Traditional ML
This type of traditional machine learning is used for regression tasks, and uses linear functions to predict the output.
Linear regression
What ML model uses convolutional layers to detect features?
CNNs
How do you initialize a Sequential model in Keras?
Model = Sequential()
True or False: Underfitting and Overfitting are the only types of issues that need to be addressed to improve ML models.
False
Deep Learning is a subfield of what field?
Machine learning
Used for classification tasks and involves the use of a decision boundary to separate different classes.
Decision trees
True or False: KNNs and Decision Trees are used for classification tasks.
True
How do you add a fully connected layer to a Keras neural network?
Model.add(Dense(number_of_neurons))
What are the ML settings decided by the programmer called?
Hyperparameters
When is Deep Learning preferred to traditional Machine Learning?
Large amount of data, complex relationships, datatypes not able to be used by traditional ML, when high accuracy is required
(students only need to say one of the above)
What are the types of ML algorithms?
Supervised, unsupervised, RL
What is a model that assigns data points to the nearest cluster based on a distance metric?
Kmeans
Which keras layer is used for tokenization and vectorization?
TextVectorization
What is the difference between a validation dataset and a test dataset in machine learning?
Validation dataset is used while improving the model to make sure model is unbiased by test set
What is the name for a network where every neuron in a layer is connected to every neuron in the next layer?
Fully connected
Type of ML Algorithm that does training by rewarding desired behaviors and punishing undesirable ones.
Reinforcement learning
What is the difference between supervised and unsupervised learning?
Supervised learning uses labeled data
Which optimizer uses an adaptive learning rate?
Adam
True or false: Weights and biases are two hyperparameters we can adjust during the training phase.
False. Weights and biases are not hyperparameters.
What is the difference between predicted output from model and the actual output called?
Loss or cost or error
Give an example of when Traditional ML is preferred over Deep Learning.
applications where traditional machine learning is preferred over deep learning:
Small data sets
Interpretable models
Linearly separable data
Extensive feature engineering required
Resource constraints
Applications where traditional machine learning may be preferred:
Credit scoring
Fraud detection
Simple image recognition
Simple speech recognition
Recommendation systems
Name one reason feature selection is an important step in building ML models.
Reduces Overfitting (fewer features), Improving Model Performance (only uses most relevant features), Reducing Training Time (less features), Improving Model Interpretability (fewer features means more interpretable), Data Quality (removing noise or irrelevant features)
Any version of the above examples of reasons is acceptable
Which loss function do we use for binary classification problems?
Binary crossentropy
Name two hyperparameters used in deep learning.
Number of neurons, number of layers, etc
Which activation function solves the vanishing gradient problem?
Adam