week 3 - chatgpt Flashcards
What limitation of linear discriminant functions motivated the development of neural networks?
Linear discriminants can only separate data with a straight line (or hyperplane); they cannot solve problems like XOR that are not linearly separable.
What is the key difference between a linear threshold unit and a multilayer neural network?
A linear threshold unit can only model linearly separable functions
Why can’t a single-layer perceptron solve the XOR problem?
Because the XOR function is not linearly separable
What innovation allows multilayer networks to solve nonlinear classification tasks?
The use of hidden layers with nonlinear activation functions enables multilayer networks to create complex decision boundaries.
How does the Delta rule improve on the Perceptron rule?
The Delta rule updates weights based on continuous error
What is the biological inspiration for competitive learning in neural networks?
It mimics the behaviour of neurons in the brain where only the most activated neuron (the winner) updates
How is a competitive network different from a standard feedforward neural network?
In a competitive network
What problem does the autoencoder solve that standard feedforward networks do not?
Autoencoders learn efficient representations by compressing data through a bottleneck
Why is an autoencoder considered a form of unsupervised learning?
Because it learns to reconstruct its input without requiring labelled data
What makes a sparse autoencoder different from a basic autoencoder?
A sparse autoencoder includes a sparsity constraint on the hidden layer
