Run this notebook online:Binder or Colab: Colab

4.6. Dropout

In Section 4.5, we introduced the classical approach to regularizing statistical models by penalizing the \(\ell_2\) norm of the weights. In probabilistic terms, we could justify this technique by arguing that we have assumed a prior belief that weights take values from a Gaussian distribution with mean \(0\). More intuitively, we might argue that we encouraged the model to spread out its weights among many features rather than depending too much on a small number of potentially spurious associations.

4.6.1. Overfitting Revisited

Faced with more features than examples, linear models tend to overfit. But given more examples than features, we can generally count on linear models not to overfit. Unfortunately, the reliability with which linear models generalize comes at a cost: Naively applied, linear models do not take into account interactions among features. For every feature, a linear model must assign either a positive or a negative weight, ignoring context.

In traditional texts, this fundamental tension between generalizability and flexibility is described as the bias-variance tradeoff. Linear models have high bias (they can only represent a small class of functions), but low variance (they give similar results across different random samples of the data).

Deep neural networks inhabit the opposite end of the bias-variance spectrum. Unlike linear models, neural networks are not confined to looking at each feature individually. They can learn interactions among groups of features. For example, they might infer that “Nigeria” and “Western Union” appearing together in an email indicates spam but that separately they do not.

Even when we have far more examples than features, deep neural networks are capable of overfitting. In 2017, a group of researchers demonstrated the extreme flexibility of neural networks by training deep nets on randomly-labeled images. Despite the absence of any true pattern linking the inputs to the outputs, they found that the neural network optimized by SGD could label every image in the training set perfectly.

Consider what this means. If the labels are assigned uniformly at random and there are 10 classes, then no classifier can do better than 10% accuracy on holdout data. The generalization gap here is a whopping 90%. If our models are so expressive that they can overfit this badly, then when should we expect them not to overfit? The mathematical foundations for the puzzling generalization properties of deep networks remain open research questions, and we encourage the theoretically-oriented reader to dig deeper into the topic. For now, we turn to the more terrestrial investigation of practical tools that tend (empirically) to improve the generalization of deep nets.

4.6.2. Robustness through Perturbations

Let us think briefly about what we expect from a good predictive model. We want it to peform well on unseen data. Classical generalization theory suggests that to close the gap between train and test performance, we should aim for a simple model. Simplicity can come in the form of a small number of dimensions. We explored this when discussing the monomial basis functions of linear models Section 4.4. Additionally, as we saw when discussing weight decay (\(\ell_2\) regularization) Section 4.5, the (inverse) norm of the parameters represents another useful measure of simplicity. Another useful notion of simplicity is smoothness, i.e., that the function should not be sensitive to small changes to its inputs. For instance, when we classify images, we would expect that adding some random noise to the pixels should be mostly harmless.

In 1995, Christopher Bishop formalized this idea when he proved that training with input noise is equivalent to Tikhonov regularization [Bishop, 1995]. This work drew a clear mathematical connection between the requirement that a function be smooth (and thus simple), and the requirement that it be resilient to perturbations in the input.

Then, in 2014, Srivastava et al. [Srivastava et al., 2014] developed a clever idea for how to apply Bishop’s idea to the internal layers of the network, too. Namely, they proposed to inject noise into each layer of the network before calculating the subsequent layer during training. They realized that when training a deep network with many layers, injecting noise enforces smoothness just on the input-output mapping.

Their idea, called dropout, involves injecting noise while computing each internal layer during forward propagation, and it has become a standard technique for training neural networks. The method is called dropout because we literally drop out some neurons during training. Throughout training, on each iteration, standard dropout consists of zeroing out some fraction (typically 50%) of the nodes in each layer before calculating the subsequent layer.

To be clear, we are imposing our own narrative with the link to Bishop. The original paper on dropout offers intuition through a surprising analogy to sexual reproduction. The authors argue that neural network overfitting is characterized by a state in which each layer relies on a specifc pattern of activations in the previous layer, calling this condition co-adaptation. Dropout, they claim, breaks up co-adaptation just as sexual reproduction is argued to break up co-adapted genes.

