week 11: neural networks + pattern completion (lectures 20 & 21)

What are the goals of unsupervised learning algorithms? Select all that apply then submit your answer.

• Discover hidden patterns in data

• Organize or cluster unlabeled data

• Predict specific labels based on past examples

• Form cell assemblies that store labeled categories

What is the primary goal of supervised learning algorithms?

• Group unlabeled data into clusters based on similarity

• Identify patterns in data without using labeled examples

• Organize data by shared characteristics without external guidance

• Map input features to known outputs for prediction or classification

Which of the following best describes how backpropagation works? Select all that apply then submit your answer.

• It computes the difference between the predicted and actual output.

• It sends error derivatives backward through the network.

• It changes the weights to reduce error.

• It stores previously correct outputs to avoid recalculating them.

How does the Perceptron learn, and when does it stop updating?

• It updates its weights after every input and stops after a fixed number of tries.

• It adjusts weights randomly and stops when the weights become constant.

• It updates weights when it makes mistakes and stops when it classifies all training examples correctly.

• It stores every answer and stops when the dataset runs out of inputs.

Imagine training a neural network is like standing in a hilly landscape and trying to find your way to the lowest point in the valley, where the error is smallest. In this analogy, what does the error derivative represent?

• Where you are on the terrain (how wrong the model currently is)

• The slope of the hill that tells the model which direction to move to reduce error

• The valley (where error is as low as it can go)

• The direction the model moves based on how much error it sees

Which of the following describe the relationship between spurious memories and top-down processing? Select all that apply then submit your answer.

• Both rely on past experience to help interpret incomplete or new information.

• Top-down processing always produces accurate and complete information.

• Like top-down processing, spurious memories can fill in gaps using learned patterns.

• Both can sometimes lead to incorrect interpretations or "false memories".

You have a system that only responds correctly when it’s given the exact input it was programmed for. What does this say about how the system works?

• It can’t generalize beyond the inputs it was programmed for.

• It uses feedback to update its behavior over time.

• It fills in missing information using stored associations.

• It adjusts its responses based on similarity to past inputs.

Why are both pattern recognition and pattern completion important in AI? Select all that apply then submit your answer.

• Pattern recognition helps identify clear input data like speech or images.

• Pattern completion allows the model to function even when input is partial or noisy.

• Pattern completion can replace recognition when input data is too noisy.

• Pattern recognition is no longer needed now that models can complete patterns.

You have a program that starts with random settings and gradually improves its responses by comparing its output to the correct answer and adjusting itself. What process is this describing?

• Running fixed instructions from top to bottom

• Training a neural network to reduce error

• Debugging code to fix mistakes manually

• Compiling a program before execution

Which of the following best describes how Hebbian learning works in an unsupervised neural network?

• Neurons reduce their activity when they repeatedly fire at the same time.

• Output units adjust their weights based on feedback from a target signal.

• Hidden units store previous firing patterns and replay them when input is missing.

• The strength of the connection between two neurons increases when they activate together.