deep-learning-with-python-meap-2nd-ed

Deep learning models only have to fit relatively simple, low-dimensional, highly-structured subspaces within their potential input data, learning to generalize via interpolation between training samples.

This phenomenon is known as the manifold hypothesis.

The central problem in machine learning is the tension between ____ and ____: the ideal model is one that stands right at the border between ____ and ____; between undercapacity and overcapacity.

Optimization and generalization; underfitting and overfitting.

When should you use the Sequential model over the Functional API?

You should use the Sequential model when your model can be expressed as a simple stack of layers.

The method used to interrupt training when the validation loss is no longer improving is called __________.

EarlyStopping

In Keras, the core building block used to create models is called a __________.

Layer.

What type of problem involves predicting a continuous value instead of a discrete label?

Regression.

The belief that by selecting only the most informative features, you can improve model performance is known as __________.

Feature selection.

A technique that applies a random fraction of units in a neural network layer to be dropped out during training is called __________.

Dropout.

In machine learning, the process of adjusting a model to get the best performance possible on the training data is called __________.

Optimization.

The type of dataset that is used to evaluate a model without any training is known as a __________.

Test set.

When training on images, the size of the data structure (tensor) is usually expressed in the format of (samples, height, width, channels) known as __________.

Channels-last format.

The function used to retrieve the gradient of a loss with respect to the model's variables in TensorFlow is called __________.

GradientTape.

If you need a model to learn complex relationships between inputs and outputs that are not linearly separable, you need to apply a __________ to your last layer.

Non-linear activation function.

A common issue that occurs when a model starts to overly fit to the noise in the training data is called __________.

Overfitting.

The process of adjusting the values of weights in a neural network based on the computed gradients of the loss function is known as __________.

Backpropagation.

The technique that adds a penalty for large weights to the loss function is known as __________.

Weight regularization.

The first step in the training process is to ______ the dataset into training and validation sets.

split

During training, the model learns by adjusting its parameters based on the ______ calculated from the predictions.

loss function

The technique used to prevent overfitting by incorporating noise in the training process is known as ______.

data augmentation

After training the model, its performance is evaluated using the ______ set.

validation

A common strategy to handle imbalanced datasets is to apply ______ techniques.

resampling

To improve model performance, you may want to perform ______ on the optimal hyperparameters.

hyperparameter tuning

When a model's performance is consistently good on the training set but poor on the validation set, it is said to be experiencing ______.

overfitting

The process of using a validation set to make adjustments during training is known as ______.

model tuning

Creating an environment to test and validate the model outside the training set is called ______ testing.

cross-validation

The final performance metric used to evaluate a model can be ______, which summarizes how well the model predicts the target variable.

accuracy

What is the purpose of splitting data into training and validation sets?

The purpose is to train the model on one subset while evaluating its performance on another to prevent overfitting.

What are the two main sets used in machine learning?

Training set and testing set.

What is cross-validation?

A technique that divides the dataset into multiple subsets to evaluate the model's performance multiple times.

What is hyperparameter tuning?

It is the process of adjusting model parameters that are not learned during training, to improve performance.

What does a validation set help with during training?

It helps with monitoring model performance and makes adjustments to improve accuracy.

What is the difference between training loss and validation loss?

Training loss measures error on the training data, while validation loss measures error on unseen validation data.

Why is monitoring validation loss important?

It helps identify overfitting; when training loss decreases while validation loss increases.

What is an epoch in the context of model training?

An epoch refers to one full pass through the entire training dataset.

How does data augmentation benefit the training process?

It increases the diversity of training data without needing more data, reducing the chance of overfitting.

What role does the test set play after training a model?

The test set evaluates the final model performance on completely unseen data to estimate how it will perform in practice.

What is the first step in preparing data for model training?

The first step is to clean and preprocess the data to handle missing values, outliers, and formatting issues.

Why is it important to have a training set and a test set?

The training set is used to train the model, while the test set evaluates its performance on unseen data.

What is validation in the context of model training?

Validation involves using a separate validation set to tune model parameters and avoid overfitting.

What technique can be applied to maximize the data generated during training?

Data augmentation increases the size of the training dataset by applying transformations to existing data, such as rotations or flips.

What is k-fold cross-validation?

K-fold cross-validation is a technique that divides the dataset into 'k' subsets, training the model k times, each time using a different subset as the validation set.

What role does normalization play in preparing data?

Normalization scales the input features to a standard range, improving model convergence during training.

How can you monitor model performance during training?

By tracking training and validation loss and accuracy metrics at the end of each epoch.

What is early stopping in model training?

Early stopping is a technique to halt training when performance on the validation set starts to degrade.

Why is hyperparameter tuning crucial in training models?

Hyperparameter tuning is essential to find the best settings for the model to improve performance on the validation set.

What is the purpose of a confusion matrix in model evaluation?

A confusion matrix provides a detailed breakdown of predictions by showing the true positive, true negative, false positive, and false negative counts.