The key challenge then is how to inject this noise. One idea is to inject the noise in an unbiased manner so that the expected value of each layer—while fixing the others—equals to the value it would have taken absent noise.

In Bishop’s work, he added Gaussian noise to the inputs to a linear model: At each training iteration, he added noise sampled from a distribution with mean zero \(\epsilon \sim \mathcal{N}(0,\sigma^2)\) to the input \(\mathbf{x}\), yielding a perturbed point \(\mathbf{x}' = \mathbf{x} + \epsilon\). In expectation, \(E[\mathbf{x}'] = \mathbf{x}\).

In standard dropout regularization, one debiases each layer by normalizing by the fraction of nodes that were retained (not dropped out). In other words, dropout with dropout probability \(p\) is applied as follows:

(4.6.1)\[\begin{split}\begin{aligned} h' = \begin{cases} 0 & \text{ with probability } p \\ \frac{h}{1-p} & \text{ otherwise} \end{cases} \end{aligned}\end{split}\]

By design, the expectation remains unchanged, i.e., \(E[h'] = h\). Intermediate activations \(h\) are replaced by a random variable \(h'\) with matching expectation.

4.6.3. Dropout in Practice

Recall the multilayer perceptron (Section 4.1) with a hidden layer and 5 hidden units. Its architecture is given by

(4.6.2)\[\begin{split}\begin{aligned} \mathbf{h} & = \sigma(\mathbf{W}_1 \mathbf{x} + \mathbf{b}_1), \\ \mathbf{o} & = \mathbf{W}_2 \mathbf{h} + \mathbf{b}_2, \\ \hat{\mathbf{y}} & = \mathrm{softmax}(\mathbf{o}). \end{aligned}\end{split}\]

When we apply dropout to a hidden layer, zeroing out each hidden unit with probability \(p\), the result can be viewed as a network containing only a subset of the original neurons. In Fig. 4.6.1, \(h_2\) and \(h_5\) are removed. Consequently, the calculation of \(y\) no longer depends on \(h_2\) and \(h_5\) and their respective gradient also vanishes when performing backprop. In this way, the calculation of the output layer cannot be overly dependent on any one element of \(h_1, \ldots, h_5\).

Fig. 4.6.1 MLP before and after dropout

Typically, *we disable dropout at test time*. Given a trained model and a new example, we do not drop out any nodes (and thus do not need to normalize). However, there are some exceptions: some researchers use dropout at test time as a heuristic for estimating the uncertainty of neural network predictions: if the predictions agree across many different dropout masks, then we might say that the network is more confident. For now we will put off uncertainty estimation for subsequent chapters and volumes.

4.6.4. Implementation from Scratch

To implement the dropout function for a single layer, we must draw as many samples from a Bernoulli (binary) random variable as our layer has dimensions, where the random variable takes value \(1\) (keep) with probability \(1-p\) and \(0\) (drop) with probability \(p\). One easy way to implement this is to first draw samples from the uniform distribution \(U[0, 1]\). Then we can keep those nodes for which the corresponding sample is greater than \(p\), dropping the rest.

In the following code, we implement a dropoutLayer() function that drops out the elements in the NDArray input X with probability dropout, rescaling the remainder as described above (dividing the survivors by 1.0-dropout).

%load ../utils/djl-imports
%load ../utils/plot-utils.ipynb
%load ../utils/
%load ../utils/
%load ../utils/
import org.apache.commons.lang3.ArrayUtils;

We can test out the dropoutLayer() function on a few examples. In the following lines of code, we pass our input X through the dropout operation, with probabilities 0, 0.5, and 1, respectively.

