Knowledge quiz

It's time to put what you've learned to the test, get 8 right to pass this unit.

Which configuration options in the Snap! Neural Network library blocks can you set to influence learning of the network?

There are three correct answers.

The number of iterations over the dataset.

How the data is partitioned for training and validation.

The number and size of hidden layers.

The color scheme for neuron visualization.

The label names after training to increase accuracy.

What is the key advantage of hyperdimensional programming?

Choose the correct answer.

It lets the same function operate on different data of any dimension such as images, sprite positions or sounds.

It combines multidimensional lists to one single value.

It uses higher order functions to easily filter data based on filtering criteria.

Because you're using loops you can always check for intermediate results of a hyperdimensional computation.

In generative AI, what is the name of the algorithm that predicts the subsequent element in a sequence, akin to choosing the next domino based on matching numbers?

Choose the correct answer.

Next token prediction algorithm

Previous word selection algorithm

Following item estimation algorithm

Subsequent piece matching algorithm

What defines the $1 Gesture Recognizer used in the lesson?

There are three correct answers.

It recognizes unistroke gestures drawn in a single continuous line.

It uses a nearest neighbor approach with Euclidean distances to identify gestures.

It was designed for rapid prototyping of gesture-controlled apps.

It natively supports multi-stroke gestures without modification.

It works without having to normalize gestures.

What do you call a block that has a ring and accepts other functions as an input?

Choose the correct answer.

A higher order function

A second class function

A command

A double reporter

What is the role of weights in a neural network?

Choose the correct answer.

They determine how strongly an input influences a neuron's output.

They decide how many layers the neural network has.

They control when the network stops training.

They convert categorical data into numerical form.

What is the main reason a single-layer perceptron cannot not learn the pancakes (XOR) dataset?

Choose the correct answer.

Because the problem is not linearly separable and requires multiple layers.

Because the dataset has too few samples to train.

Because the pancake data is not numerical.

Because XOR problem outputs are random and unpredictable.

What is the primary reason to resample a sketch to a fixed number of evenly distributed points?

Choose the correct answer.

To normalize drawings so they are comparable point by point across examples.

To make drawings the same size.

To increase the number of points captured for every sketch.

To remove the need for different training examples.

Which of the following statements about the context window are true?

There are two correct answers.

The shorter the context window the less coherent the generated text.

The context window is the amount of tokens taken into consideration when predicting the next token.

The context window decides which of the possible next tokens will be selected based on the topic of the previous conversation.

The longer the context window the less coherent the generated text.

10.

What is a word embedding?

Choose the correct answer.

A numerical representation of words in a vector space that captures semantic meaning.

A dictionary that lists all possible definitions of a word.

A list of all words in a given corpus.

The amount of words taken into consideration when predicting a next token in a text.