 3, 639, 516)
        filtered_img = f(img_)

        # plot original image and first and second components of output
        pylab.subplot(1, 3, 1); pylab.axis('off'); pylab.imshow(img)
        pylab.gray();
        # recall that the convOp output (filtered image) is actually a "minibatch",
        # of size 1 here, so we take index 0 in the first dimension:
        pylab.subplot(1, 3, 2); pylab.axis('off'); pylab.imshow(filtered_img[0, 0, :, :])
        pylab.subplot(1, 3, 3); pylab.axis('off'); pylab.imshow(filtered_img[0, 1, :, :])
        pylab.show()


This should generate the following output.

.. image:: images/3wolfmoon_output.png
    :align: center

Notice that a randomly initialized filter acts very much like an edge detector!

Note that we use the same weight initialization formula as with the MLP.
Weights are sampled randomly from a uniform distribution in the range
[-1/fan-in, 1/fan-in], where fan-in is the number of inputs to a hidden unit.
For MLPs, this was the number of units in the layer below. For CNNs however, we
have to take into account the number of input feature maps and the size of the
receptive fields.


MaxPooling
++++++++++

Another important concept of CNNs is *max-pooling,* which is a form of
non-linear down-sampling. Max-pooling partitions the input image into
a set of non-overlapping rectangles and, for each such sub-region, outputs the
maximum value.

Max-pooling is useful in vision for two reasons: 
  #. By eliminating non-maximal values, it reduces computation for upper layers.

  #. It provides a form of translation invariance. Imagine
     cascading a max-pooling layer with a convolutional layer. There are 8
     directions in which one can translate the input image by a single pixel.
     If max-pooling is done over a 2x2 region, 3 out of these 8 possible
     configurations will produce exactly the same output at the convolutional
     layer. For max-pooling over a 3x3 window, this jumps to 5/8.

     Since it provides additional robustness to position, max-pooling is a
     "smart" way of reducing the dimensionality of intermediate representations.

Max-pooling is done in Theano by way of
``theano.tensor.signal.downsample.max_pool_2d``. This function takes as input
an N dimensional tensor (where N >= 2) and a downscaling factor and performs
max-pooling over the 2 trailing dimensions of the tensor.

An example is worth a thousand words:

.. code-block:: python

    from theano.tensor.signal import downsample

    input = T.dtensor4('input')
    maxpool_shape = (2, 2)
    pool_out = downsample.max_pool_2d(input, maxpool_shape, ignore_border=True)
    f = theano.function([input],pool_out)

    invals = numpy.random.RandomState(1).rand(3, 2, 5, 5)
    print 'With ignore_border set to True:'
    print 'invals[0, 0, :, :] =\n', invals[0, 0, :, :]
    print 'output[0, 0, :, :] =\n', f(invals)[0, 0, :, :]

    pool_out = downsample.max_pool_2d(input, maxpool_shape, ignore_border=False)
    f = theano.function([input],pool_out)
    print 'With ignore_border set to False:'
    print 'invals[1, 0, :, :] =\n ', invals[1, 0, :, :]
    print 'output[1, 0, :, :] =\n ', f(invals)[1, 0, :, :]

This should generate the following output:

.. code-block:: bash

    With ignore_border set to True:
        invals[0, 0, :, :] =
        [[  4.17022005e-01   7.20324493e-01   1.14374817e-04   3.02332573e-01 1.46755891e-01]
         [  9.23385948e-02   1.86260211e-01   3.45560727e-01   3.96767474e-01 5.38816734e-01]
         [  4.19194514e-01   6.85219500e-01   2.04452250e-01   8.78117436e-01 2.73875932e-02]
         [  6.70467510e-01   4.17304802e-01   5.58689828e-01   1.40386939e-01 1.98101489e-01]
         [  8.00744569e-01   9.68261576e-01   3.13424178e-01   6.92322616e-01 8.76389152e-01]]
        output[0, 0, :, :] =
        [[ 0.72032449  0.39676747]
         [ 0.6852195   0.87811744]]

    With ignore_border set to False:
        invals[1, 0, :, :] =
        [[ 0.01936696  0.67883553  0.21162812  0.26554666  0.49157316]
         [ 0.05336255  0.57411761  0.14672857  0.58930554  0.69975836]
         [ 0.10233443  0.41405599  0.69440016  0.41417927  0.04995346]
         [ 0.53589641  0.66379465  0.51488911  0.94459476  0.58655504]
         [ 0.90340192  0.1374747   0.13927635  0.80739129  0.39767684]]
        output[1, 0, :, :] =
        [[ 0.67883553  0.58930554  0.69975836]
         [ 0.66379465  0.94459476  0.58655504]
         [ 0.90340192  0.80739129  0.39767684]]

Note that compared to most Theano code, the ``max_pool_2d`` operation is a
little *special*. It requires the downscaling factor ``ds`` (tuple of length 2
containing downscaling factors for image width and height) to be known at graph
build time. This may change in the near future.


The Full Model: LeNet
+++++++++++++++++++++

Sparse, convolutional layers and max-pooling are at the heart of the LeNet
family of models. While the exact details of the model will vary greatly,
the figure below shows a graphical depiction of a LeNet model.

