Chapter 10: Deep Learning Flashcards
what is deep learning
a representation learning technique
number of layers
what is representation learning
does more of the work of choosing features for us. done by a model and not by hand
give example representation learning techniques (3)
linear basis function model
kernel method
neural network
what is a convolutional neural network
learn feature vector for image
2 or 3 d data
what is a recurrent neural network
learns from sequential data.
neurons have internal memory about what came before
what are the two types of recurrent neural network neuron memories
LSTM long short term
GRU gated current unit
what are the layers in a CNN
convolutional
pooling
fully connected
what is a convolutional layer
local connectivity- each neuron in a layer is connected only to a receptive field of previous layer
describe weight sharing
a convolutional filter slides over all neurons in the layer
what is stride
the number of blocks you move across by
what is an activation map
the result of applying a filter to a layer
how do we calculate the size of an activation map
inputsize - filtersize
—————————— + 1
Stride
what is a pooling layer
input is the activation map
reduces the size of each activation map
downsampling
what is a fully connected layer (CNN)
connects activation maps and maps to a 1D array.
then uses this as input to a MLP model
what data is representational learning modernly used for
structured
sequential
graphical