NDManager manager = NDManager.newBaseManager();

public NDArray dropoutLayer(NDArray X, float dropout) {
    // In this case, all elements are dropped out
    if (dropout == 1.0f) {
        return manager.zeros(X.getShape());
    // In this case, all elements are kept
    if (dropout == 0f) {
        return X;

    NDArray mask = manager.randomUniform(0f, 1.0f, X.getShape()).gt(dropout);
    return mask.toType(DataType.FLOAT32, false).mul(X).div(1.0f - dropout);
NDArray X = manager.arange(16f).reshape(2, 8);
System.out.println(dropoutLayer(X, 0));
System.out.println(dropoutLayer(X, 0.5f));
System.out.println(dropoutLayer(X, 1.0f));
ND: (2, 8) gpu(0) float32
[[ 0.,  1.,  2.,  3.,  4.,  5.,  6.,  7.],
 [ 8.,  9., 10., 11., 12., 13., 14., 15.],

ND: (2, 8) gpu(0) float32
[[ 0.,  2.,  0.,  6.,  8.,  0., 12.,  0.],
 [ 0., 18., 20.,  0., 24.,  0., 28.,  0.],

ND: (2, 8) gpu(0) float32
[[0., 0., 0., 0., 0., 0., 0., 0.],
 [0., 0., 0., 0., 0., 0., 0., 0.],
] Defining Model Parameters

Again, we work with the Fashion-MNIST dataset introduced in Section 3.6. We define a multilayer perceptron with two hidden layers containing 256 outputs each.

int numInputs = 784;
int numOutputs = 10;
int numHiddens1 = 256;
int numHiddens2 = 256;

NDArray W1 = manager.randomNormal(0, 0.01f, new Shape(numInputs, numHiddens1), DataType.FLOAT32);
NDArray b1 = manager.zeros(new Shape(numHiddens1));
NDArray W2 = manager.randomNormal(0, 0.01f, new Shape(numHiddens1, numHiddens2), DataType.FLOAT32);
NDArray b2 = manager.zeros(new Shape(numHiddens2));
NDArray W3 = manager.randomNormal(0, 0.01f, new Shape(numHiddens2, numOutputs), DataType.FLOAT32);
NDArray b3 = manager.zeros(new Shape(numOutputs));

NDList params = new NDList(W1, b1, W2, b2, W3, b3);

for (NDArray param : params) {
} Defining the Model

The model below applies dropout to the output of each hidden layer (following the activation function). We can set dropout probabilities for each layer separately. A common trend is to set a lower dropout probability closer to the input layer. Below we set it to 0.2 and 0.5 for the first and second hidden layer respectively. By using the isTraining boolean variable described in Section 2.5, we can ensure that dropout is only active during training.

float dropout1 = 0.2f;
float dropout2 = 0.5f;

public NDArray net(NDArray X, boolean isTraining) {

    X = X.reshape(-1, numInputs);
    NDArray H1 = Activation.relu(;

    if (isTraining) {
        H1 = dropoutLayer(H1, dropout1);

    NDArray H2 = Activation.relu(;
    if (isTraining) {
        H2 = dropoutLayer(H2, dropout2);

} Training and Testing

This is similar to the training and testing of multilayer perceptrons described previously.

int numEpochs = Integer.getInteger("MAX_EPOCH", 10);
float lr = 0.5f;
int batchSize = 256;

double[] trainLoss;
double[] testAccuracy;
double[] epochCount;
double[] trainAccuracy;

trainLoss = new double[numEpochs];
trainAccuracy = new double[numEpochs];
testAccuracy = new double[numEpochs];
epochCount = new double[numEpochs];

Loss loss = new SoftmaxCrossEntropyLoss();

FashionMnist trainIter = FashionMnist.builder()
        .setSampling(batchSize, true)
        .optLimit(Long.getLong("DATASET_LIMIT", Long.MAX_VALUE))

FashionMnist testIter = FashionMnist.builder()
        .setSampling(batchSize, true)
        .optLimit(Long.getLong("DATASET_LIMIT", Long.MAX_VALUE))

float epochLoss = 0f;
float accuracyVal = 0f;

for (int epoch = 1; epoch <= numEpochs; epoch++) {
    // Iterate over dataset
    System.out.print("Running epoch " + epoch + "...... ");
    for (Batch batch : trainIter.getData(manager)) {
        NDArray X = batch.getData().head();
        NDArray y = batch.getLabels().head();

        try (GradientCollector gc = Engine.getInstance().newGradientCollector()) {
            NDArray yHat = net(X, true); // net function call

