topic 5 Flashcards
(34 cards)
what is an artificial neuron
- an individual building block if a neutral network
- made up of bias and activation function
what are weights
- give importance to features that contribute more towards learning
- represent strength of connections between neurons
what is transfer (summation) function
combines multiple inputs into one output value, so activation function can be applied
what is activation function
introduces non-linearity to the network
what is bias
shifts the values produced by activation function, allows neurons to make predicitions even when all input is zero
what is a layer
combination of multiple neurons stacked together in a row
what is input layer
each neuron corresponds to a feature in the input dataset
what is hidden layers
intermediate layers that do computations and extract the features from the data
what is the output layer
maps the learned features from hidden layer to final output
example of supervise learning algorithms
- backpropagation
- gradient descent
- stochastic gradient descent
examples of unsupervised learning algorithms
- autoencoders
- generative adversarial
examples of reinforcement learning algorithms
- q- learning
- policy gradient networks
types of connection pattern
- feedforward (graphs have no loops)
- recurrent (loops occur because of feedback)
types of weights
- fixed (not changed at all)
- adaptive (update weights though out training)
types of memory unit
- static (memoryless, current output depends on current input)
- dynamic (output depends upon current input as well as current output)
what are feed forward neutral networks
- data flows from input layer to output layer without loops
LOOK AT NOTES FOR PIC OF IT
What is multi-layer perceptron
- it is an FFNN with fully connected neutrons with non-linear activation functions
what are convolutional neutral networks
- extract features from images, such as edges, textures and shapes. Features used to recognise objects, patterns and classify images
What are recurrent neutral networks
- process sequential data, where order of inputs matters
- contain loops that allow info to pass from one step to the next
- suitable for tasks that involve time - series data or sequences of information
what does encoder - decoder model do
what is an encoder
- read input sequence
- Summarises info in fixed-length vector
- passed along as input for decoder
what is a decoder
- interprets context vector and generates output sequence
autoencoders
- encoder-decoder models in which input an output domains are the same
large language models
designed for natural language processing tasks