.. image:: images/mylenet.png
    :align: center

The lower-layers are composed to alternating convolution and max-pooling
layers. The upper-layers however are fully-connected and correspond to a
traditional MLP (hidden layer + logistic regression). The input to the
first fully-connected layer is the set of all features maps at the layer
below.

From an implementation point of view, this means lower-layers operate on 4D
tensors. These are then flattened to a 2D matrix of rasterized feature maps,
to be compatible with our previous MLP implementation.


Putting it All Together
+++++++++++++++++++++++

We now have all we need to implement a LeNet model in Theano. We start with the
LeNetConvPoolLayer class, which implements a {convolution + max-pooling}
layer.

.. literalinclude:: ../code/convolutional_mlp.py
  :pyobject: LeNetConvPoolLayer

Notice that when initializing the weight values, the fan-in is determined by
the size of the receptive fields and the number of input feature maps.

Finally, using the LogisticRegression class defined in :doc:`logreg` and
the HiddenLayer class defined in :doc:`mlp` , we can
instantiate the network as follows.

.. literalinclude:: ../code/convolutional_mlp.py
  :start-after: start-snippet-1
  :end-before: end-snippet-1
  
We leave out the code that performs the actual training and early-stopping,
since it is exactly the same as with an MLP. The interested reader can
nevertheless access the code in the 'code' folder of DeepLearningTutorials.

Running the Code
++++++++++++++++

The user can then run the code by calling:

.. code-block:: bash

    python code/convolutional_mlp.py

The following output was obtained with the default parameters on a Core i7-2600K
CPU clocked at 3.40GHz and using flags 'floatX=float32':

.. code-block:: bash

    Optimization complete.
    Best validation score of 0.910000 % obtained at iteration 17800,with test
    performance 0.920000 %
    The code for file convolutional_mlp.py ran for 380.28m

Using a GeForce GTX 285, we obtained the following:

.. code-block:: bash

    Optimization complete.
    Best validation score of 0.910000 % obtained at iteration 15500,with test
    performance 0.930000 %
    The code for file convolutional_mlp.py ran for 46.76m

And similarly on a GeForce GTX 480:

.. code-block:: bash

    Optimization complete.
    Best validation score of 0.910000 % obtained at iteration 16400,with test
    performance 0.930000 %
    The code for file convolutional_mlp.py ran for 32.52m

Note that the discrepancies in validation and test error (as well as iteration
count) are due to different implementations of the rounding mechanism in
hardware. They can be safely ignored.

Tips and Tricks
+++++++++++++++

Choosing Hyperparameters
------------------------

CNNs are especially tricky to train, as they add even more hyper-parameters than
a standard MLP. While the usual rules of thumb for learning rates and
regularization constants still apply, the following should be kept in mind when
optimizing CNNs.

Number of filters
*****************
When choosing the number of filters per layer, keep in mind that computing the
activations of a single convolutional filter is much more expensive than with
traditional MLPs !

Assume layer :math:`(l-1)` contains :math:`K^{l-1}` feature
maps and :math:`M \times N` pixel positions (i.e.,
number of positions times number of feature maps),
and there are :math:`K^l` filters at layer :math:`l` of shape :math:`m \times n`.
Then computing a feature map (applying an :math:`m \times n` filter
at all :math:`(M-m) \times (N-n)` pixel positions where the
filter can be applied) costs :math:`(M-m) \times (N-n) \times m \times n \times K^{l-1}`.
The total cost is :math:`K^l` times that. Things may be more complicated if
not all features at one level are connected to all features at the previous one.

For a standard MLP, the cost would only be :math:`K^l \times K^{l-1}`
where there are :math:`K^l` different neurons at level :math:`l`.
As such, the number of filters used in CNNs is typically much
smaller than the number of hidden units in MLPs and depends on the size of the
feature maps (itself a function of input image size and filter shapes).

Since feature map size decreases with depth, layers near the input layer will tend to
have fewer filters while layers higher up can have much more. In fact, to
equalize computation at each layer, the product of the number of features
and the number of pixel positions is typically picked to be roughly constant
across layers. To preserve the information about the input would require
keeping the total number of activations (number of feature maps times
number of pixel positions) to be non-decreasing from one layer to the next
(of course we could hope to get away with less when we are doing supervised
learning). The number of feature maps directly controls capacity and so
that depends on the number of available examples and the complexity of
the task.


Filter Shape
************
Common filter shapes found in the litterature vary greatly, usually based on
the dataset. Best results on MNIST-sized images (28x28) are usually in the 5x5
range on the first layer, while natural image datasets (often with hundreds of pixels in each
dimension) tend to use larger first-layer filters of shape 12x12 or 15x15.

The trick is thus to find the right level of "granularity" (i.e. filter
shapes) in order to create abstractions at the proper scale, given a
particular dataset.


Max Pooling Shape
*****************
Typical values are 2x2 or no max-pooling. Very large input images may warrant
4x4 pooling in the lower-layers. Keep in mind however, that this will reduce the
dimension of the signal by a factor of 16, and may result in throwing away too
much information.


.. rubric:: Footnotes

.. [#f1] For clarity, we use the word "unit" or "neuron" to refer to the
         artificial neuron and "cell" to refer to the biological neuron.


Tips
****

If you want to try this model on a new dataset, here are a few tips that can help you get better results:

 * Whitening the data (e.g. with PCA)
 * Decay the learning rate in each epoch