            NDArray lossValue = loss.evaluate(new NDList(y), new NDList(yHat));
            NDArray l = lossValue.mul(batchSize);

            epochLoss += l.sum().getFloat();

            accuracyVal += Training.accuracy(yHat, y);
            gc.backward(l); // gradient calculation

        Training.sgd(params, lr, batchSize); // updater

    trainLoss[epoch-1] = epochLoss/trainIter.size();
    trainAccuracy[epoch-1] = accuracyVal/trainIter.size();

    epochLoss = 0f;
    accuracyVal = 0f;

    for (Batch batch : testIter.getData(manager)) {
        NDArray X = batch.getData().head();
        NDArray y = batch.getLabels().head();

        NDArray yHat = net(X, false); // net function call
        accuracyVal += Training.accuracy(yHat, y);

    testAccuracy[epoch-1] = accuracyVal/testIter.size();
    epochCount[epoch-1] = epoch;
    accuracyVal = 0f;
    System.out.println("Finished epoch " + epoch);

System.out.println("Finished training!");
Running epoch 1...... Finished epoch 1
Running epoch 2...... Finished epoch 2
Running epoch 3...... Finished epoch 3
Running epoch 4...... Finished epoch 4
Running epoch 5...... Finished epoch 5
Running epoch 6...... Finished epoch 6
Running epoch 7...... Finished epoch 7
Running epoch 8...... Finished epoch 8
Running epoch 9...... Finished epoch 9
Running epoch 10...... Finished epoch 10
Finished training!
String[] lossLabel = new String[trainLoss.length + testAccuracy.length + trainAccuracy.length];

Arrays.fill(lossLabel, 0, trainLoss.length, "train loss");
Arrays.fill(lossLabel, trainAccuracy.length, trainLoss.length + trainAccuracy.length, "train acc");
Arrays.fill(lossLabel, trainLoss.length + trainAccuracy.length,
                trainLoss.length + testAccuracy.length + trainAccuracy.length, "test acc");

Table data = Table.create("Data").addColumns(
    DoubleColumn.create("epochCount", ArrayUtils.addAll(epochCount, ArrayUtils.addAll(epochCount, epochCount))),
    DoubleColumn.create("loss", ArrayUtils.addAll(trainLoss, ArrayUtils.addAll(trainAccuracy, testAccuracy))),
    StringColumn.create("lossLabel", lossLabel)

render(LinePlot.create("", data, "epochCount", "loss", "lossLabel"),"text/html");

4.6.5. Concise Implementation

Using DJL, all we need to do is add a Dropout layer after each fully-connected layer, passing in the dropout probability as the only argument to its constructor. During training, the Dropout layer will randomly drop out outputs of the previous layer (or equivalently, the inputs to the subsequent layer) according to the specified dropout probability. When the model is not in training mode, the Dropout layer simply passes the data through during testing.

SequentialBlock net = new SequentialBlock();
net.setInitializer(new NormalInitializer(0.01f), Parameter.Type.WEIGHT);

Next, we train and test the model.

Map<String, double[]> evaluatorMetrics = new HashMap<>();

Tracker lrt = Tracker.fixed(0.5f);
Optimizer sgd = Optimizer.sgd().setLearningRateTracker(lrt).build();

Loss loss = Loss.softmaxCrossEntropyLoss();

DefaultTrainingConfig config = new DefaultTrainingConfig(loss)
                .optOptimizer(sgd) // Optimizer (loss function)
                .optDevices(Engine.getInstance().getDevices(1)) // single GPU
                .addEvaluator(new Accuracy()) // Model Accuracy
                .addTrainingListeners(TrainingListener.Defaults.logging()); // Logging

    try (Model model = Model.newInstance("mlp")) {

        try (Trainer trainer = model.newTrainer(config)) {

            trainer.initialize(new Shape(1, 784));
            trainer.setMetrics(new Metrics());

