CS 224D: Deep Learning for NLP1 1

Course Instructor: Richard Socher

Lecture Notes: Part III2 2

Author: Rohit Mundra, Richard Socher

Spring 2015

Keyphrases: Neural networks. Forward computation. Backward

propagation. Neuron Units. Max-margin Loss. Gradient checks.
Xavier parameter initialization. Learning rates. Adagrad.
This set of notes introduces single and multilayer neural networks,
and how they can be used for classification purposes. We then dis-
cuss how they can be trained using a distributed gradient descent
technique known as backpropagation. We will see how the chain
rule can be used to make parameter updates sequentially. After a
rigorous mathematical discussion of neural networks, we will discuss
some practical tips and tricks in training neural networks involving:
neuron units (non-linearities), gradient checks, Xavier parameter
initialization, learning rates, Adagrad, etc. Lastly, we will motivate
using recurrent neural networks as a language model.

1 Neural Networks: Foundations

We established in our previous discussions the need for non-linear

classifiers since most data are not linearly separable and thus, our
classification performance on them is limited. Neural networks are
a family of classifiers with non-linear decision boundary as seen in
Figure 1: We see here how a non-linear
Figure 1. Now that we know the sort of decision boundaries neural decision boundary separates the data
networks create, let us see how they manage doing so. very well. This is the prowess of neural
networks.
Fun Fact:
1.1 A Neuron Neural networks are biologically in-
spired classifiers which is why they are
A neuron is a generic computational unit that takes n inputs and often called "artificial neural networks"
to distinguish them from the organic
produces a single output. What differentiates the outputs of different kind. However, in reality human neural
neurons is their parameters (also referred to as their weights). One of networks are so much more capable
and complex from artificial neural net-
the most popular choices for neurons is the "sigmoid" or "binary lo-
works that it is usually better to not
gistic regression" unit. This unit takes an n-dimensional input vector draw too many parallels between the
x and produces the scalar activation (output) a. This neuron is also two.

associated with an n-dimensional weight vector, w, and a bias scalar,

b. The output of this neuron is then:
1
a=
1 + exp(w T x + b)
We can also combine the weights and bias term above to equiva- Neuron:
lently formulate: A neuron is the fundamental building
block of neural networks. We will
1 see that a neuron can be one of many
a= functions that allows for non-linearities
1 + exp([w T b] · [ x 1]) to accrue in the network.
 cs 224d: deep learning for nlp 2

This formulation can be visualized in the manner shown in Fig-

ure 2.

1.2 A Single Layer of Neurons

We extend the idea above to multiple neurons by considering the
case where the input x is fed as an input to multiple such neurons as
shown in Figure 3. Figure 2: This image captures how in
a sigmoid neuron, the input vector x is
If we refer to the different neurons’ weights as {w(1) , · · · , w(m) } first scaled, summed, added to a bias
and the biases as {b1 , · · · , bm }, we can say the respective activations unit, and then passed to the squashing
sigmoid function.
are { a1 , · · · , am }:
1
a1 =
1 + exp(w(1)T x + b1 )
..
.
1
am =
1 + exp(w(m)T x + b m)

Let us define the following abstractions to keep the notation simple

and useful for more complex networks:
 1 
1+exp(z1 )
 .. 
σ(z) = 
 .


1
1+exp(zm )

b1
 .  m
 ..  ∈ R
b= 
bm
 
− w (1) T −
m×n
W= ··· ∈R
 
− w(m) T −
We can now right the output of scaling and biases as:

z = Wx + b
Figure 3: This image captures how
The activations of the sigmoid function can then be written as: multiple sigmoid units are stacked on
  the right, all of which receive the same
a (1) input x.
 . 
 .  = σ (z) = σ (Wx + b)
 . 
a(m)

So what do these activations really tell us? Well, one can think
of these activations as indicators of the presence of some weighted
combination of features. We can then use a combination of these
activations to perform classification tasks.
 cs 224d: deep learning for nlp 3

