Deep Learning Flashcards
What is the biological inspiration for Deep Learning?
It is inspired by the function and structure of the brain.
What are two types of Deep Learning?
Layered Representations Learning and Hierarchical Representations Learning
What are the three types of layers in a Deep Learning Model?
Input Layer, Hidden Layer, Output Layer
What is the Loss Function?
A function used to calculate how accurate the output of a model is.
Which Loss function is used for Binary output?
Binary Cross Entropy
What is the result of the loss function called?
Loss Score
How is the Loss Score used?
The loss score is fed into an optimizer function and used to update the weights.
What is the function of the input layer?
To receive the data.
What is the function of the Hidden Layers?
To extract features form the images of increasing complexity.
What is the function of the output layer?
To generate probabilities for the specified classes.
What is backpropogation used for?
Backpropogation is used to find a set of weights that minimises the error of the model.
What does SGD stand for?
Stochastic Gradient Descent
Why is an activation function used?
An activation function is used to cope with complex real tasks that require non-linearity.
How does SGD work?
SGD updates the weights of the whole network by looking at every example in the training set.
What is a Hyperparameter?
The specifications of the Architecture of the model
What are Epochs?
The number of time the training samples pass through he network
What is the Batch Size?
Partitions of the traning data that pass through the network.
What is Overfitting?
An issue that arises where a model becomes to specific and memorises the correct classification for the training data, and is unable to correctly classify the actual data being used.
Why is Dropout used?
It is used to overcome overfitting
How does Dropout work?
It randomly drops connections that have been created in order to force generalisation rather than memorisation
What are the three main Loss Functions?
Mean Squared Error, Binary Cross Entropy, Logarithmic Cross Entropy.
What are the two sections contained in a Convolutional Neural Network?
Feature Extraction and Classification
What are the three layers contained within the feature extraction on a CNN?
Input, Convolution, Pooling
How does a Convolutional Layer extract features?
A square filter is passed over the image and the dot product of the features contained in the filter is taken.
What is the advantage of a convolutional layer?
Convolutional Layers retain information about the spatial relationship between the pixels in the image, whereas previous models have converted everything into vectors which loses spatial knowledge.
What is a pooling layer?
The third layer in the feature extraction section it decreases the size of the feature map to reduce computational cost.
What are the two types of pooling?
Max Pooling and Average Pooling.
What is Dimensionality Reduction?
Representing Multi-Dimensional data in a lower dimension.
What is the Curse of Dimensionality?
Sample Density decreases exponentially with in the increase in dimensionality.
What is an Autoencoder?
A types of unsupervised ANN that aims to learn representation of the data in a way that captures the essential features, reducing the dimensionality of the data.
What is the Objective Function used for in an Autoencoder?
It measures the difference between the Input data and its reconstruction and aims to minimise this difference.