  , numEpochs, trainIter, testIter);

            Metrics metrics = trainer.getMetrics();

                    .forEach(evaluator -> {
                        evaluatorMetrics.put("train_epoch_" + evaluator.getName(), metrics.getMetric("train_epoch_" + evaluator.getName()).stream()
                                            .mapToDouble(x -> x.getValue().doubleValue()).toArray());
                        evaluatorMetrics.put("validate_epoch_" + evaluator.getName(), metrics.getMetric("validate_epoch_" + evaluator.getName()).stream()
                                            .mapToDouble(x -> x.getValue().doubleValue()).toArray());
INFO Training on: 1 GPUs.
INFO Load MXNet Engine Version 1.8.0 in 0.062 ms.
Training:    100% |████████████████████████████████████████| Accuracy: 0.57, SoftmaxCrossEntropyLoss: 1.10
Validating:  100% |████████████████████████████████████████|
INFO Epoch 1 finished.
INFO Train: Accuracy: 0.57, SoftmaxCrossEntropyLoss: 1.10
INFO Validate: Accuracy: 0.74, SoftmaxCrossEntropyLoss: 0.68
Training:    100% |████████████████████████████████████████| Accuracy: 0.79, SoftmaxCrossEntropyLoss: 0.57
Validating:  100% |████████████████████████████████████████|
INFO Epoch 2 finished.
INFO Train: Accuracy: 0.79, SoftmaxCrossEntropyLoss: 0.57
INFO Validate: Accuracy: 0.76, SoftmaxCrossEntropyLoss: 0.58
Training:    100% |████████████████████████████████████████| Accuracy: 0.82, SoftmaxCrossEntropyLoss: 0.48
Validating:  100% |████████████████████████████████████████|
INFO Epoch 3 finished.
INFO Train: Accuracy: 0.82, SoftmaxCrossEntropyLoss: 0.48
INFO Validate: Accuracy: 0.81, SoftmaxCrossEntropyLoss: 0.49
Training:    100% |████████████████████████████████████████| Accuracy: 0.84, SoftmaxCrossEntropyLoss: 0.44
Validating:  100% |████████████████████████████████████████|
INFO Epoch 4 finished.
INFO Train: Accuracy: 0.84, SoftmaxCrossEntropyLoss: 0.44
INFO Validate: Accuracy: 0.82, SoftmaxCrossEntropyLoss: 0.49
Training:    100% |████████████████████████████████████████| Accuracy: 0.85, SoftmaxCrossEntropyLoss: 0.41
Validating:  100% |████████████████████████████████████████|
INFO Epoch 5 finished.
INFO Train: Accuracy: 0.85, SoftmaxCrossEntropyLoss: 0.41
INFO Validate: Accuracy: 0.83, SoftmaxCrossEntropyLoss: 0.47
Training:    100% |████████████████████████████████████████| Accuracy: 0.86, SoftmaxCrossEntropyLoss: 0.39
Validating:  100% |████████████████████████████████████████|
INFO Epoch 6 finished.
INFO Train: Accuracy: 0.86, SoftmaxCrossEntropyLoss: 0.39
INFO Validate: Accuracy: 0.82, SoftmaxCrossEntropyLoss: 0.47
Training:    100% |████████████████████████████████████████| Accuracy: 0.86, SoftmaxCrossEntropyLoss: 0.37
Validating:  100% |████████████████████████████████████████|
INFO Epoch 7 finished.
INFO Train: Accuracy: 0.86, SoftmaxCrossEntropyLoss: 0.37
INFO Validate: Accuracy: 0.84, SoftmaxCrossEntropyLoss: 0.44
Training:    100% |████████████████████████████████████████| Accuracy: 0.87, SoftmaxCrossEntropyLoss: 0.36
Validating:  100% |████████████████████████████████████████|
INFO Epoch 8 finished.
INFO Train: Accuracy: 0.87, SoftmaxCrossEntropyLoss: 0.36
INFO Validate: Accuracy: 0.86, SoftmaxCrossEntropyLoss: 0.38
Training:    100% |████████████████████████████████████████| Accuracy: 0.87, SoftmaxCrossEntropyLoss: 0.35
Validating:  100% |████████████████████████████████████████|
INFO Epoch 9 finished.
INFO Train: Accuracy: 0.87, SoftmaxCrossEntropyLoss: 0.35
INFO Validate: Accuracy: 0.85, SoftmaxCrossEntropyLoss: 0.40
Training:    100% |████████████████████████████████████████| Accuracy: 0.88, SoftmaxCrossEntropyLoss: 0.33
Validating:  100% |████████████████████████████████████████|
INFO Epoch 10 finished.
INFO Train: Accuracy: 0.88, SoftmaxCrossEntropyLoss: 0.33
INFO Validate: Accuracy: 0.86, SoftmaxCrossEntropyLoss: 0.38
INFO forward P50: 0.523 ms, P90: 0.585 ms
INFO training-metrics P50: 0.017 ms, P90: 0.019 ms
INFO backward P50: 0.653 ms, P90: 0.764 ms
INFO step P50: 1.259 ms, P90: 1.378 ms
INFO epoch P50: 19.917 s, P90: 20.670 s
trainLoss = evaluatorMetrics.get("train_epoch_SoftmaxCrossEntropyLoss");
trainAccuracy = evaluatorMetrics.get("train_epoch_Accuracy");
testAccuracy = evaluatorMetrics.get("validate_epoch_Accuracy");