1.3 Feed-forward Computation

So far we have seen how an input vector x ∈ Rn can be fed to a
layer of sigmoid units to create activations a ∈ Rm . But what is the
intuition behind doing so? Let us consider the following named-
entity recognition (NER) problem in NLP as an example:

"Museums in Paris are amazing"

Dimensions for a single hidden layer
Here, we want to classify whether or not the center word "Paris" is neural network: If we represent each
word using a 4-dimensional word
a named-entity. In such cases, it is very likely that we would not just vector and we use a 5-word window
want to capture the presence of words in the window of word vectors as input, then the input x ∈ R20 . If we
but some other interactions between the words in order to make the use 8 sigmoid units in the hidden layer
and generate 1 score output from the
classification. For instance, maybe it should matter that "Museums" activations, then W ∈ R8×20 , b ∈ R8 ,
is the first word only if "in" is the second word. Such non-linear de- U ∈ R8×1 , s ∈ R. The stage-wise
feed-forward computation is then:
cisions can often not be captured by inputs fed directly to a Softmax
function but instead require the scoring of the intermediate layer z = Wx + b
discussed in Section 1.2. We can thus use another matrix U ∈ Rm×1 a = σ(z)
to generate an unnormalized score for a classification task from the s = UT a
activations:
s = U T a = U T σ (Wx + b)
Analysis of Dimensions: If we represent each word using a 4-
dimensional word vector and we use a 5-word window as input (as
in the above example), then the input x ∈ R20 . If we use 8 sigmoid
units in the hidden layer and generate 1 score output from the activa-
tions, then W ∈ R8×20 , b ∈ R8 , U ∈ R8×1 , s ∈ R.

1.4 Maximum Margin Objective Function

Like most machine learning models, neural networks also need an Figure 4: This image captures how a
simple feed-forward network might
optimization objective, a measure of error or goodness which we compute its output.
want to minimize or maximize respectively. Here, we will discuss a
popular error metric known as the maximum margin objective. The
idea behind using this objective is to ensure that the score computed
for "true" labeled data points is higher than the score computed for
"false" labeled data points.
Using the previous example, if we call the score computed for the
"true" labeled window "Museums in Paris are amazing" as s and the
score computed for the "false" labeled window "Not all museums in
Paris" as sc (subscripted as c to signify that the window is "corrupt").
Then, our objective function would be to maximize (s − sc ) or to
minimize (sc − s). However, we modify our objective to ensure that
error is only computed if sc > s ⇒ (sc − s) > 0. The intuition
behind doing this is that we only care the the "true" data point have
a higher score than the "false" data point and that the rest does not
 cs 224d: deep learning for nlp 4

matter. Thus, we want our error to be (sc − s) if sc > s else 0. Thus,

our optimization objective is now:

minimize J = max(sc − s, 0)

However, the above optimization objective is risky in the sense that

it does not attempt to create a margin of safety. We would want the
"true" labeled data point to score higher than the "false" labeled data
point by some positive margin ∆. In other words, we would want
error to be calculated if (s − sc < ∆) and not just when (s − sc < 0).
Thus, we modify the optimization objective:

minimize J = max(∆ + sc − s, 0)

We can scale this margin such that it is ∆ = 1 and let the other
parameters in the optimization problem adapt to this without any
change in performance. For more information on this, read about
functional and geometric margins - a topic often covered in the study
of Support Vector Machines. Finally, we define the following opti- The max-margin objective function
mization objective which we optimize over all training windows: is most commonly associated with
Support Vector Machines (SVMs)

minimize J = max(1 + sc − s, 0)

In the above formulation sc = U T σ (Wxc + b) and s = U T σ (Wx + b).

