Deep Learning (DL)
able to extract hierarchical features of complex datasets through its multiple layer training with ANN
Artificial Neural Networks (ANN)
a computational model in which neurons, that hold the input, process information through weighted connections
How to train DL
Back-propagation
Back-propagation
is the process of taking incorrect paths and retracing the cycle to adjust your layers/weights
Sequence Models
input data in the form of sequences. The goal is to find patterns to make predictions
Recurrent Neural Network (RNN)
not a Feed-Foward Network; Designed to handle sequential data, their feedback loop allows information to persists across time steps; Contains a hidden state/memory to retain dependencies /patterns
Feed-forward neural network
a type of artificial neural network where information flows in only one direction from input to output layers without any cycles or feedback loops.
One-to-One RNN
produces a single input at each time-step and produces a single output at each corresponding time-step (Real-Time Prediction)
One-to-Many RNN
takes a single input @ generates a sequence of multiple outputs (Content Generation)
Many-to-Many RNN
takes in a sequence of inputs @ produces a sequence of outputs (Sequence Transformation)
Long Short-Term Memory (LSTM)
unlike RNN’s, this can handle long-term dependencies/memory by actively choosing what to remember
LSTM Gates Names
Input, Forget, Output
LSTM Input Gate
decides what new information from the current input should be stored
LSTM Forget Gate
decides what information in the current cell should be discarded/forgotten
LSTM Output Gate
regulates how much the current memory cell should be exposed as the output
Convolutional Neural Network (CNN)
type of DL model designed for processing grid-like data like images @ videos
CNN Layers (In, FE)
Input Layer → Feature Extraction
CNN Input Layer
accepts 3D images w/ height ,width,depth representing RGB color channels
CNN Feature Extraction
repeating pattern of convolutional layers that extract meaningful layers from input layer
CNN Feature Extraction Convolutional Layer
applied convolutional operations to the input image using small filters (kernels)
Convolutional Layer Kernel
slides across the input to detect specific features such as edges,corners, or textures
CNN Feature Extraction Activation Function
allows the network to learn more complex @ non-linear relationships in the data
CNN Feature Extraction Pooling Layer
helps reduce the spatial dimensions of feature maps generated by the convolutional layer
CNN Feature Extraction Fully Connected Layer
responsible for making final predictions or classifications based on the learned features