String[] lossLabel = new String[trainLoss.length + testAccuracy.length + trainAccuracy.length];

Arrays.fill(lossLabel, 0, trainLoss.length, "test acc");
Arrays.fill(lossLabel, trainAccuracy.length, trainLoss.length + trainAccuracy.length, "train acc");
Arrays.fill(lossLabel, trainLoss.length + trainAccuracy.length,
                trainLoss.length + testAccuracy.length + trainAccuracy.length, "train loss");

Table data = Table.create("Data").addColumns(
    DoubleColumn.create("epochCount", ArrayUtils.addAll(epochCount, ArrayUtils.addAll(epochCount, epochCount))),
    DoubleColumn.create("loss", ArrayUtils.addAll(testAccuracy , ArrayUtils.addAll(trainAccuracy, trainLoss))),
    StringColumn.create("lossLabel", lossLabel)

render(LinePlot.create("", data, "epochCount", "loss", "lossLabel"),"text/html");

4.6.6. Summary

  • Beyond controlling the number of dimensions and the size of the weight vector, dropout is yet another tool to avoid overfitting. Often all three are used jointly.

  • Dropout replaces an activation \(h\) with a random variable \(h'\) with expected value \(h\) and with variance given by the dropout probability \(p\).

  • Dropout is only used during training.

4.6.7. Exercises

  1. What happens if you change the dropout probabilities for layers 1 and 2? In particular, what happens if you switch the ones for both layers? Design an experiment to answer these questions, describe your results quantitatively, and summarize the qualitative takeaways.

  2. Increase the number of epochs and compare the results obtained when using dropout with those when not using it.

  3. What is the variance of the activations in each hidden layer when dropout is and is not applied? Draw a plot to show how this quantity evolves over time for both models.

  4. Why is dropout not typically used at test time?

  5. Using the model in this section as an example, compare the effects of using dropout and weight decay. What happens when dropout and weight decay are used at the same time? Are the results additive, are there diminished returns or (worse), do they cancel each other out?

  6. What happens if we apply dropout to the individual weights of the weight matrix rather than the activations?

  7. Invent another technique for injecting random noise at each layer that is different from the standard dropout technique. Can you develop a method that outperforms dropout on the Fashion-MNIST dataset (for a fixed architecture)?