1.5 Training with Backpropagation – Elemental

In this section we discuss how we train the different parameters in
the model when the cost J discussed in Section 1.4 is positive. No
parameter updates are necessary if the cost is 0. Since we typically
update parameters using gradient descent (or a variant such as SGD),
we typically need the gradient information for any parameter as
required in the update equation:

θ ( t +1) = θ ( t ) − α ∇ θ ( t ) J

Backpropagation is technique that allows us to use the chain rule

of differentiation to calculate loss gradients for any parameter used
in the feed-forward computation on the model. To understand this
further, let us understand the toy network shown in Figure 5 for
which we will perform backpropagation.
Here, we use a neural network with a single hidden layer and a
single unit output. Let us establish some notation that will make it
Figure 5: This is a 4-2-1 neural network
easier to generalize this model later: where neuron j on layer k receives input
(k) (k)
zj and produces activation output a j .
• xi is an input to the neural network.

• s is the output of the neural network.

 cs 224d: deep learning for nlp 5

• Each layer (including the input and output layers) has neurons
which receive an input and produce an output. The j-th neuron
(k)
of layer k receives the scalar input z j and produces the scalar
(k)
activation output aj .

(k) (k)
• We will call the backpropagated error calculated at z j as δj .

• Layer 1 refers to the input layer and not the first hidden layer. For
(1) (1)
the input layer, x j = z j = aj .

• W (k) is the transfer matrix that maps the output from the k-th layer
to the input to the (k + 1)-th Thus, W (1) = W and W (2) = U to put
this new generalized notation in perspective of Section 1.3.
Backpropagation Notation:
Let us begin: Suppose the cost J = (1 + sc − s) is positive and • xi is an input to the neural network.
(1)
we want to perform the update of parameter W14 (in Figure 5 and • s is the output of the neural net-
(1) (2) work.
Figure 6), we must realize that W14 only contributes to z1 and
(2) • The j-th neuron of layer k receives
thus a1 . This fact is crucial to understanding backpropagation – (k)
the scalar input z j and produces
backpropagated gradients are only affected by values they contribute (k)
(2) the scalar activation output a j .
to. a1 is consequently used in the forward computation of score by
(1) (1)
(2) • For the input layer, x j = z j = aj .
multiplication with W1 . We can see from the max-margin loss that:
• W (k) is the transfer matrix that maps
∂J ∂J the output from the k-th layer to
=− = −1 the input to the (k + 1)-th. Thus,
∂s ∂sc
W (1) = W and W (2) = U T using
Therefore we will work with ∂s
here for simplicity. Thus, notation from Section 1.3.
(1)
∂Wij

(2) (2) (2)

∂s ∂W (2) a(2) ∂Wi ai (2) ∂ai
(1)
= (1)
= (1)
= Wi (1)
∂Wij ∂Wij ∂Wij ∂Wij
(2) (2) (2)
(2) ∂ai (2) ∂ai ∂zi
⇒ Wi (1)
= Wi (2) (1)
∂Wij ∂zi ∂Wij
(2) (2)
(2) σ ( z i ) ∂zi
= Wi (2) (1)
∂zi ∂Wij
(2)
(2) (2) ∂zi
= Wi σ0 (zi ) (1)
∂Wij
(2) (2) ∂ (1) (1) (1) (1) (1) (1) (1) (1) (1)
= Wi σ0 (zi ) (1)
( bi + a1 Wi1 + a2 Wi2 + a3 Wi3 + a4 Wi4 )
∂Wij
∂
+ ∑ ak Wik )
(2) (2) (1) (1) (1)
= Wi σ0 (zi ) (1)
( bi
∂Wij k
(2) (2) (1)
= Wi σ0 (zi ) a j
(2) (1)
= δi · aj
 cs 224d: deep learning for nlp 6

(2) (1)
We see above that the gradient reduces to the product δi · aj where
(2)
δi is essentially the error propagating backwards from the i-th neu-
(1)
ron in layer 2. a j is an input fed to i-th neuron in layer 2 when
scaled by Wij .
Let us discuss the "error sharing/distribution" interpretation of
backpropagation better using Figure 6 as an example. Say we were to
(1) Figure 6: This subnetwork shows the
update W14 :
relevant parts of the network required
(1)
to update Wij
1. We start with the an error signal of 1 propagating backwards from
(3)
a1 .

2. We then multiply this error by the local gradient of the neuron

(3) (3)
which maps z1 to a1 . This happens to be 1 in this case and thus,
(3)
the error is still 1. This is now known as δ1 = 1.
(3)
3. At this point, the error signal of 1 has reached z1 . We now need
to distribute the error signal so that the "fair share" of the error
(2)
reaches to a1 .
(3) (3) (2) (2)
4. This amount is the (error signal at z1 = δ1 )×W1 = W1 . Thus,
(2) (2)
the error at a1 = W1 .

5. As we did in step 2, we need to move the error across the neuron

(2) (2)
which maps z1 to a1 . We do this by multiplying the error signal
(2)
at a1 by the local gradient of the neuron which happens to be
(2) (2)
σ (z1 )(1 − σ(z1 )).
(2) (2) (2) (2)
6. Thus, the error signal at z1 is σ (z1 )(1 − σ (z1 ))W1 . This is
(2)
known as δ1 .
(1)
7. Finally, we need to distribute the "fair share" of the error to W14
by simply multiplying it by the input it was responsible for for-
(1)
warding, which happens to be a4 .
(1)
8. Thus, the gradient of the loss with respect to W14 is calculated to
(1) (2) (2) (2)
be a4 σ(z1 )(1 − σ (z1 ))W1 .

Notice that the result we arrive at using this approach is exactly

the same as that we arrived at using explicit differentiation earlier.
Thus, we can calculate error gradients with respect to a parameter
in the network using either the chain rule of differentiation or using
an error sharing and distributed flow approach – both of these ap-
proaches happen to do the exact same thing but it might be helpful
 cs 224d: deep learning for nlp 7

to think about them one way or another.

(1)
Bias Updates: Bias terms (such as b1 ) are mathematically equivalent
(2)
to other weights contributing to the neuron input (z1 ) as long as the
input being forwarded is 1. As such, the bias gradients for neuron
(k) (1)
i on layer k is simply δi . For instance, if we were updating b1
(1) (2)
instead of W14 above, the gradient would simply be σ (z1 )(1 −
(2) (2)
σ(z1 ))W1 .

Generalized steps to propagate δ(k) to δ(k−1) :

(k) (k)
1. We have error δi propagating backwards from zi , i.e. neuron i Figure 7: Propagating error from δ(k) to
at layer k. See Figure 7. δ ( k −1)

( k −1) (k)
2. We propagate this error backwards to a j by multiplying δi by
( k −1)
the path weight Wij .

( k −1) (k) ( k −1)

3. Thus, the error received at a j is δi Wij .

( k −1)
4. However, a j may have been forwarded to multiple nodes in
the next layer as shown in Figure 8. It should receive responsibility
for errors propagating backward from node m in layer k too, using
the exact same mechanism.
( k −1) (k) ( k −1) (k) ( k −1)
5. Thus, error received at a j is δi Wij + δm Wmj .

(k) ( k −1)
6. In fact, we can generalize this to be ∑i δi Wij .

( k −1)
7. Now that we have the correct error at a j , we move it across
neuron j at layer k − 1 by multiplying with with the local gradient
( k −1)
σ0 (z j ).
( k −1) ( k −1)
8. Thus, the error that reaches z j , called δj is
( k −1) (k) ( k −1)
σ0 (z j ) ∑i δi Wij

1.6 Training with Backpropagation – Vectorized

So far, we discussed how to calculate gradients for a given parameter Figure 8: Propagating error from δ(k) to
in the model. Here we will generalize the approach above so that δ ( k −1)
we update weight matrices and bias vectors all at once. Note that
these are simply extensions of the above model that will help build
intuition for the way error propagation can be done at a matrix-
vector level.
 cs 224d: deep learning for nlp 8

(k)
For a given parameter Wij , we identified that the error gradient is
( k +1) (k)
simply δi · a j . As a reminder, W (k) is the matrix that maps a(k)
to z(k+1) . We can thus establish that the error gradient for the entire
matrix W (k) is:
 ( k +1) ( k ) ( k +1) ( k )

δ1 a1 δ1 a2 ···
 ( k +1) ( k ) ( k +1) ( k )  ( k +1) ( k ) T
∇W ( k ) = 
δ2 a1 δ2 a2 · · ·
=δ a
.. .. ..
. . .
Error propagates from layer (k + 1) to
Thus, we can write an entire matrix gradient using the outer prod- (k) in the following manner:
uct of the error vector propagating into the matrix and the activations
δ ( k ) = f 0 ( z ( k ) ) ◦ (W ( k ) T δ ( k + 1 ) )
forwarded by the matrix.
Of course, this assumes that in the
Now, we will see how we can calculate the error vector δ(k) . We forward propagation the signal z(k) first
(k) (k) ( k +1) ( k )
established earlier using Figure 8 that δj = σ0 (z j ) ∑i δi Wij . goes through activation neurons f to
generate activations a(k) and are then
This can easily generalize to matrices such that:
linearly combined to yield z(k+1) via
transfer matrix W (k) .
δ ( k ) = σ 0 ( z ( k ) ) ◦ (W ( k ) T δ ( k + 1 ) )

In the above formulation, the ◦ operator corresponds to an element

wise product between elements of vectors (◦ : R N × R N → R N ).

Computational efficiency: Having explored element-wise updates

as well as vector-wise updates, we must realize that the vectorized
implementations run substantially faster in scientific computing
environments such as MATLAB or Python (using NumPy/SciPy
packages). Thus, we should use vectorized implementation in prac-
tice. Furthermore, we should also reduce redundant calculations
in backpropagation - for instance, notice that δ(k) depends directly
on δ(k+1) . Thus, we should ensure that when we update W (k) using
δ(k+1) , we save δ(k+1) to later derive δ(k) – and we then repeat this for
(k − 1) . . . (1). Such a recursive procedure is what makes backpropa-
gation a computationally affordable procedure.

2 Neural Networks: Tips and Tricks

Having discussed the mechanics of neural networks, we will now

dive into some tips and tricks commonly employed when using neu-
ral networks in practice.

2.1 Gradient Check

We have discussed how to calculate error gradients for parameters
in a neural network model using calculus. Here we discuss another
 cs 224d: deep learning for nlp 9

technique of approximating the gradient without the use of error

backpropagation:

∂J J (θ + h) − J (θ − h) J (θ + h) − J (θ − h)
≈ lim = lim
∂θ h →0 ( θ + h ) − ( θ − h ) h →0 2h
We know the above holds true from the definition of differentia-
tion, but how does this apply to getting an error gradient? Well, the
term J (θ + h) is simply the error calculated on a forward pass for a
given dataset when we change perturb the parameter θ by +h. Sim-
ilarly, the term J (θ − h) is the error calculated on a forward pass for
the same dataset when we change perturb the parameter θ by −h.
Thus, using two forward passes, we can approximate the gradient
with respect to any given parameter in the model. Of course, consid-
ering how many operations are required for even a forward pass, this
is an extremely expensive procedure of evaluating gradients – this
is a great way of verifying implementations of the backpropagation
however. Gradient checks are a great way to
A naive implementation of gradient check can be implemented in compare analytical and numerical
gradients. Analytical gradients should
the following manner: be close and numerical gradients can be
calculated using:
∂J J (θ + h) − J (θ − h)
Snippet 2.1 ≈ lim
∂θ h →0 2h
def eval_numerical_gradient(f, x): J (θ + h) and J (θ − h) can be evalu-
""" ated using two forward passes. An
implementation of this can be seen in
a naive implementation of numerical gradient of f at x
Snippet 2.1.
- f should be a function that takes a single argument
- x is the point (numpy array) to evaluate the gradient
at
"""

fx = f(x) # evaluate function value at original point

grad = np.zeros(x.shape)
h = 0.00001

# iterate over all indexes in x

it = np.nditer(x, flags=[’multi_index’],
op_flags=[’readwrite’])
while not it.finished:

# evaluate function at x+h

ix = it.multi_index
old_value = x[ix]
x[ix] = old_value + h # increment by h
fxh = f(x) # evaluate f(x + h)
x[ix] = old_value # restore to previous value (very
 cs 224d: deep learning for nlp 10

important!)

# compute the partial derivative

grad[ix] = (fxh - fx) / h # the slope
it.iternext() # step to next dimension
return grad

2.2 Regularization
Like most classifiers, neural networks are also prone to overfitting
which causes their validation and test performances to be subop-
timal. We can implement L2 regularization such that the loss with
regularization, JR , is calculated to be:

L
JR = J + λ ∑ W (i)

F
i =1

In the above formulation, W (i) is the Frobenius norm of the

F
matrix W (i) and λ is the relative weight assigned to regularization
in the weighted-sum objective. Adding this type of regularization
penalizes weights for being large in magnitude by contributing
to the cost quadratically with it. This reduces the flexibility of the
target function (i.e. classifier) thereby reducing the overfitting phe-
nomenon. Imposing such a constraint can be interpreted as the prior
Bayesian belief that the optimal weights are close to zero. How close
you wonder? Well, that is what λ controls – large values of λ lead to
a stronger belief that the weights should be chose to zero. It must be
noted that the bias terms are not regularized and do not contribute to
the cost term above – try thinking about why this is the case.

2.3 Neuron Units

So far we have discussed neural networks that contain sigmoidal
neurons to allow nonlinearities, however in many applications better
networks can be designed using other activation functions. Some
common choices are listed here with their function and gradient
definitions and these can be substituted with the sigmoidal functions
discussed above.

Sigmoid: This is the default choice we have discussed and the activa-
tion function σ is:

1
σ(z) =
1 + exp(−z)
Figure 9: The response of a sigmoid
nonlinearity
 cs 224d: deep learning for nlp 11

where σ (z) ∈ (0, 1)

The gradient of σ (z) is:

− exp(−z)
σ0 (z) = = σ(z)(1 − σ(z))
1 + exp(−z)

Tanh: The tanh functionis an alternative to the sigmoid function that

is often found to converge faster in practice. The primary difference
between tanh and sigmoid is that tanh output ranges from −1 to 1
while the sigmoid ranges from 0 to 1.

exp(z) − exp(−z)
tanh(z) = = 2σ(2z) − 1
exp(z) + exp(−z) Figure 10: The response of a tanh
nonlinearity
where tanh(z) ∈ (−1, 1)

The gradient of tanh(z) is:

2
exp(z) − exp(−z)

0
tanh (z) = 1 − = 1 − tanh2 (z)
exp(z) + exp(−z)

Hard tanh: The hard tanh function is sometimes preferred over the
tanh function since it is computationally cheaper. It does however
saturate for magnitudes of z greater than 1. The activation of the
hard tanh is:

 −1
 : z < −1
hardtanh(z) = z : −1 ≤ z ≤ 1
 Figure 11: The response of a hard tanh
 1 :z>1
nonlinearity
The derivative can also be expressed in a piecewise functional form:
(
0 1 : −1 ≤ z ≤ 1
hardtanh (z) =
0 : otherwise

Soft sign: The soft sign function is another nonlinearity which can
be considered an alternative to tanh since it too does not saturate as
easily as hard clipped functions:

z
softsign(z) =
1 + |z|
The derivative is the expressed as: Figure 12: The response of a soft sign
nonlinearity
sgn(z)
softsign0 (z) =
(1 + z )2

where sgn is the signum function which returns ± 1 depending on the sign of z
 cs 224d: deep learning for nlp 12

ReLU: The ReLU (Rectified Linear Unit) function is a popular choice

of activation since it does not saturate even for larger values of z and
has found much success in computer vision applications:

rect(z) = max(z, 0)

The derivative is then the piecewise function: Figure 13: The response of a ReLU
( nonlinearity
1 :z>0
rect0 (z) =
0 : otherwise

Leaky ReLU: Traditional ReLU units by design do not propagate

any error for non-positive z – the leaky ReLU modifies this such
that a small error is allowed to propagate backwards even when z is
negative:

leaky(z) = max(z, k · z)

where 0 < k < 1 Figure 14: The response of a leaky

ReLU nonlinearity
This way, the derivative is representable as:
(
0 1 :z>0
leaky (z) =
k : otherwise

2.4 Xavier Parameter Initialization

In Understanding the difficulty of training deep feedfor-
ward neural networks (2010), Xavier et al study the effect
of different weight and bias initialization schemes on training dy-
namics. The empirical findings suggest that for sigmoid and tanh
activation units, lower error rates are achieved and faster convergence
( l +1) × n ( l )
occurs when the weights of a matrix W ∈ Rn are initialized
randomly with a uniform distribution with the following range:
 √ √ 
6 6
W ∼ U − (l ) ,
n + n ( l +1) n ( l ) + n ( l +1)

Where n(l ) is the number of input units to W

and n(l +1) is the number of output units from W

In this parameter initialization scheme, bias units are initialized to 0.

This approach attempts to maintain activation variances as well as
backpropagated gradient variances across layers. Without such ini-
tialization, the gradient variances (which are a proxy for information)
generally decrease with backpropagation across layers.
 cs 224d: deep learning for nlp 13

2.5 Learning Rate

The speed at which a parameter in the model updates can be con-
trolled using the learning rate. In the following naive gradient de-
scent formulation, α is the learning rate:

θ ( t +1) = θ ( t ) − α ∇ θ ( t ) J

You might think that for fast convergence rates, we should set α to
larger values however faster convergence is not guaranteed with
larger convergence rates. In fact, with very large learning rates, we
might experience that the loss function actually diverges because the
parameters update causes the model to overshoot the convex minima
as shown in Figure 15. In non-convex models (most of those we work
with), the outcome of a large learning rate is unpredictable but the
chances of diverging loss functions are very high.
The simple solution to avoiding a diverging loss is to use a very
small learning rate so that we carefully scan the parameter space.
Furthermore, we can keep this learning rate fixed for all parameters Figure 15: Here we see that updating
in the model instead of having a more complex scheme in which each parameter w2 with a large learning rate
can lead to divergence of the error.
parameter has its own learning rate.
Since training is the most expensive phase in a deep learning
system, some research has attempted to improve this naive approach
to setting learning learning rates. For instance, Ronan Collobert
( l +1) × n ( l )
scales the learning rate of a weight Wij (where W ∈ Rn ) by
( l )
the inverse square root of the fan-in of the neuron (n ). Another
approach is to allow the learning rate to decrease over time such that:

α (0) τ
α(t) =
max(t, τ )

In the above scheme, α(0) is a tunable parameter and represents the

starting learning rate. τ is also a tunable parameter and represents
the time at which the learning rate should start reducing. In practice,
this method has been found to work quite well. In the next section
we discuss another method for adaptive gradient descent which does
not require hand-set learning rates.

2.6 Subgradient Optimization Using AdaGrad

AdaGrad is an implementation of standard stochastic gradient de-
scent (SGD) with one key difference: the learning rate can vary for
each parameter. The learning rate for each parameter depends on
the history of gradient updates of that parameter in a way such that
parameters with a scarce history of updates are updated faster using
a larger learning rate. In other words, parameters that have not been
 cs 224d: deep learning for nlp 14

updated much in the past are likelier to have higher learning rates
now. Formally:

α
θ ( t ) = θ ( t −1) − q ∇θ J (t)
∑tτ =1 (∇θ J (τ ) )2
In this technique, we see that if the RMS of the history of gradients
is extremely low, the learning rate is very high. A simple implemen-
tation of this technique is:

Snippet 2.2
# Assume the gradient dx and parameter vector x
cache += dx**2
x += - learning_rate * dx / np.sqrt(cache + 1e-8)